More IBM 64-bit Chip Speculation [Archive]

DharvaBinky

Oct 9, 2002, 05:08 PM

Originally posted by MacManiac1224
This just adss fuel to the fire. I hope that Apple will use this chip, and that it will be out by MWNY. That is my bope, if it is, they can count on me to buy it. The G4 is apporaching 2 years old.

My question is: If Apple wanted to change OS X into a 64-bit operating system, how long would that take? And if so, when they release the IBM chip, do they have to have a 64 bit operating system, or will they just wait. Personally, I think it is perfect: the IBM chip comes out at MWNY with a shipping date in august, and at the same time, 10.3 comes out, with that same shipping date, therefore a 64-bit operating system.

What do you guys think?

As I understand it... MacOS X is already what Apple refers to as "64-bits clean". Meaning, that... Although, they're not issuing 64-bit instructions, all registers and memory spaces use 64-bit addresses padded with 32-bits of zeros (this is could be entirely inaccurate as to how that works). I remember hearing lots of stuff about how MacOS System 6 and above were "32-bits clean" when we started getting the 68020 and 68030s which were 32-bit processors.

Regardless, I do know that Apple claims OS X is "64-bits clean" so it would run unmodified on a 64-bit processor.

How long would it take them to retool the software for 64-bit instructions? <grin> Since it's the same compiler and basically the same instruction set... it could be as easy as... a recompile.

: )

Binky

Falleron

Oct 9, 2002, 05:10 PM

Originally posted by DharvaBinky

As I understand it... MacOS X is already what Apple refers to as "64-bits clean". Meaning, that... Although, they're not issuing 64-bit instructions, all registers and memory spaces use 64-bit addresses padded with 32-bits of zeros (this is could be entirely inaccurate as to how that works). I remember hearing lots of stuff about how MacOS System 6 and above were "32-bits clean" when we started getting the 68020 and 68030s which were 32-bit processors.

Regardless, I do know that Apple claims OS X is "64-bits clean" so it would run unmodified on a 64-bit processor.

How long would it take them to retool the software for 64-bit instructions? <grin> Since it's the same compiler and basically the same instruction set... it could be as easy as... a recompile.

: )

Binky
Now thats what I like to hear! Hope you are correct.

retaks

Oct 9, 2002, 05:22 PM

okay let me get this straight im getting confused with these bits. So we want a 64bit chip, what bit do pc's use? and what bit is mac using now??

skunk

Oct 9, 2002, 05:25 PM

Gimme, gimme, gimme!!! :p :) :D

MacCoaster

Oct 9, 2002, 05:26 PM

Originally posted by DharvaBinky
I remember hearing lots of stuff about how MacOS System 6 and above were "32-bits clean" when we started getting the 68020 and 68030s which were 32-bit processors.
Uh. The entire M68k family is 32bit, even the original Motorola 68000.

scem0

Oct 9, 2002, 05:36 PM

Now, will this come out early next year, or late next year, or next year at all? I am betting that the g5 will come Jan of next year, and then the power4lite/GPUL powermac will come out later next year, but it will keep the title of 'g5'.

Mr.Hey

Oct 9, 2002, 05:38 PM

My name is on the front page....I'm famous

Furious Tiger

Oct 9, 2002, 05:40 PM

Originally posted by scem0

I couldn't agree more.

Interesting quote you added. Might you have a bit more insight that you can tell us?

retaks

Oct 9, 2002, 05:40 PM

okay let me get this straight im getting confused with these bits. So we want a 64bit chip, what bit do pc's use? and what bit is mac using now??

Microsoft_Windows_Hater

Oct 9, 2002, 05:41 PM

A movement of any BSD or 'nix in general simply requires a recompile. If apple has been playing their cards correctly then all kernel memory calls will be in place correctly for 64bit and it would only require a recompile. As for Mac OS X itself, not darwin, I don't think there would be any major problems. The only big problem would be what applications would be.

Today's app's are compiled in 32bit. If you go and compile in 64bit then it isn't backwards compatible. Would apple have to have 64bit app's compiled on install? I doubt it, really messy. I guess they will just introduce 64bit across the entire line at once, to enable everyone to get the required speed boost without any nasty consequences.

Furious Tiger

Oct 9, 2002, 05:43 PM

Originally posted by gandalf55
let's see... Apple goes this route to prove they aren't crazy... and then there is a shortage of 64bit chips... pissing all the people off who ordered new macs. i'd hope this wouldn't be something that would happen.

HAHAHAHA, Please don't remind us of the past.:(

vniow

Oct 9, 2002, 05:43 PM

Originally posted by retaks
okay let me get this straight im getting confused with these bits. So we want a 64bit chip, what bit do pc's use? and what bit is mac using now??

32 biy is what all current CPUs are at now, PC or Mac.
AMD's upcoming Clawhammer along with Intel's Itanium and this IBM chip are all 64 bit.

barkmonster

Oct 9, 2002, 05:44 PM

okay let me get this straight im getting confused with these bits. So we want a 64bit chip, what bit do pc's use? and what bit is mac using now??

The G3, G4, Celeron, Pentium III, Pentium 4, Athlon XP and Athlon MP are all 32bit cpus, this covers all the desktop cpus in PCs you can buy right now (the mac is Personal Computer aswell).

As for these endless speculations about both the chip and whether apple will use it. I just hope this isn't all BS or wishful thinking and something actually happens next year. Just after the G3 was introduced there was talk of the G4 and a future version of it with the features IBM have in the chip available by late 2002. The article was so old it even had 1200MHZ!!!!! in big letters in the article as if that was blistering speed. At the time G3s were at 266Mhz and the Pentium II and K6 cpus were at 600Mhz.

I summarised it in this message (http://forums.macrumors.com/showthread.php?postid=149887#post149887)

There's been talk of the G5 in one form or another for years, people were even swept up into the idea that apple would be the first company to release a desktop PC with a 64bit cpu and OS because of it. Baring in mind how close the windows world is to getting 64bit desktop cpus and how far away it seems before apple will, this seems very disappointing with the situation we're in now.

Saying that and being WAY too optimistic, when a next generation powermac comes out and intel keep using design compromises to push their clockspeeds higher we could be in a situation of apple running those snail ads again by this time next year.

Imagine it :

5Ghz Pentium 4, 1 instruction per clockcycle, huge power requirements and elaborate cooling methods.

2Ghz PowerPC, 8 instructions per clockcycle, low power and turbine cooled.

Jimong5

Oct 9, 2002, 05:44 PM

Originally posted by retaks
okay let me get this straight im getting confused with these bits. So we want a 64bit chip, what bit do pc's use? and what bit is mac using now??
32 and 32.

Furious Tiger

Oct 9, 2002, 05:45 PM

Originally posted by retaks
okay let me get this straight im getting confused with these bits. So we want a 64bit chip, what bit do pc's use? and what bit is mac using now??
32 retaks. Intel's Itanium uses 64 and we (macs) use 32 as well.

DharvaBinky

Oct 9, 2002, 05:52 PM

Originally posted by MacCoaster

Uh. The entire M68k family is 32bit, even the original Motorola 68000.

I guess sorta... I found this:

"The 68000 has 32-bit registers but only a 16-bit ALU and external data bus. It has 24-bit addressing and a linear address space, with none of the evil segment registers of Intel's contemporary processors that make programming them unpleasant. That means that a single directly accessed array or structure can be larger than 64KB in size. Addresses are computed as 32 bit, but the top 8 bits are cut to fit the address bus into a 64-pin package (address and data share a bus in the 40 pin packages of the 8086 and Zilog Z8000). "

32-bit registers... 16-bit math and external databus. Because the databus and execution unit are 16-bit, I think it would qualify more as a 16-bit processor, although, I'm sure people would disagree...

I *do* remember the this-n-that about System 7 being 32-bits clean, though, and that being a big deal for transition to PPC.

Sooo... with the attention given to MacOS X and it being 64-bit clean, I think it all sounds good to me.

:)

Binky

DharvaBinky

Oct 9, 2002, 06:05 PM

Originally posted by Microsoft_Windows_Hater
A movement of any BSD or 'nix in general simply requires a recompile. If apple has been playing their cards correctly then all kernel memory calls will be in place correctly for 64bit and it would only require a recompile. As for Mac OS X itself, not darwin, I don't think there would be any major problems. The only big problem would be what applications would be.

Today's app's are compiled in 32bit. If you go and compile in 64bit then it isn't backwards compatible. Would apple have to have 64bit app's compiled on install? I doubt it, really messy. I guess they will just introduce 64bit across the entire line at once, to enable everyone to get the required speed boost without any nasty consequences.

Again, its my understanding from having done a layman's read of Book E white paper, that all PPC chips adhering to Book E (the 64-bit implementation of PPC) are able to execute 32-bit instructions. I would imagine, though, that unless the apps are written 64-bit clean as well (I've found numerous articles about ftp servers that aren't etc...) they might take a speed hit while the CPU pads the registers out to 64-bit... <shrug> However, the GPUL will apparantly be working in the same capacity as AMD's clawhammer where 32-bit and 64-bit instructions will be running in a mixed environment.

I think it will be pretty interesting to see how this affects memory and storage performance. Currently, 4Gb is the physical limit for RAM in a machine due to 32-bit mapping. Some high-end chipsets for servers can address more than 4Gb by doing a kind of "virtual RAM" mapping technique onto real RAM. This will definately allow Macs to have a theoretical RAM limit of vast numbers (way over even Terabytes, what's that... Petabytes?). Also, this would directly affect storage as well. Remember drives having partitions limited to 4Gb? 32-bit addressing again... now we do a map to maintain 64-bit addressing across the drives (IDE uses LBA addressing to make it happen)... with full 64-bit ops, the drives could be addressed directly with no translation.

This doesn't sound like a big deal, but... if the OS is having to do a little look-up everytime the drive is accessed to map out the 64-bit address to 32-bit storage... those could add up significantly... removing those (and other little tricks that had to be put into 32-bit operations) could aggregate to a nice speed boost for OS X...

Binky

Dave Marsh

Oct 9, 2002, 06:47 PM

It's my understanding that the PPC spec was designed to permit a clean move to 64-bit from the beginning. Over the past month, several techs have stated repeatedly on these forums that today's 32-bit applications will run cleanly on the new PPC64 chip without modification.

Concerning the chip availability fiasco we have repeatedly suffered through because Moto hasn't been able to provide chips in the quantities needed, the new PPC64 GPUL chip will be manufactured by IBM, most likely at their new Fishkill, NY facility. IBM is a master at this. I would really doubt they would fail to supply contracted quantities of any chip, barring events beyond their control (e.g., the plant burning down, or whatever). They plan, and they execute...professionally.

Finally, it is extremely doubtful to me that IBM would have invested the time/money to build in the SIMD vector processing unit (Altivec-compatible) without a customer in mind to use it. While IBM will certainly use this chip themselves, it seems clear this modification was included to meet a key customer's requirements...Apple's.

I am disappointed to hear that this chip may not be in production for another 8-9 months. From prior posts, it appears this chip has been available (at least in small quantities) for testing for many months. I wonder what's been taking so long to get the final version right? I too am looking forward with great interest to what IBM has to say next week.

prewwii

Oct 9, 2002, 07:09 PM

Originally posted by MacCoaster

Uh. The entire M68k family is 32bit, even the original Motorola 68000.

You're kind of right. Mac's were 32 bit from the git go BUT only used the first 24 bits for a long time. I think system 7 was the first of the real 32 usage. Lots of programs used the upper 8 bits for their own special purpose until the full use of 32 came into being.

Jim

Chisholm

Oct 9, 2002, 07:39 PM

Originally posted by Dave Marsh

I am disappointed to hear that this chip may not be in production for another 8-9 months. From prior posts, it appears this chip has been available (at least in small quantities) for testing for many months. I wonder what's been taking so long to get the final version right? I too am looking forward with great interest to what IBM has to say next week.

I completely agree with you. But in all fairness these major changes in the processors take YEARS to bring to market. P5 has been in development for at least a year. I read a really long article a year or so ago that was predicting the leap frog AMD is about to do over Intel. It was way too technical for me, but I was able to nod my head and follow the major points. It was the author's view that Intel made a decision to dominate speed wise in the short term without planning well enough for the longer term. Course the guy could be just a disgruntled engineer who wasn't getting enough at home, but it was intresting to see just how much planning is required.

A coworker sent me a PDF the other day showing Moto's in the beginning stages of planning the G6 (VERY early...).

Anyway, my evil peece is begging for my attention. Sorry if I didn't really say anything.

Cheers!
-john

j763

Oct 9, 2002, 07:40 PM

this all sounds great but the big question is -- when will we get our hands on it? Jan? July? or later?

kenohki

Oct 9, 2002, 07:46 PM

Originally posted by Microsoft_Windows_Hater
The only big problem would be what applications would be.

Today's app's are compiled in 32bit. If you go and compile in 64bit then it isn't backwards compatible. Would apple have to have 64bit app's compiled on install? I doubt it, really messy. I guess they will just introduce 64bit across the entire line at once, to enable everyone to get the required speed boost without any nasty consequences.

I would assume that they'd like to have software shipped in the Mac OS X bundle system. Then they can have both the 32-bit and 64-bit versions of the software in one "icon". The OS knows which microarchitecture it's running on so it just executes the proper binaries when the application is launched. That would make the user experience a lot better during the migration. Sort of like fat binaries during the 68k to PPC transition or bundles between 68k, x86, SPARC, and PA-RISC with NeXTstep/OpenStep. 32 bit binaries should run just fine, just like SPARC binaries run on UltraSPARC or SPARC64 without a recompile. But they're going to have to maintain backward compatibility with the installed 32-bit base.

However, the majority of users won't need the extra memory or precision at this point.

Macmaniac

Oct 9, 2002, 08:18 PM

Originally posted by j763
this all sounds great but the big question is -- when will we get our hands on it? Jan? July? or later?

If I know Apple correctly here is a possible path they may take:
MWSF Apple updates iMacs, laptops, and XServe
MWT Apple increases the powermacs to 1.5ghz
MWNY Laptops possibly, Hints at 10.3 and provides some details (Little do we know it is a 64 bit OS X)
Sometime after MWNY before October Apple will release Powermacs with Power4 lite(64bit) at maybe 1ghz to 1.5ghz with 10.3. This my IBM 64bit speculation. If I am right the possiblilites are endless(We may be reading this post a yeart from now;) ;)

One question what happnes to the G5 from Moto??? Does it go into the iMacs and laptops?

oldMac

Oct 9, 2002, 08:28 PM

Originally posted by DharvaBinky

"The 68000 has 32-bit registers but only a 16-bit ALU and external data bus. It has 24-bit addressing and a linear address space, ... Addresses are computed as 32 bit, but the top 8 bits are cut to fit the address bus into a 64-pin package (address and data share a bus in the 40 pin packages of the 8086 and Zilog Z8000). "
...
I *do* remember the this-n-that about System 7 being 32-bits clean, though, and that being a big deal for transition to PPC.

You're kind of right. Mac's were 32 bit from the git go BUT only used the first 24 bits for a long time. I think system 7 was the first of the real 32 usage. Lots of programs used the upper 8 bits for their own special purpose until the full use of 32 came into being.

These are both accurate statements. The 68000 was internally 32 bit, but was restricted to an effective 24 bits of external addressing because Motorola decided to cram it into a 64-pin package. This was resolved with the 68020.

As an old-school trick, some programmers took advantage of the extra 8 bits as a sort of "extra register" where they could cram through a little more information to gain performance. Of course, when System 7 started utilizing 32 bit addressing, this "extra space" would get clobbered causing the programs to crash.

"32-bit clean" refers to an application that doesn't attempt to take advantage of the extra space that's left open by the 24-bit System 6 running on a semi-32 bit processor (68030).

In fact, some of the early Macintosh ROMs also used these tricks. This caused those machines to not be 32-bit clean and to require either running in System 7's 24-bit compatibility mode or by installing a piece of software made by Connectix (Mode32) and later purchased by Apple (32 bit enabler) that "patched out" the dirty ROM routines.

Supporting 32-bit apps on 64-bit hardware should really not be a big deal at all.

Catfish_Man

Oct 9, 2002, 08:59 PM

...the big deal with this chip is NOT that it's 64 bit (although that makes a darn nice marketing number :)). It's that it's based on the POWER4 architecture which has modern features that the G4+ is lacking (out of order execution, multiple FPUs, high bandwidth memory bus), as well as being capable of an insane 8 instructions/cycle (both the G4+ and the Pentium 4 can do 3). I don't know how much of this the GPUL will include, but any of it will be a major improvement (dual FPUs will mean that doing a single FP divide will no longer block floating point heavy apps for 25 cycles while it finishes, they'll just use the other FPU, oooe will also help with that).

Scottgfx

Oct 9, 2002, 10:45 PM

Originally posted by Catfish_Man
...the big deal with this chip is NOT that it's 64 bit (although that makes a darn nice marketing number :))

Mmmmmmm, Atari Jaguar. :)

BigFish

Oct 9, 2002, 11:08 PM

Originally posted by Macmaniac

If I know Apple correctly here is a possible path they may take:
MWSF Apple updates iMacs, laptops, and XServe
MWT Apple increases the powermacs to 1.5ghz
MWNY Laptops possibly, Hints at 10.3 and provides some details (Little do we know it is a 64 bit OS X)
Sometime after MWNY before October Apple will release Powermacs with Power4 lite(64bit) at maybe 1ghz to 1.5ghz with 10.3. This my IBM 64bit speculation. If I am right the possiblilites are endless(We may be reading this post a yeart from now;) ;)

One question what happnes to the G5 from Moto??? Does it go into the iMacs and laptops?

I think Apple will bring out Moto's G5 in March as the 1.5GHz. Reports have stated that they would be delivering them in supply in early '03. Then possibly IBM units in January '04. Apple could then replace the Moto with the IBM and move the Moto to the iMac and PBook. And maybe the iBook will get a G4 sometime in '04.

P-Worm

Oct 9, 2002, 11:28 PM

Originally posted by Scottgfx

Mmmmmmm, Atari Jaguar. :)

Very good point indeed. However i feel that this entire board has fallen into a lapse of wanting big numbers right now. I thought that a board of Mac heads would surely know that overall system performance is what. For example, we have an elephant at the circus that can jump through hoops (Bare with me here...). He can jump through an amazing amount of 1500 hoops each minute. Wow! I gotta see this in action! But when I go to the show, it turns out that the elephant is terribly trained and won't do what the ring master says. My point is that Ghz is a theoretical value. Just like this elephant CAN jump through 1500 hoops a minute, but won't.

Sorry guys, but I think that was the worst analogy that has ever been used on these boards. But if you dug how confusing it was, then the movie Mulholland Drive is for you.

P-Worm

bretm

Oct 10, 2002, 12:05 AM

Booga

Oct 10, 2002, 01:45 AM

Originally posted by DharvaBinky

As I understand it... MacOS X is already what Apple refers to as "64-bits clean". Meaning, that...

The basic assumption in "64-bit clean" code is that no assumption is ever made that a pointer is 32 bits. Such an assumption can be manifest in certain data structures (such as ones that leave 32 bits as "user data", intended to be a pointer) or can be in code where an assumption is made about comparing the size of a void* to an int. I don't know how gcc or Metrowerks handles it, but some of the old 64-bit Alpha compilers had "int" be 32 bits, void* be 64 bits, and "long long" be 64 bits. Badly written code could store a void* in an int or vice-versa, or perform bit or bytewise operations on the pointer as if it were an int.

In answer to another question, currently all home computers use 32 bit ABIs, with "Wintel" machines using the IA32 (aka x86) instruction set, and Macintoshes using the 32bit PowerPC instruction set. Since 32 bit chips can only address a flat 4GB of space, it makes that the hard upper limit for RAM in most of these machines. Due to this limitation, the lifetime in years of any 32 bit ABI can be counted on one hand. The replacements on the other side of the fence are x86-32 (AMD) and IA64 (Intel/HP). If IBM comes in with a viable 64 bit offering, since everyone else must soon change ABIs anyway, who knows...

Scottgfx

Oct 10, 2002, 02:12 AM

Originally posted by P-Worm
[B]

Very good point indeed. However i feel that this entire board has fallen into a lapse of wanting big numbers right now.

That's natural, we always want what the other doesn't have. (yet) I just glad that someone got the reference. :)

I think I was rooting for the Amiga CD32 at the time. 68020 I believe, right? Man, if Commodore had just added up all the registers and used that number instead... Well, by that time Mehdi Ali had already mucked things up too much already. :)

pianojoe

Oct 10, 2002, 02:22 AM

Originally posted by oldMac

As an old-school trick, some programmers took advantage of the extra 8 bits as a sort of "extra register" where they could cram through a little more information to gain performance. Of course, when System 7 started utilizing 32 bit addressing, this "extra space" would get clobbered causing the programs to crash.

Applause! Applause! Applause!

And this was the reason why Tempus was the fastest text editor I ever worked with! And it ran on my 8MHz Atari...

Supporting 32-bit apps on 64-bit hardware should really not be a big deal at all.

Two thoughts:

If a program writes out a memory location to a file, the 32bit version will write out 4 bytes, and the 64bit version will write out 8 bytes. The data files will not be compatible without extra effort.

Early PPC systems provided an emulation mode for 680x0 code. If a software company didn't have the manpower to port their whole product to the new processor, they could at least port the performance-intensive parts first. I remember that a good portion of System 7.5 was still 680x0 code running in emulation even on the PPC. Same goes for altivec optimization. It only pays out for, say, 10% of the code, so why bother with the rest, since the G4 can behave as a perfect G3?

This helps migration a lot. (Similar to introducing the carbon lib very early and providing classic mode in OS X.) I guess Apple would do the same thing with a 64-bit-processor: Create an environment to run your 32-bit-stuff, but make it possible to hand-taylor specific subroutines to run natively on 64-bit for performance reasons...

nerm

Oct 10, 2002, 04:33 AM

Originally posted by pianojoe

If a program writes out a memory location to a file, the 32bit version will write out 4 bytes, and the 64bit version will write out 8 bytes. The data files will not be compatible without extra effort.

Well, if you would write down memory locations from a 64bit program to a file and try to read it on another (32bit) computer you couldn't do anything with them since they're locations in another (physical) memory. If you would try to read the file from a 32bit application on the same computer you still wouldn't be able to get any usage from the memory addresses since they're in different processes and therefore protected by the operating system...

Early PPC systems provided an emulation mode for 680x0 code. If a software company didn't have the manpower to port their whole product to the new processor, they could at least port the performance-intensive parts first. I remember that a good portion of System 7.5 was still 680x0 code running in emulation even on the PPC. Same goes for altivec optimization. It only pays out for, say, 10% of the code, so why bother with the rest, since the G4 can behave as a perfect G3?

This helps migration a lot. (Similar to introducing the carbon lib very early and providing classic mode in OS X.) I guess Apple would do the same thing with a 64-bit-processor: Create an environment to run your 32-bit-stuff, but make it possible to hand-taylor specific subroutines to run natively on 64-bit for performance reasons...

Yes, but I don't think the operating system can (or should) cope with a process changing from 32bit to 64bit mode. This would mean a lot of work for the scheduler. Besides, I for one think that the biggest benefit of 64bit processing isn't necessarily performance but the possibility to address more than 4GB memory. I am a developer and I have a lot of memory hungry applications running (not to mention the system I am developing) so even if I currently can't have more than 1GB physical memory on my PowerBook I would certainly like to have more virtual memory instead of getting "Out of memory errors"...

/ Nerm

pianojoe

Oct 10, 2002, 08:36 AM

Originally posted by nerm

Well, if you would write down memory locations from a 64bit program to a file and try to read it on another (32bit) computer you couldn't do anything with them since they're locations in another (physical) memory. If you would try to read the file from a 32bit application on the same computer you still wouldn't be able to get any usage from the memory addresses since they're in different processes and therefore protected by the operating system...

Yes, but I don't think the operating system can (or should) cope with a process changing from 32bit to 64bit mode. This would mean a lot of work for the scheduler. Besides, I for one think that the biggest benefit of 64bit processing isn't necessarily performance but the possibility to address more than 4GB memory. I am a developer and I have a lot of memory hungry applications running (not to mention the system I am developing) so even if I currently can't have more than 1GB physical memory on my PowerBook I would certainly like to have more virtual memory instead of getting "Out of memory errors"...

/ Nerm

I completely agree with you. Thank you for proving my point, that not all software written for 32bit can be used on a 64bit machine, no matter how clever the "emulation handler" will be.

(Again, this is what people want: Buy the latest Mac, copy over your applications, documents and prefs folder, and continue working with minimal downtime.)

Rocketman

Oct 10, 2002, 08:40 AM

Originally posted by Dave Marsh

I am disappointed to hear that this chip may not be in production for another 8-9 months. From prior posts, it appears this chip has been available (at least in small quantities) for testing for many months. I wonder what's been taking so long to get the final version right? I too am looking forward with great interest to what IBM has to say next week.

Steve's constant search for the next great thing or a processor that is "insanely great" coupled with Apples need for a large stock of CPU's at the moment of product release combine to make this condition.

Steve wants a million 2.0 Ghz G5's and IBM can supply a range of CPU's from 1.0 to 1.4 Ghz now. To get closer to insanely great Steevie-poo has to wait and thus makes us wait too.

Maybe by then hypertransport, firewire2 and other enabling technologies will also be ready.

Ah, the problems of MASS MARKET supercomputers.

Rocketman

prewwii

Oct 10, 2002, 08:52 AM

Originally posted by bretm

Me thinks your math is a little off. The G4 IS over 3 years old. Sept '99.

When I was a kid one of the advertiser for the Friday Night fights was General Electric. One of their lines was something about 10 years ago 90% of their products didn't exist. In the PC world that product life cycle time has shorten, for most vendors, to less that 3 years. Apple got caught with their proverbial equipment in a wringer by using Motorola as their primary vendor and has not kept up to the industry standard.

Apple has another problem. They have few second sources for their critical parts. Further Apple's customers do not have a second source for their computer needs. When new hot products come out Apple often lacks the capacity to meet customer demand. Imagine what would happen if Apple was able to just double their market penetration and they were able to bring new products to market at an industry standard rate. Apple would be grid lock with their present business model.

I think Apple's business model will prevent them from ever being a sustained industry leader in performance. Apple has been behind for over a year and this thread talks about another year for this new processor to come on line. The rest of the PC industry does not have Apple's loyal customer base. They have to innovate or suffer the rath of fickled customers who have the dollars to buy innovation some place else.

Around here folks buy PC's because their neighbor did. They are just doing email, surfing the web and balancing their check book. Nothing special. Yet Apple can not crack that market with a $1,000...$2,000 iMac because Apple is an unknown commodity that nobody can help them with when it breaks. In my opinion Mac is not paying attention to their market in either the performance or common use arena's. G4's for 3 years....... common on..... it's getting so marriages don't last that long.

ryan

Oct 10, 2002, 09:20 AM

Originally posted by P-Worm

For example, we have an elephant at the circus that can jump through hoops (Bare with me here...)
The real problem is, elephants can't jump.
;)

nerm

Oct 10, 2002, 09:28 AM

kenohki

Oct 10, 2002, 09:29 AM

Originally posted by pianojoe

I completely agree with you. Thank you for proving my point, that not all software written for 32bit can be used on a 64bit machine, no matter how clever the "emulation handler" will be.

(Again, this is what people want: Buy the latest Mac, copy over your applications, documents and prefs folder, and continue working with minimal downtime.)

UGH! This is not how it works! The architects of PowerPC had enough foresight to see 64-bit computing coming. The PowerPC ISA (Instruction Set Architecture) is 64-bit. All PowerPC processors follow this ISA. The difference between a 32-bit number and a 64-bit number is the precision with which it can be "described". The more ones and zeros you have to work with, the more combinations of ones and zeros you can have. A 64-bit number is a combinarion of ones and zeros that is 64-bits long. Obviously you can have exponentially more combinations of ones and zeros with 64 bits to work with than you can with 32 bits. Thus you can be more precise about describing your number. But a number is still a number and a command is still a command. They're just different data types. The ISA was created so that the processor "knows" this.

Thus, 32-bit applications run NATIVELY on a 64-bit PowerPC. There is no "emmulation handler" required. You may have heard stuff about Intel moving to a 64-bit processor that requires emmulation but that's because Intel is moving to a completely different ISA. Opteron (AMD's 64-bit chip) on the other hand, is seriously being looked at because it is a 64-bit extension of the x86 ISA. Meaning, it allows 32-bit x86 applications to run natively and also allows you to take advantage of 64-bit addressing with new or rewritten applications. Thus, your investment in software is preserved.

Apple doesn't need to worry about moving to a 64-bit PPC as much as they need to worry about backward compatibility. Once developers start to take advantage of 64-bit addressing and precision, your installed base of 32-bit processors is subject to the dustbin because of software support. So, most software unless it explicitly needs the advantages of being able to work with huge numbers will probably still be "32-bit". For those that will benefit from "big math", you'll probably see developers compile both a 32-bit and 64-bit version and package them inside one Mac OS X bundle. Then you maintain backward compatibility with your installed base for new or enhanced apps, and still get the Mac OS simplicity of a drag-n-drop install.

I have a Sun/SPARC/Solaris (as well as Apple ;) ) background. Sun made the transition to a 64-bit implementation of the SPARC ISA when it introduced UltraSPARC I. All the software that ran on the 32-bit SuperSPARC, TurboSPARC, HyperSPARC etc. ran just fine on the 64-bit UltraSPARC. Besides some different core files in Solaris, all you had to do was "buy the latest Sun, copy over your applications, documents and prefs folder, and continue working with minimal downtime." In fact, most SPARC applications are still 32 bit because it's easier to maintain one binary since they run full throttle on either the 64-bit or 32-bit implementations. Most other companies' transitions to a 64-bit processor were relatively painless. I'll bet that Apple's will be even smoother. No worries here.

gropo

Oct 10, 2002, 10:38 AM

........Most other companies' transitions to a 64-bit processor were relatively painless. I'll bet that Apple's will be even smoother. No worries here. Yes, THANK YOU... Finally someone who knows what's up! I read over the dev docs @ the ADC regarding the G4 fpu addressing protocol - 32-bit fpu calls are referred to as "single" and 64-bit calls are referred to as "double," indicating that the groundwork has already been laid for the advent of 64-bit integer register space.

nawk

Oct 10, 2002, 10:45 AM

Here is what IBM says about the processer:

Maintain binary compatibility for both 32-bit and 64-bit applications with prior PowerPC and PowerPCAS systems: Several internal IBM task forces in the first half of the 1990s had concluded that the PowerPC architecture did not have any technical impediments to allow it to scale up to significantly higher frequencies with excellent performance. With no technical reason to change, in order to keep our customers software investment in tact, we accepted the absolute requirement of maintaining binary compatibility for both 32-bit and 64-bit applications, from a hardware perspective.

See for yourself:
http://www-1.ibm.com/servers/eserver/pseries/hardware/whitepapers/power4.html

ffakr

Oct 10, 2002, 11:21 AM

Originally posted by barkmonster

At the time G3s were at 266Mhz and the Pentium II and K6 cpus were at 600Mhz.

There wasn't ever a time when 266MHz G3s competed against Pentium IIs and K6s at 600MHz. The only desktop that the 266MHz G3 shipped in was the Beige tower and the iMac (iMac may be the only exception to my argument here). The Beige G3 competed against the PII 233-333MHz machines. Apple moved to the 100MHz bus right after Intel and the 300+ MHz Blue and White G3s competed with the 400+ MHz PIIs (though slower PIIs, LX based machines, were still being sold for a while).
Aside from this, the PII NEVER ran faster than 450MHz anyway. The PIII came to market at 450MHz, replacing the PII.
The K6/2 ran at similar speeds during the processor race, though sometimes clocking slightly faster. It was WAY easier to overclock since Intel started locking the multipliers on the Pentium cartridges. The K6/2 stayed in production far longer than the PII since AMD didn't have a moble version of the Athlon for a LONG time. This is why K6/2s reached higher clock speeds... up to 550MHz, though they didn't get that fast until after the K7 was available.

Saying that and being WAY too optimistic, when a next generation powermac comes out and intel keep using design compromises to push their clockspeeds higher we could be in a situation of apple running those snail ads again by this time next year.

Imagine it :

5Ghz Pentium 4, 1 instruction per clockcycle, huge power requirements and elaborate cooling methods.

2Ghz PowerPC, 8 instructions per clockcycle, low power and turbine cooled. [/B]

I'd say that IS way too optimistic. The P4 can execute 3 operations per clock cycle. It may average slightly over 1 operation/clock but it can do more.

The PowerPC64 platform is supposed to be 8-way superscaler though no one seems clear on what IBM means by that (I'm not a CPU designer). The definition of superscaler that I've seen simply say that superscaler processors are designed to run multiple instruction at once. I've not seen any wording indicating that x-way superscaler means it can run x operations per clock cycle. I've also not seen any wording that says it can run a sub-set of x possible operations per cycle.

Look at the current G4 for example. It is superscaler and it can issue 3 or 4 operations max per clock cycle. It can issue one operation in any one of 8 area's fo the CPU [load/fetch, integer, fp, fp, altivec, altivec, altivec, altivec]. Does this mean the a G4 is 3 way superscaler or is it 8 way superscaler? I've heard both arguments and I honestly don't know which is true.

If you consider both definitions you'll realize that definition one, the 8 instructions/clock, would require a MASSIVE processor to pull off. There would be multiple load/fetch units, multiple integer, multiple fp, and altivec on the chip. It'd be a monster. This would mean that it would be very big, very hot, and very expensive. This isn't the ideal chip for Apple... It'd never go in a laptop.

If you consider definition two, the 8 instruction units with a subset of instructions/clock ala the current G4, that would be much easer to pull off. The chip size would be manageable. It could be made to run more efficiently than a G4, higher average # of instructions/cycle. It would be more powerful as a result. It won't run 8X as many instructions/cycle as a P4 though.

I think that IBM's next chip will be able to issue a subset of 8 possible instructions per clock cycle. We won't know for sure until next week though. We also won't know what the average number of instructions/cycle actually is for quite a while.

ffakr.

Santiago

Oct 10, 2002, 01:01 PM

Originally posted by MacCoaster

Uh. The entire M68k family is 32bit, even the original Motorola 68000.

Right, but the early ones only had 24-bit addressing, and some programs used the upper eight bits of the address word for things that would be very bad if the entire 32-bit word was being interpreted as an address.

As to the PowerPC, there should be no need to change the OS at all. The PowerPC was always designed to be implementable as either a 32- or 64-bit chip, with 64-bit implementations being backwards compatible with 32-bit code.

DakotaGuy

Oct 10, 2002, 01:20 PM

Even if Apple has to sacrifice chip design to get ultra high megahertz they need to do it. If you want to sell computers you have to have the most megahertz. Alti-vec...and all this other stuff Apple has been feeding us will not improve speed only more Megahertz actually makes a faster computer. I love my Macs but am thinking PC more and more because they are SO much more powerful right now. It costs thousands more for a 1.25 dual then what you can get a cheap PC for that is going to run a lot faster. Heck I even saw a $399 PC with a 1.7 GHZ processor in it...that has more speed and will kick the **** out of a PowerMac.

I love my Apples, but am afraid that they are not as good of values as they used to be. And also am afraid that this new IBM chip will actually be slower then the current ones, since according to many on these boards, Apples have actually been becoming slower and slower over each revision. I have thought about replacing my iMac DV 400 with a new eMac 800, but from what people are saying it does not sound like it will be any faster.

I have never used a PC and have always enjoyed Macs, but acording to even most Mac people PC's and XP is better now. Too bad, I like supporting the little company.

When I see a $399 PC with 1.7 GHz and a $3,500+ Mac with only 1.25 GHz, then I can see these peoples points.

I guess I am just getting down on Apple, because other people are.

iShater

Oct 10, 2002, 02:48 PM

Originally posted by Abercrombieboy
Even if Apple has to sacrifice chip design to get ultra high megahertz they need to do it. If you want to sell computers you have to have the most megahertz. Alti-vec...and all this other stuff Apple has been feeding us will not improve speed only more Megahertz actually makes a faster computer. I love my Macs but am thinking PC more and more because they are SO much more powerful right now. It costs thousands more for a 1.25 dual then what you can get a cheap PC for that is going to run a lot faster. Heck I even saw a $399 PC with a 1.7 GHZ processor in it...that has more speed and will kick the **** out of a PowerMac.

I love my Apples, but am afraid that they are not as good of values as they used to be. And also am afraid that this new IBM chip will actually be slower then the current ones, since according to many on these boards, Apples have actually been becoming slower and slower over each revision. I have thought about replacing my iMac DV 400 with a new eMac 800, but from what people are saying it does not sound like it will be any faster.

I have never used a PC and have always enjoyed Macs, but acording to even most Mac people PC's and XP is better now. Too bad, I like supporting the little company.

When I see a $399 PC with 1.7 GHz and a $3,500+ Mac with only 1.25 GHz, then I can see these peoples points.

I guess I am just getting down on Apple, because other people are.

Whoaaaaaaaaaaaaa! Slow down there buddy! :D

1) $399 PC, I won't touch one with a stick, unless I am building it and I got some serious deals on the parts, and I am willing to make major sacrifices.
2) $399 PC will not be faster than your powermac, it will be cheaper, and will have smaller HD, slower components and everything built into the mother board.
3) If you have never used a PC, I don't think you should do a reverse-switch. Most people I know who did that miss their old days when they were Mac users and are too much invested in their current envirornment to switch back.
4) The eMac will smoke your iMac DV 400. If you go by megahertz as you like to believe the speed of a system is measured, then it will be twice as fast as your iMac and would probably cost you less than you paid for your iMac.

Don't believe everything you read, and you DO get what you pay for.

my 2.5 cents.
:D

P.S. You will pay way more than $399 to get it to pefform the way you want it to, and by that time your computing experience would go down the dell ... errr ... toilet. :p

kenohki

Oct 10, 2002, 02:49 PM

ffakr

Oct 10, 2002, 02:53 PM

Originally posted by Abercrombieboy
I love my Macs but am thinking PC more and more because they are SO much more powerful right now. It costs thousands more for a 1.25 dual then what you can get a cheap PC for that is going to run a lot faster. Heck I even saw a $399 PC with a 1.7 GHZ processor in it...that has more speed and will kick the **** out of a PowerMac.
I'm the first to admit that a higher clocked processor is probably going to be faster in most common tasks... a 2GHz P4 will run a bubble sort (or even Netscape) faster than a 1GHz G4.
You are totally on a different page though. You are comparing a 1.7GHz celeron with almost no L2 cache to a a Pro line Mac tower. I guarantee your $399 tower is going to have crap video, crap hard drives, crap sound, and if it even has a NIC.. that is going to be crap too. You're comparing an Escort to a Mercedes. Apple computers are made from very high quality components. There is a reason why I see 8 year old macs still running every day on my campus.

I love my Apples, but am afraid that they are not as good of values as they used to be. And also am afraid that this new IBM chip will actually be slower then the current ones, since according to many on these boards, Apples have actually been becoming slower and slower over each revision. I have thought about replacing my iMac DV 400 with a new eMac 800, but from what people are saying it does not sound like it will be any faster.
I'm not sure where you get these notions from. Apple were NEVER good values. You pay for the user experience, and you pay for machines that are built better. You do pay though, and you always have.

The new IBM chips will NOT be slower... I'll guarantee that. They are built using the Power4 processor as their design reference. The Power4 (though ONLY clocked at slightly over 1GHz right now) is the fastest computer chip in production today. It toasts G4s, P4s, Athlons, Itaniums, HP/RISC (though not by much any more)... Strange how a chip that only runs at 1GHz could be so fast isn't it?? Guess there is more to processor design that speed at any cost.

Also, why would you think that an eMac 800 won't be any faster than an iMac DV 400? The processor is twice as fast (clocked), it has altivec support, it has faster drives, the eMacs video is at least two generations newer [faster] than the iMacDV's video. The eMac isn't going to impress someone who uses a 2.4 GHz P4, but it is a nice machine (I've used them quite a bit), and it is marketed at the low end educational buyer and more recently the low end consumer.

I have never used a PC and have always enjoyed Macs, but acording to even most Mac people PC's and XP is better now. Too bad, I like supporting the little company.

Most people say that XP is much better than other versions of Windows, but most reviews of XP vs. OS X (even by PC magazines) give the edge to OS X. I have seen a LOT of reviews that basically say that XPs user experience is a poor copy of OS X's. I'd, for the most part, agree with that.. though there are things about Windows that I prefer.
As for supporting the company, that is admirable... especially since Apple has always driven innovation in the computer industry. MS would be selling OSes that were still DOS shells if other people didn't push them to make better products. I'm a realist though. I always tell people to get the best product for their needs. If you can't afford a Mac, get a PC. If you have used PCs for years with out major issues, stick with what you know. This isn't to say that I don't think that the Mac is a better tool overall.

When I see a $399 PC with 1.7 GHz and a $3,500+ Mac with only 1.25 GHz, then I can see these peoples points.

Again, you are comparing a single processor Celeron to a dual processor 1.25 GHz G4 tower. You are comparing a bargan bones PC made with the cheepest components to a Pro Mac. I'd wager that if you purchased both and used them equally, only one computer will be useful in 2 years... only one computer will still be running in 3 years. Guess which one?

...just my 2cents.
I am, after all, just a stupid ffakr.

nixd2001

Oct 10, 2002, 03:40 PM

Originally posted by gropo
Yes, THANK YOU... Finally someone who knows what's up! I read over the dev docs @ the ADC regarding the G4 fpu addressing protocol - 32-bit fpu calls are referred to as "single" and 64-bit calls are referred to as "double," indicating that the groundwork has already been laid for the advent of 64-bit integer register space.

i think this might be slightly confused. 32 bit floating point values are traditionally called single precision floats (or "single" or even just "float" in C) whereas a 64 bit floating point value is a double precision floating point value (or "double" in C). Support for both single and double precision FP has been common for ages. It is independent of how many bits are present in integer/address registers.

That said, Apple appears to have been very careful to ensure "type opacity" for pointer values, so they can easily change from being a 32 bit quantity to a 64 bit quantity with not much more than a recompile operation.

Telomar

Oct 10, 2002, 03:44 PM

People keep saying the POWER4 can manage 8 instructions/cycle, it can't. It can fetch 8 and dispatch up to 5 (1 group).

nixd2001

Oct 10, 2002, 03:48 PM

Originally posted by pianojoe

If a program writes out a memory location to a file, the 32bit version will write out 4 bytes, and the 64bit version will write out 8 bytes. The data files will not be compatible without extra effort.

What you say is accurate, but misses the point. Each program should write 32 bits. It's NOT a problem to define data quanties as being 8, 16, 32 or 64 bit independent of the native integer/pointer register size, provided you've not been plain lazy/dumb at the start.

Or put another way, a 64 bit processor can still read/write images (ETC) that a 32 bit processor can read/write and vice versa (etc etc). Otherwise it would really be an incompetent mess!

If there are any issues encountered with specific programs here, the original authors didn't deserve a salary.

penguin341

Oct 10, 2002, 04:06 PM

Tiauguinho

Oct 10, 2002, 04:31 PM

Ffkar,

When they say that this new processor chip will be a 8 way supercalar it means that it can perform 8 indepent instructions per clock cycle. To do this, the processor must have an instruction fetching unit that can fetch more than one instruction at a time, built-in logic to determine if instructions are independent, and multiple execution units to execute all the independent instructions. Of course, that instrucions that depend on the result of the previous instruction can't be processed at the same time. You cant have the chicken and the egg at the same time. First the chicken, then the egg.
The current G4 is a 4 way Superscalar chip.
Imagine this: You have an instruction (a instruction that takes advantage of the superscalar technology) that you would process on both Chips (a 4 way and a 8 way superscalar). The 4 way superscalar was a 1Ghz and the 8 was a 500Mhz... they would both dispatch the instruction at the same time! That's the benefit of the Superscalar technology.

suzerain

Oct 10, 2002, 06:47 PM

Originally posted by ffakr

The new IBM chips will NOT be slower... I'll guarantee that. They are built using the Power4 processor as their design reference. The Power4 (though ONLY clocked at slightly over 1GHz right now) is the fastest computer chip in production today. It toasts G4s, P4s, Athlons, Itaniums, HP/RISC (though not by much any more)...

Not to be argumentative, but this simply isn't true. I totally dig the concept of Macs blowing the doors off Windows machines, but it's simply not going to happen. There's too much inertia on the side of Intel/AMD right now.

I think it's more reasonable to assume these chips (meaning, the IBM GPUL) will benchmark close to current Intels. Look at these SPECfp2000 (http://www.hp.com/products1/itanium/performance/architecture/speccpu.html) numbers.

The Itanium is beating the Power4, though not by tons.

And the important thing to remember is that the new IBM chip IS NOT A POWER4. It is a new chip which takes design cues from the Power4. It is being referred to as GPUL (GigaProcessor UltraLite), which is a reference to it being the "light" version of the Power4.

Basically, that means they scaled it back from the heat-producing electrical black hole it is now into a more reasonable chip that doesn't require you to also build your own nuclear reactor to run it. Then they added on an AltiVec-alike, and appear to be extending the pipeline in order to boost Mhz performance, probably to make it easier to market to the average consumer.

(I don't understand why people on this list constantly use the current Power4 numbers...1-1.3Ghz, when IBM has, itself, said that they expect the chip to DEBUT at 2 Ghz. To me, this means that this processor has a longer pipeline, and will not match the Power4 in terms of operating efficiency.)

Undoubtedly, this chip will be fast, but not as fast as the Power4.

Yes, the Itanium stumbled out of the gate. Yes, moving to the Itanium is going to be a bitch for the PC world, because of addressing incompatibilities. Yes, the PowerPC move to 64 bit should be easy.

But no, the GPUL is not going to "toast" the Itanium in raw performance...the Itanium is pretty bad-assed.

(As a side note, the Itanium ALSO runs at slower clock speeds...1-1.5 Ghz. Intel's going to have to completely shift its marketing strategy if they plan to mainstream Itaniums after they milk the P4.)

Chryx

Oct 10, 2002, 08:17 PM

Originally posted by suzerain
And the important thing to remember is that the new IBM chip IS NOT A POWER4. It is a new chip which takes design cues from the Power4. It is being referred to as GPUL (GigaProcessor UltraLite), which is a reference to it being the "light" version of the Power4.

(I don't understand why people on this list constantly use the current Power4 numbers...1-1.3Ghz, when IBM has, itself, said that they expect the chip to DEBUT at 2 Ghz. To me, this means that this processor has a longer pipeline, and will not match the Power4 in terms of operating efficiency.)

You forget, the GPUL will be
1) smaller (no second core, ?less L2 cache ondie?)
2) made on a finer (.13 or .09) process, the current POWER4's are .18.. and frickin' HUGE as a result.. something in the region of 400mm^2

a much smaller die due to not having a second core, perhaps a bit less L2 cache, and a smaller process.. and it could well be the same pipeline length as the POWER4, less signal propogation issues could alone add a fair bit of headroom.

infact, I'd put money (but not much) on it being a POWER4 design with the second core chopped and altivec/VMX execution units added.

and a lone POWER4 core @ 2Ghz is (theoretically) a very fearsome piece of silicon indeed.

Now, the memory bandwidth will be scaled back from the POWER4, and I'd imagine L3 cache for it in Macs won't top 4MB or so, so it'll perform "differently" to a real POWER4, slower on _large_ bandwidth bound workloads, but possibly faster on stuff that'll fit in the L1/L2/L3? caches. (see 500Mhz Celeron/Mendicino verses 400Mhz Pentium 2/Deutches as an x86 example of that behaviour.)

ffakr

Oct 10, 2002, 08:25 PM

Originally posted by suzerain
Not to be argumentative, but this simply isn't true. I totally dig the concept of Macs blowing the doors off Windows machines, but it's simply not going to happen. There's too much inertia on the side of Intel/AMD right now.

I think it's more reasonable to assume these chips (meaning, the IBM GPUL) will benchmark close to current Intels. Look at these SPECfp2000 (http://www.hp.com/products1/itanium/performance/architecture/speccpu.html) numbers.
I can't get to that link for some reason. It's OK to be argumentative too... it's a rumor discussion list :-)
And... if you look back at my original statement (which you quoted), I was refering to the contention of the previous poster.. that the PowerPC64 will be slower than the G4. I said that I guaranteed that whatever IBM has up it's sleeve won't be slower than the G4. You twisted what I said. :-)
The Itanium is beating the Power4, though not by tons.
It wasn't the last time I looked, but I haven't checked out the SPEC on Itanium2 yet.

And the important thing to remember is that the new IBM chip IS NOT A POWER4. It is a new chip which takes design cues from the Power4. It is being referred to as GPUL (GigaProcessor UltraLite), which is a reference to it being the "light" version of the Power4.

Basically, that means they scaled it back from the heat-producing electrical black hole it is now into a more reasonable chip that doesn't require you to also build your own nuclear reactor to run it. Then they added on an AltiVec-alike, and appear to be extending the pipeline in order to boost Mhz performance, probably to make it easier to market to the average consumer.

(I don't understand why people on this list constantly use the current Power4 numbers...1-1.3Ghz, when IBM has, itself, said that they expect the chip to DEBUT at 2 Ghz. To me, this means that this processor has a longer pipeline, and will not match the Power4 in terms of operating efficiency.)
I don't know either. I didn't use 1 GHz in reference to the upcomming PowerPC64, I only used that number to point out that a processor (the Power4 in my example) can be very very powerful with out being clocked very very fast.

It may have a longer pipeline, though IBM was never a big fan of long pipelines in the past. evne the latest PPC750s don't have very long pipes... I think they are like 7 stages in the 750fx but don't quote me on that. As far as I know, pipeline is not the only element affecting clock speed. It's a way to boost it, but the overall design is going to affect how fast the chip clocks too.
Undoubtedly, this chip will be fast, but not as fast as the Power4.
I don't think it will be as fast as the Power4 either. I don't think I ever said that. It won't be competing against the Power4 though, just like it won't be competing against the Itanium or the PA-RISC (which I call the wrong name earlier :-(
This is a small server/workstation processor. Merced (and the Power4) are still being marketed as big iron, or at least mini-big iron.. if that makes sense.
I do, however, think it will be fast. Really fast. I think it will be a smoking processor because:
1) It takes design cues from the Power4, but it will build on the Power4 design. It isn't unreasonable to expect that next gen processors benefit from lessons learned on the last project. Sure it will be a 'lite' processor, but it will benefit from all that IBM has learned from the Power4 (and the power3, and the 750s, and the 604... and so on).
2) If it can really do multiple fetch/stores and issue as many ops per cycle as people say, it will be much more efficient than the x86 processors it will compete with. I recall seeing that the P4 really doesn't average much more than one operation per cycle in the real world. If you put a PowerPC that could _potentially_ execute 3 or 4 times as many operations per cycle as a P4 that was clocked twice as fast as the PowerPC... guess which processor will perform better?
3. it will retain Altivec which has always seemed to perform better than SSE2. Many of the SSE2 instructions take MANY more cycles to complete than the same or similar operation in Altivec.Yes, the Itanium stumbled out of the gate. Yes, moving to the Itanium is going to be a bitch for the PC world, because of addressing incompatibilities. Yes, the PowerPC move to 64 bit should be easy.
It's not addressing incompatabilities that will hold up acceptance of EPIC. EPIC is totally different from CISC (x86 in this case). Itaniums are VLIW processors. They are HUGE, hot, and they require exquisitely complex compilers to write fast code. Itanium is probably the most complex processor ever created and many people think that the compiler writers will need to work harder than the chip engineers to make Itanium work well.
But no, the GPUL is not going to "toast" the Itanium in raw performance...the Itanium is pretty bad-assed.

(As a side note, the Itanium ALSO runs at slower clock speeds...1-1.5 Ghz. Intel's going to have to completely shift its marketing strategy if they plan to mainstream Itaniums after they milk the P4.)
Itanium doesn't run at 1.5 GHz. I'm pretty sure Itanium 2 is only available at ~900 and 1000MHz so far.

But who cares if it toasts the Itanium. I didn't say GPUL will toast itanium.. I have no Idea about that. I said that the power4 toasts Itanium **which as you tried to point out, may not be true anymore**
If GPUL ends up in macs next year, it won't compete with itanium. I don't expect to see Itanium in the average computer until well after Deerfield arrives. I think the software will hold up acceptance of Itanium at least as much as the hardware will. Intell will push x86 at consumers for AT LEAST two more years, probably longer.
PowerPC64 will go up against P4 and Athlon. It will likely also go head to head with clawhammer. This is where the comparisons need to be made. From what we think we all know, I think it will take them all. clawhammer is the real unknown. It has only sampled at 800MHz so far so who knows how it will perform when it comes out of the gate (about mid way through 2003).
If the PowerPC can do as much work as people claim it can (per cycle), I think it is going to put a lot of PCs to shame.

This is all, of course, only my opinion. Afterall, what do I know, I'm just an average ffakr with the same sketchy info and innuendo that everyone else has. :-)

suzerain

Oct 10, 2002, 08:40 PM

Originally posted by Chryx

You forget, the GPUL will be
1) smaller (no second core, ?less L2 cache ondie?)
2) made on a finer (.13 or .09) process, the current POWER4's are .18.. and frickin' HUGE as a result.. something in the region of 400mm^2

a much smaller die due to not having a second core, perhaps a bit less L2 cache, and a smaller process.. and it could well be the same pipeline length as the POWER4, less signal propogation issues could alone add a fair bit of headroom.

So, what sort of relationship is there between the manufacturing process and speed? Does it have a greater impact on speed of the chip, or power consumption? Or is it one of those situations where the designers have to choose where to take their advantage (i.e., going to a smaller manufacturing process allows you to choose either lower power consumption or added speed or some combination thereof)?

I ask, because is it reasonable to assume that going from a .18 micron process to .13 (or .09) will automatically allow them to boost speeds from 1.3Ghz to 2.0 Ghz without extending the pipeline?

I'm just wondering if you could illuminate on that a bit...on how much speed such a change in manufacturing would allow you to gain (all theoretically, of course, since this whole discussion is basically about vaporware, until we see IBM talk about it).

suzerain

Oct 10, 2002, 08:57 PM

Originally posted by ffakr

This is all, of course, only my opinion. Afterall, what do I know, I'm just an average ffakr with the same sketchy info and innuendo that everyone else has. :-)

Yeah, well, this is all our opinions, since we have no real data, save for cryptic messages about a product announcement.

However, I didn't "twist" anything you said; I just quoted it. :p

My real point could have been summed up this way:

IBM has said [paraphrasing] that they are going to discuss a new chip which is based on the Power4, but it is destined for desktop workstations rather than servers.

To me, "based on" means, pretty specifically, that it is NOT a Power 4. Otherwise they would have just said, "We're announcing a new version of the Power4 for workstations."

It's obviously different enough that they bothered to make that distinction.

Therefore, accepting all of that logic, it doesn't seem to me that it's particularly helpful to look at Power4 benchmarks AT ALL, since the Power4 is a big fat power-wasting behemoth of a chip that is used in $15,000 servers.

Obviously, the GPUl will be a very different beast indeed. And, as the other poster who responded to my post said, it may well be true that the chip outperforms the Power4 for smaller operations, but would probably be toasted by it on applications where it has to process gigantic amounts of data.

Then again, maybe not. But how would IBM continue to sell servers based on the Power4 if they had a cheaper chip that outperformed it?

So, since the Itanium 2 is edging the Power4 on benchmarks, I think it's also doubtful that the GPUL will outbench that chip, either.

Instead, I'd guess that we'll see performance similar to a Pentium4, but with better FPU perfformance.

And, IMO it's also unlikely that it'll blow the doors off Intel...notice how the whole industry is considering a 64 bit leap at the same time? I think these developments all kinda happen together...

It would seem reasonable to assume, though, that a 2 Ghz GPUL would be able to be considered an equal to a 3-4 Ghz Pentium4, which is likely what'll be on the other side of the fence at the time.

Of course, I'd like it to be...oh I dunno...1,000 TIMES faster than the then-fastest P4...

Chryx

Oct 10, 2002, 09:25 PM

Originally posted by suzerain

So, what sort of relationship is there between the manufacturing process and speed? Does it have a greater impact on speed of the chip, or power consumption? Or is it one of those situations where the designers have to choose where to take their advantage (i.e., going to a smaller manufacturing process allows you to choose either lower power consumption or added speed or some combination thereof)?

I ask, because is it reasonable to assume that going from a .18 micron process to .13 (or .09) will automatically allow them to boost speeds from 1.3Ghz to 2.0 Ghz without extending the pipeline?

Well, I'll give you an x86 example (since their processors are the ones I'm most familiar with, my knowledge of which PPC chips were which process is kinda flaky)

the "Coppermine" Pentium 3, on .18, topped out at ~1.05ghz (that's as high as overclockers could push them in a stable manner with anything other than EXTREME cooling, Intel tried to push it to 1.13Ghz and then had to recall it as the review samples they'd sent out didn't work properly. :p)

the "Tualatin" Pentium 3, on .13 (with twice the L2 cache, 512K) tops out at ~1.7Ghz (though Intel cut the P3 off at 1.4Ghz to avoid making the P4 look too pathetic ;))

the Tualatin is a more complex design due to the extra cache, and yet it gained ~700Mhz of headroom from the die shrink.

If the GPUL is.. half a Power4, and it's on a smaller process, I'd expect it to massively gain headroom.

As for heat verses clock, it works something like
smaller process = generates less heat at whatever clockspeed BUT then gets clocked higher which balances out the overall heat output.

Chryx

Oct 10, 2002, 09:31 PM

Originally posted by suzerain
To me, "based on" means, pretty specifically, that it is NOT a Power 4. Otherwise they would have just said, "We're announcing a new version of the Power4 for workstations."

It's obviously different enough that they bothered to make that distinction.

Or it could just be that they want to label it as a PowerPC chip, rather than a POWER chip?

Since the POWER4 is a PowerPC-AS chip anyway, could it be that the distinction between the two is merely the intended market?

Ah well, we'll know more on the 15th.

(in other news, did the macrumors server just start a nightly cron job to back everything up or something?.. it's getting very unresponsive from where I'm sitting.)

Telomar

Oct 10, 2002, 09:36 PM

Chryx

Oct 10, 2002, 09:47 PM

The POWER4 is more powerful than the Itanium 2 in the real world. Furthermore Spec isn't the greatest benchmark of all times and in practice when you are looking at picking between those two chips people rarely base their purchasing decision on those numbers (you'd be pretty dumb if you did).

Indeed, but there isn't really anything that qualifies as a "perfect" benchmark, and unless someone has any better suggestions, SPEC serves as a halfway decent "it's fast.. no.. it's slow.. no, wait, it's fast!" test.. even if it is rather compiler dependent :)

All that said what we will see Oct 15 won't be a POWER4. What I expect it will be is a core very similar to the POWER4 core with less L2 cache, a SIMD unit and a new memory subsystem. I think they'd be silly not to plan dual core but they may have other factors preventing them.

The top factor there would be die size/yield

Plus, the interconnects required to keep a pair of cores properly fed with data would be.. excessively complex for a desktop part (high pin count on the processor package, large number of traces on the mainboard.. etc)

It's a different enough chip that you would say it is POWER4 derived and certainly would be incredibly powerful. Most of the cost reduction would come through just removing the large caches the POWER4 uses.

the L3 cache on the POWER4 is external, although 128MB of EDram can't be cheap.

the majority of the price reduction for the processor itself will come from

1) less complex design (single core, less cache on die)
2) process size reduction.

Any guesses at L3 cache speeds for GPUL?, I'm thinking Powermacs with 4MB of DDR Sram at 1/4th the clockspeed? (or perhaps 1/3rd?)

zerokelvin

Oct 10, 2002, 11:56 PM

I don't know that it's really just as simple as all this discussion.

The G4 has a 64-bit core, a 64-bit data bus, a 36-bit (64G?! John Holmes got nuttin' on you!) address bus (along with a bunch of other specialized control doodadz), 128-bit path to L3, a 128-bit butt whoopin math chip - and an arbiter chip between the two CPUs in the current systems fat piped to RAM, where the 2.7GB/s bandwidth is shared with all the fancy new schmangled DMA (A dirty trick SGI taught everybody with their Revolution in Octane arch.) devices hung off an UberBridge.

The G4 on the CPU side is not going any further 'physically' without major chicanery.

All WPs (with the exeption of Book E which we don't care about (what's that for? WAPtv?) on G5 were pulled from HelloMoto well over a year ago (anyone remember the writs on the working protos with the skinny mic fab?) - speculatively, since this is a rumor site, due to issues with getting RapidIO to play well with AltiVec and MPX. Second runner up might be that their having difficulties with running G3/G4 user mode binaries.

I'd bet they do it, and I wouldn't count HelloMoto out just yet. Too much of the ground work is already lain, and the big strength of the Power4 lies in cascading near infinite arrays of CPUs (anyone remember adding mass quantities of 604's in AS/400s?). This doesn't do us much good when we're using AMD's commodity memory architecture.

Apple could definitely do some magic with BSD/Mach, but they need to be able to continue to make cheap computers for cheapo customers. I predict that you may see a stress fissure turn into a rift were OS X gets ported to some high dollar warez - but the consumer side can never get away from widely available commodity sub-parts... same ones Wintel board designers are using.

Apple used to try and compete by making premium warez, but the Great God Cheapoquatl kicked them in the junk.

If you've read this far you're crazy as a loon. Thank you.
c

Chryx

Oct 11, 2002, 01:36 AM

Originally posted by zerokelvin
The G4 on the CPU side is not going any further 'physically' without major chicanery.

They aren't trying to take the <b>G4</b> any further, they are looking to replace that router-controller-with-an-oversized-ego completely with something derived from a high-end mainframe chip.

Or is that chicanery? :p

BTW, the POWER4 might scale well in an SMP environment, but it's still damned impressive on a per-processing-core basis, hopefully the GPUL will preserve that aspect of it's performance.

Telomar

Oct 11, 2002, 05:24 AM

Originally posted by Chryx
Indeed, but there isn't really anything that qualifies as a "perfect" benchmark, and unless someone has any better suggestions, SPEC serves as a halfway decent "it's fast.. no.. it's slow.. no, wait, it's fast!" test.. even if it is rather compiler dependent :)No there are no perfect benchmarks beyond testing the system specifically for task you have planned. Of course when testing desktops you usually have dozens of applications planned and performance can vary.

I have a particular dislike for SPEC though. I've seen too many funny things done with it to really have any faith in it.

The top factor there would be die size/yield

Plus, the interconnects required to keep a pair of cores properly fed with data would be.. excessively complex for a desktop part (high pin count on the processor package, large number of traces on the mainboard.. etc)Die size wouldn't be too big an issue. It would be around the size of the initial PIV's probably.

I agree the main disadvantage with a dual core chip would be yield. If one core doesn't work the chip is no longer dual core. You can sell it as a single core processor but still you've just wasted money adding the second core that doesn't work. That's really what I expect would be the driving force behind a decision not to produce a dual core GPUL.

Memory handling/interconnects are another problem but in all honesty I wouldn't consider that the worst.

On the other hand from a design and cost point of view there can be substantial benefits in moving to multi-core chips. Over the coming years chip manufactures are increasingly going to be talking about it for desktop chips (Intel I know are looking at it and IBM's views are well known).

If you were going to begin with any chip it would make sense to do it with one derived from a dual core chip as well.

the L3 cache on the POWER4 is external, although 128MB of EDram can't be cheap.

the majority of the price reduction for the processor itself will come from

1) less complex design (single core, less cache on die)
2) process size reduction.

Any guesses at L3 cache speeds for GPUL?, I'm thinking Powermacs with 4MB of DDR Sram at 1/4th the clockspeed? (or perhaps 1/3rd?) I didn't so much mean with respect to the L3 cache itself, although it is expensive.

The L3 controller and a lot of the circuitry due to the complexity of the POWER4's memory system would be simplified in a desktop chip.

The question of L3 cache is an interesting one as it's the area I would expect is most open to change. I doubt 4MB early on although I expect it will probably handle 2 - 8MB and I'd assume 1/3rd clockspeed.

ffakr

Oct 11, 2002, 10:17 AM

Originally posted by Telomar
No there are no perfect benchmarks beyond testing the system specifically for task you have planned. Of course when testing desktops you usually have dozens of applications planned and performance can vary.

I have a particular dislike for SPEC though. I've seen too many funny things done with it to really have any faith in it.

Intel has a nasty habit of optimising their compilers to specifically perform better in SPEC. One can only assume that such funny business hurts performance in other areas at the expense of better spec marks. All the other big iron processors tend to have spec scores submitted from boxes that use common tools.. like gcc to compile the spec suite. ...at least that is what I've always heard.
originally posted by suzerain
However, I didn't "twist" anything you said; I just quoted it.
Yes, you just reposted what I said, but you posted my comments out of context. Your previous post started with a quote from me stating that 'I guarantee that the GPUL will be faster'. You then went on to argue that the GPUL won't be faster than Itanium. My quote was in direct reference to the post I was commenting on, in which the poster asserted that the upcomming IBM processor won't be faster than the current G4. This is a VERY different argument from the one that you insinuated I was making. In fact, I'm pretty sure that I never made any claims about the upcomming IBM chip in comparison to any processor except the G4 and possibly existing 32bit x86 processors.

I'm just trying to be accurate here. :D

[quote]originally posted by zerokelvin

The G4 has a 64-bit core, a 64-bit data bus, a 36-bit (64G?! John Holmes got nuttin' on you!) address bus (along with a bunch of other specialized control doodadz), 128-bit path to L3, a 128-bit butt whoopin math chip - and an arbiter chip between the two CPUs in the current systems fat piped to RAM, where the 2.7GB/s bandwidth is shared with all the fancy new schmangled DMA (A dirty trick SGI taught everybody with their Revolution in Octane arch.) devices hung off an UberBridge.

Owch, hold on. The G4 is considered a 32bit processor. It has 32 bit registers for integers. It can do double precision floats but so can 32bit x86 processors. I certainly wouldn't call the G4 a 64 bit chip.

It also doesn't have a 38 bit address BUS, it has address registers that are 38 bits in size, meaning it could theoretically address quite a bit of memory (though I don't know how to do this in a 32bit ansi C program.

As for the 128 bit butt whompin math chip, it has SIMD registers that are 128 bits wide but the whole point of SIMD is that you perform the same operation on multiple data. You have to have 128 bits of data, at the same time, that needs the same operation done on it. SSE2 on the P4 is also 128 bits wide too.

And, you're being way to nice with your 'fat piped to RAM' comment. The G4 is still only available with a 64bit SDR path to the main chipset and therefore only 64bit SDR access to main system memory.

I'd say that the G4 certainly 'ain't all that' like you seem to think it is. :-(
It's a nice, elegant chip, but it isn't where it should be at this stage in the evolution of the CPU.

DharvaBinky

Oct 11, 2002, 11:00 AM

Intel Corp is to appeal a judge's ruling yesterday that its Itanium processors infringe on patents owned by Intergraph Corp, risking an additional $100m payout on top of the $150m it has already agreed to pay Intergraph, Kevin Murphy writes. Texas District Court Judge John Ward ruled that two Intergraph US patents are "valid and enforceable" and that Intel's products "literally infringe" upon them, Integraph said. Judge Ward also said Intergraph is entitled to an injunction on the Itanium and Itanium 2.

Hrmmm... No Itanium? <grin> I bet intergraph would file the injunction just out of spite. :)

With Itanium gone... that would leave... ummm... AMD and Apple in the 64-bit market (not counting Sun, IBM, and their high end stuff).

Sooo since AMD has had such a hard time in the market, this would bolster *both* Opterons and G5s software support.

Ultracute...

Binky

beatle888

Oct 11, 2002, 03:08 PM

i dont see intel dropping their itanium project.
they will simply just pay the other company off.

ffakr

Oct 11, 2002, 10:19 PM

I agree, there won't be any injunction... Intel has put BILLIONs, with a big fat B, into the itanium. They will pay off Intergraph long before it gets that far. They are currently appealing anyway so that right to injunction is most likely on hold.

The whole intergraph/intel thing is pretty funny. I highly recommend taking the time to read up on the whole conflict between the two companies.

It reads like a classic david and golaith story. Intel screwed Intergraph hard and often. Intel essentially forced Intergraph out of the hardware business. They were stealing their IP during this time and they continued to for long afterward (still do). The thing is, Intergraph keeps winning court cases against them. It's like the big bully who is threatening the littler kids, but the bully is really stupid and gets caught by the teachers every time.

zerokelvin

Oct 13, 2002, 01:15 AM

...durned cat.

Anywayz...

I'd rather see improvements in SMP implementation from the software side both at Apple and with 3rd party developers at the next expo even more so that faster hardwarez.

Mac users would stand more to gain, and I'm pessimistically assuming that it's too much to ask for both,
c:p

Chryx

Oct 13, 2002, 06:31 AM

Originally posted by zerokelvin
...durned cat.

Anywayz...

I'd rather see improvements in SMP implementation from the software side both at Apple and with 3rd party developers at the next expo even more so that faster hardwarez.

Mac users would stand more to gain, and I'm pessimistically assuming that it's too much to ask for both,
c:p

The thing that would REALLY help the SMP performance would be a less crippled path between the processors and main memory, a 1.3GB/s link for _two_ processors to a 2.7GB/s memory system is the height of inefficiency.

compare to the Athlon/Alpha SMP model where each chip has dedicated link of 2.1GB/s (or more for the Alpha and AthlonXP 2700+/2800+) to the memory controller, which then has 2.1GB/s or more of bandwidth.

sure, there's still the potential for them to fight over the memory bandwidth, but at least the bottleneck is moved as deep into things as possible, instead of being stage #1 in the process.

steve53e

Oct 13, 2002, 08:27 AM

Interesting that HP would choose to switch the computers and OS'es from one test to the other while all others remained the same.

Chisholm

Oct 13, 2002, 02:38 PM

Originally posted by ffakr
I agree, there won't be any injunction... Intel has put BILLIONs, with a big fat B, into the itanium. They will pay off Intergraph long before it gets that far. They are currently appealing anyway so that right to injunction is most likely on hold.

The whole intergraph/intel thing is pretty funny. I highly recommend taking the time to read up on the whole conflict between the two companies.

It reads like a classic david and golaith story. Intel screwed Intergraph hard and often. Intel essentially forced Intergraph out of the hardware business. They were stealing their IP during this time and they continued to for long afterward (still do). The thing is, Intergraph keeps winning court cases against them. It's like the big bully who is threatening the littler kids, but the bully is really stupid and gets caught by the teachers every time.

I have a friend/ coworker that worked at Intergraph for 7 years or so before being laid off. He said the company was always having money problems.

Not that it has anything to do with this, but its funny that a company makes more money in court than producing a product. God bless America.:p

MOM

Oct 13, 2002, 05:10 PM

check out http://www.forbes.com/technology/newswire/2002/10/13/rtr749520.html
MOM

beatle888

Oct 13, 2002, 05:28 PM

thanks MOM :D

" IBM said the new PowerPC 970 microchip is a "lite" version of its Power4 chip..........IBM said its new PowerPC chip would go into production late next year and process 64 bits of data at a time at 1.8 Gigahertz, or 1.8 billion cycles per second.........An industry source said Cupertino, California-based Apple would use the chip in its Macintosh computers."

:p

scem0

Oct 13, 2002, 07:16 PM

I don't understand how they can already know the speed of the chip when it won't be in a computer till a year from now...

Chryx

Oct 13, 2002, 07:42 PM

Originally posted by scem0
I don't understand how they can already know the speed of the chip when it won't be in a computer till a year from now...

IBM most likely have them up and running in the labs.

how the media know?.. I have no idea.

scem0

Oct 13, 2002, 07:48 PM

Yeah, I am very sure they have them running in labs. But in 2 months wont they have 2 GHz, and in 4 months 2.5 GHz? And then they can manufacture a whole bunch of 2.5 GHz pmacs in august and september and release them to the public in october/november.

Chryx

Oct 13, 2002, 07:53 PM

Originally posted by scem0
Yeah, I am very sure they have them running in labs. But in 2 months wont they have 2 GHz, and in 4 months 2.5 GHz? And then they can manufacture a whole bunch of 2.5 GHz pmacs in august and september and release them to the public in october/november.

Not really, semiconductor R&D is always at least a year ahead of what we can buy.

eg, Intel recently demonstrated a P4 @ 4.7Ghz, fastest P4 we can buy is 2.8Ghz

nixd2001

Oct 13, 2002, 07:55 PM

Originally posted by scem0
I don't understand how they can already know the speed of the chip when it won't be in a computer till a year from now...

Because it's not all shoot-and-hope. A lot of effort is spent "characterising processes" - learning what a given level can do and tweaking it to improve it. IBM will likely have targets for the capabilities of the chips to be produced by the new Fishkill fab line (assuming this is where they are made) and, given IBM has a reasonable amount of experience in this sort of area, their targets are probably achievable (if not already proven on small fab runs). I imagine that when Ford or GM design a new car engine, they can make a pretty good prediction of how powerful it will be before they build the first one.

The variation is most likely to be in the speed yields. That is, they will probably produce chips graded at (say) three different speeds. Know the relative ratio of these speeds in advance is where experience will really come in to play. And if something goes wrong, they'll just have more low end parts than they wanted. But the press release will still claim all three speeds, even if the top end one is in "short supply".

Or, in short, it's not totally random - experience counts a lot.