IBM Producing 65nm Chips?

[DeepIn2U]

What No street cred??

WHat no credit for posting this info before Macrumors?? CHeck my last post and the time for it, my source was from ircspy.com. heuer007, not sure when you posted info on this cannot tell from your previous 5 posts, but hey you should get credit as well for finding out this info and bringing it here.

Now on topic, what I gathered from this was, smaller chips and relative heat problems solved as normally gaming systems don't get much cooling and run for hours at maximum performance. Also the technology that IBM learns from what goes into the neXT Box, and especially from the multi-core tech that is going into the PS3 the "Core" will more than likely go into the PowerPC for Macs. Or is all of this technology already from the hyped PowerPC 5??

I'm expecting faster 32-bit emulation, along with more efficient computations from slightly higher clock speeds, and inroads to see more recognition for Apple that their cpu (IBM) is the best way to go for game/office apps/research/music/security processing!! Yes I know that music and lately research has always been better on a Mac, but its about time analysts investors and programmers (web site, c++, etc) learn this and further port or start their projects first on Mac.

 

[foniks2020]

Originally posted by Henriok
What family of processors is that? IBM doesn't have any other family of processors than PowerPCs. It doesn't say that the Xbox 2 will be PowerPC, but it will be based on PowerPC. I see no reason why the Xbox 2-processor won't be a PowerPC chip, but they very well might add some cool custom technology. I count on it. They will certainly add som obscure technology to make piracy harder.

Well it *could* be based on a Power4/5 chip but damn if that wouldn't be overkill and the heat diss.. would be impossible, so yeah PowerPC it is.

But IBM doesn't need to add anything funky to prevent piracy, look at Apple, they've never had a hacked ASIC ROM (I think that's the term) which is the chip on the MBoard that makes Apple Macs proprietary, in effect. SO M$ just has to do what they've already done and add their little chip that's already in the XBOX or an equivalent that works with PPC and all is secure.

 

[ffakr]

Re: What No street cred??

Originally posted by Prom1
Now on topic, what I gathered from this was, smaller chips and relative heat problems solved as normally gaming systems don't get much cooling and run for hours at maximum performance.

wattage is just amperage x voltage. smaller ships should require lets current but the smaller processes tend to 'bleed' more electrons so they may not use significantly less power. This is why Intel's Prescott won't likely be any cooler P4 (and it will be bigger of course).
.065 micron _should_ be quite a bit cooler though, if they solve the problems associated with such a small process.

Also the technology that IBM learns from what goes into the neXT Box, and especially from the multi-core tech that is going into the PS3 the "Core" will more than likely go into the PowerPC for Macs. Or is all of this technology already from the hyped PowerPC 5??

Not much is known about the PS3 cpu yet, but I've heard that it may not be a multi-core processor in the traditional sense. Rather, Cell may have multiple specialized cores. This is oversimplifying.. but imagine an integer math processor that did int work really well.. and a vector processor (like supercomputers use) or a DSP for mpeg decoding... and then think of each of these cpu's tacked together on one die. Cell might be like this, but who the hell outside sony and IBM knows :shrug:

I'm expecting faster 32-bit emulation, along with more efficient computations from slightly higher clock speeds..

It's entirely possible that if the processor incorporates different 'cores' on the xbox2 cpu, it could add logic to speed the emulation of x86 code. Heck, it could even support some instructions natively.
Remember, IBM added hardware to the first PPCs to speed the emulation of Motorola's 68k code.. even though that was a CISC processor family like x86. Original PPCs often ran 68k code faster than the cpus they replaced and they weren't clocked significantly faster.
The xbox2 cpu will have to emulate a P3 733-800MHz (I've heard its actually a celeron). This wouldn't be too hard on a .065 micron PPC that would likely run at least twice that fast.. if it had hardware support for emulation that is.
It may even be easier to do hardware emulation now than when the first PPCs ship. To keep x86 alive, AMD and Intel actually design their CISC processors to function in a very RISC-like way at their cores. Complex CISC instructions are broken down into simple micro-ops inside P4s and Athlons. The 970 also breaks down PPC instructions into micro-ops. IBM could provide hardware support to decode x86 instructions into a compatible PPC micro-op so that it could run x86 code at nearly native speed.
How cool would that be?
That's something I'd like to see funnel back into the desktop processors. :smile:

 

[tortoise]

Re: What No street cred??

Originally posted by Prom1
I'm expecting faster 32-bit emulation, along with more efficient computations from slightly higher clock speeds, and inroads to see more recognition for Apple that their cpu (IBM) is the best way to go for game/office apps/research/music/security processing!! Yes I know that music and lately research has always been better on a Mac, but its about time analysts investors and programmers (web site, c++, etc) learn this and further port or start their projects first on Mac.

Actually, the G5 is less than stellar for many types of research and high-performance computing purposes, and for a very specific reason. The PPC970 has mediocre memory performance (particularly compared to the top-notch memory performance of AMD64), and for any kind of large memory app or app that uses the memory subsystem hard, many x86 derived systems will run rings around the G5 because of this weakness in the PPC970 design. For our high-performance research systems, Opteron systems school the G5 clock-for-clock. Why? Because none of our apps require much in the way of DSP but they do require transforms on large and fine-grained data structures.

PPC970 rocks for DSP-like computations (e.g. audio, video, and useless benchmarks like LINPACK), and the AMD64 architecture has wicked fast memory. Pick the processor according to your actual needs. The G5 has strengths and weaknesses, and is clearly mediocre in some applications.

 

[jero]

i love how the ps1 outsells the xbox in japan.

 

[Rocketman]

Re: IBM Producing 65nm Chips?

Originally posted by Macrumors
Team Xbox claims that IBM will be producing 65-nm chips for the upcoming "Xbox 2" and start delivering commercial parts in the next 12-18 months.

It seems that 65um is here. Apple is always 3+ months behind the curve on adoption due to testing and packaging.

Powerbook G5 soon. Microsoft having priority is not a problem as they tend to place full orders in advance and Apple more buys as it goes.

Rocketman

 

[GregA]

Originally posted by MacsRgr8
he he.. wouldn't it be gr8 if XBox 2 games were playable on a G5?

<can't imagine M$ ever tolerating that... but we can dream>

I'm surprised MS doesn't sell "Virtual XBox" for Windows... and then of course next year "Virtual XBox2 for Mac". Surely they could sell some software that would make them some nice side profit.

Originally posted by nighthawk
They had a retail PPC version of Windows NT all the way up to NT 4.0. The problem was that there was so little software written for the PPC version, so few people bought it.

Didn't the Alpha version run Windows-x86 Apps through emulation? Certainly, MS has the pieces to make emulation work now.

It would be very interesting if IBM (like Digital) subsidised Windows-for-PowerPC development, to sel their own PowerPC computers. IBM's PowerPCs, bundled with Windows XP (or MacOS X?).

 

[AidenShaw]

FX!32

Originally posted by GregAussie
Didn't the Alpha version [of NT] run Windows-x86 Apps through emulation? Certainly, MS has the pieces to make emulation work now.

The Digital Alpha systems had a subsystem called FX!32.

It had an x86 emulator that ran 32-bit x86 apps on the Alpha NT systems. (16-bit apps were emulated by the NTVDM (or WOW) subsystem)

Even better, it had a translator, compiler and optimizer. When you first ran an app, it was emulated. After you exited, the translator looked at the code paths that you ran, and converted those to optimized native Alpha code.

The next time you ran the app, those code paths would run at native speed. When you hit new code, the emulator would kick in - but then later the offline optimizer would convert the new stuff to native code.

Another great optimization was the fact that you were running Win32 applications - when the application called a Win32 (or other NT) routine, there was no need to emulate or translate - FX!32 would just jacket the x86 call and do a direct transfer to the appropriate Alpha native NT function.

Unfortunately, the Pentium Pro was a big fraction of the performance of the Alpha systems, for a much smaller fraction of the cost of the Alphas. Running x86 code on the Alpha really sucked for price-performance, and there wasn't enough native Alpha NT code to compensate.

I'm not sure where the FX!32 IP ended up, but Intel bought Digital's compiler assets, and FX!32 was close to the compiler group. I wouldn't be surprised to hear that Intel's new software "x86 on Itanium" offering is based on FX!32 technology.

 

[Zech Marquis]

Gamecube--HA!

I've seen a few people here bragging about the little Gamecube,oh please! It's a kiddie game console. No online gaming, no hard drive, and NO compelling grown up titles. And yes, I'm a proud Mac owner, as the Mac simply won't see the likes of Doom III until...? Halo 2--nope, althugh the mac Halo does have online play now. Xbox will soon have Ninja Gaiden, DOA Online Ultimate, Full Spectrum Warrior, Sudeki, Fable, Stacraft Ghost...you get the idea.

Now, back on th main topic--yes the Xbox 2 will have a variant of the G5--with some custo designed features IBM hasn't revealed to the public yet. 65 nm will be very helpful for the evetual Powerbook G5s ad iMac G5s.

A few more Cube talk--when Nintendo has something like Doom III, then we can talk. Until then, just enjoy your little Mario Gokart, and tell mommy you need some more milk and cookies. I'm going to hit the slopes in Amped 2!

 

[sanford]

Re: Gamecube--HA!

Originally posted by Zech Marquis
I've seen a few people here bragging about the little Gamecube,oh please! It's a kiddie game console. No online gaming, no hard drive, and NO compelling grown up titles.

Right, no hard drive. There is a network adapter for online gaming, though; just few online titles and only one that I know of outside Japan. And Eternal Darkness is one of the most compelling "grown up" titles ever created. Metroid Prime is no slouch, either.

GCN has an extremely solid if somewhat limited catalog of gaming titles, some specifically for children, some for adults and some that can be enjoyed by both.

That being stated, I do play my Xbox more than my GCN, likely due in large part to Xbox Live.

 

[ffakr]

Re: Re: What No street cred??

Originally posted by tortoise
Actually, the G5 is less than stellar for many types of research and high-performance computing purposes, and for a very specific reason. The PPC970 has mediocre memory performance (particularly compared to the top-notch memory performance of AMD64), and for any kind of large memory app or app that uses the memory subsystem hard, many x86 derived systems will run rings around the G5 because of this weakness in the PPC970 design. For our high-performance research systems, Opteron systems school the G5 clock-for-clock. Why? Because none of our apps require much in the way of DSP but they do require transforms on large and fine-grained data structures.

PPC970 rocks for DSP-like computations (e.g. audio, video, and useless benchmarks like LINPACK), and the AMD64 architecture has wicked fast memory. Pick the processor according to your actual needs. The G5 has strengths and weaknesses, and is clearly mediocre in some applications.

Both AMD64 and the 970 systems produced by Apple have more cpu bandwidth than the memory contained in them. A dual proc G5 has 2 independent CPU buses each comprised of 2 unidirectional 32bit wide buses at 1GHz (16GB/sec) while the dual channel DDR 400 has 6.4 GB/sec of bandwidth.
http://www.apple.com/powermac/architecture.html
The Athlon64 has an on-die controller so the memory bus runs at CPU speed.
I think what you are trying to refer to is the decreased latency of the Athlon64 compared to the G5 system, not bandwidth... since both architectures have more than enough bandwidth with existing memory technologies.
Saying the G5 has mediocre memory performance is fud and crap and you should know this if you rely on machines for research.
From what I've seen ( http://www.fftw.org and BigMac and work done in house ), the PPC 970 seems to do just fine as a high performance research machine.

Could you possibly tell us why, exactly, you've come to believe that a) the G5 has a mediocre memory subsystem, and b) what you've done to try and optimise your benchmarks on the G5?
One of our unix admins are currently evaluating machines to run in a cluster. He's told me that they've seen a 35% increase in performance on the G5s after trying out Beta 6 of IBMs xlf compiler (compared to F77). He's also said that they are currently having a lot of trouble benchmarking the Opteron which may be a bug in x86 GCC (generating faulty code - underflow issues??)

BTW, Linpack solves linear equations... yet the G5 excellent linpack performance isn't relevent to its use in research?
huh? Maybe you were thinking about the Pro-PC benchmarks of MS Word when "useless benchmarks" come to mind.

 

[ffakr]

Re: FX!32

Originally posted by AidenShaw
Unfortunately, the Pentium Pro was a big fraction of the performance of the Alpha systems, for a much smaller fraction of the cost of the Alphas. Running x86 code on the Alpha really sucked for price-performance, and there wasn't enough native Alpha NT code to compensate.

FX!32 was cool stuff. I remember reading about it back in the day. I'd always heard that it never fully lived up to it's promise though.... I don't remember if it generated flaky code or if it just didn't run as fast as hoped.
I don't recall PPros being all that cheap though. I do remember them being quite a bit more than Pentium systems, even though they didn't run all that great unless they had 32bit clean systems (they ran 16bit code like crap, which Win95 contained). As a result, it was NT or Linux only for PPros. .. but I'm getting OT...
I'm pretty sure you could get single proc Alpha's back then for not much more than high end single proc PCs.. like $2500. I remember because I wanted one to play with and I was surprised that it wasn't totally out of my range.

speaking of 'where is it now?'. If you think about it, the Transmeta chips act a lot like FX!32. They translate from one ISA to another, and they optimise the translation as the code is run.

 

[ffakr]

Originally posted by jero
i love how the ps1 outsells the xbox in japan.

well, to be fair, Sony has a cult following in Japan. This is true of a lot of 'retro tech'.
Also, XBox games tend to be developed for the US market. There are a lot of Japanese only PS/PS2 games but that's not true the other way around. Also, I don't think that Japan is as big into the architypal US games (First person shooters and such).

 

[iMeowbot]

Re: Re: FX!32

Originally posted by ffakr
FX!32 was cool stuff. I remember reading about it back in the day. I'd always heard that it never fully lived up to it's promise though.... I don't remember if it generated flaky code or if it just didn't run as fast as hoped.

Honestly I think the biggest problem is that it was too good at what it did! It allowed ISVs to be lazy; since the IA32 version of a program ran well enough on AXP, even if not up to full potential, why bother running the code through a native Alpha compiler and have to support an additional set of binaries? This, IMHO, is what made Alpha NT not worth the bother for DigiPAQ to continue developing.

 

[sanford]

Originally posted by ffakr
well, to be fair, Sony has a cult following in Japan. This is true of a lot of 'retro tech'.

Oh, ffakr, but you can't be fair because anything Microsoft does most be monopoly-grinding garbage! Seriously, you're right: just as Nintendo's Super Famicom once held the hearts and minds of millions of Japanese, now does Sony. And developers continue, even stateside, to produce new games for PS1 -- usually at bargain prices compared to top-tier current generation titles. Why the Intellivsion Lives and Midway Arcade Treasures nostalgia collections aren't PS1 rather than PS2 titles is beyond me. Creating the Sony console versions for PS1 would have made the product compatible with both systems (PS2 being backward compatible with PS1, of course).

You're also right about retro-tech being de rigeur in Japan. Space Invaders is not a video game so much as it is a mark, a mission, a brand, a scar... Hmm... Apparently a Dashboard Confessional record. Anyway, I wouldn't quite yet classify PS1 as retro-tech as it is not only still in production but still has active developers. In video games, the Japanese have rather consistently been less impressed than the West with wowser graphics effects as they have gameplay, keeping the PS1 a viable option for the Japanese household: tiny, simple and a huge library of games. The Japanese go for all the hot cutting-edge tech in their mobile phones!

Note that in the odd hours of the day, it's not unusual to find yourself playing Crimson Skies on Xbox Live with a match full of Japanese opponents. Skies is the sort of game that has Japanese appeal. Xbox is making inroads in Japan. It's the only US-made console to have ever done this well in the East. It's hardly a failure.

It would be somewhat pleasant if a) all posters to this forum could look at Microsoft as a developer of technology -- some good, some bad, some ugly; and b) the current crop of video game consoles as having their unique strengths and weaknesses and not all bad except the particular unit one owns.

 

[Photorun]

However, XFlops was such an initial disappointment in Japan for sales of units, M$ had supposedly predicted selling 3 million in the first month or something but it wasn't even 1/6th that, at one point M$ was thinking about pulling it from the shelves (which I wish they would have so I could have done a Nelson "Ha Ha"). I tend to think of Xbox's mediocre at best welcome and still by Microcraps estimation underwhelming sales in Japan is just proof that the Japanese have more taste, sense, or brains than Americans/world, at least as far as electronics, and they know their electronics. Any right minded person would see anything with Microsh** on it, in it, or running it and take a sledge hammer to it.

 

[sanford]

Originally posted by Photorun
I tend to think of Xbox's mediocre at best welcome and still by Microcraps estimation underwhelming sales in Japan is just proof that the Japanese have more taste, sense, or brains than Americans/world, at least as far as electronics, and they know their electronics. Any right minded person would see anything with Microsh** on it, in it, or running it and take a sledge hammer to it.

The Japanese personal computer market is overwhelmingly Windows on Intel-compatible just as the rest of the first world, and the second and third worlds for that matter. Following your logic, Linux/Mac OS/Solaris would predominate in Japanese personal computing. They don't, Windows does. QED.

 

[Henriok]

Originally posted by foniks2020
Well it *could* be based on a Power4/5 chip but damn if that wouldn't be overkill and the heat diss.. would be impossible, so yeah PowerPC it is.

Actually.. ever since Power3 all Power-chips are PowerPCs too, so.. For the time beeing, the only other processor-technology IBM have is the one that they use in their zSeries mainframes. But.. It seems that they'll get their share of Power/PowerPCs with the introduction of the Power5.

Come to think of it, the PowerPC 970 (aka G5) is also a Power-chip.

Originally posted by foniks2020
But IBM doesn't need to add anything funky to prevent piracy, look at Apple, they've never had a hacked ASIC ROM (I think that's the term) which is the chip on the MBoard that makes Apple Macs proprietary, in effect. SO M$ just has to do what they've already done and add their little chip that's already in the XBOX or an equivalent that works with PPC and all is secure.

I didn't mean someone would pirate the actual Xbox hardware (make clones), I meant that they could implement stuff to hinder software piracy and stop folks from installing any other OS than whatever Microsoft wants, ie, make it harder to mod them to run Linux and such.

And.. of course implement the Palladium security plattform.

 

[sjonni]

bio

Originally posted by VIREBEL661
I'm no processor manufacturer, or claim to be any kind of expert on that subject, however I've heard we were reaching the limitations of silicon as far as speed and smaller processes. I bet we'll see a revolutionary step in the next couple of years - something that makes current processor tech obsolete. I remember hearing something, but don't know where. For the sake of conversation, perhaps someone with a better memory than I can post some information on emerging processor technologies? Here's to G5 Powerbooks soon (hopefully) - I'll be buying one at that time!

It´s called biological cpu´s. It will not be sillicon but some sort of living cpu. A BOI CHIP! And then it will all end like in the movie T3.

 

[eric67]

Originally posted by Stoffel
I don't bother the xbox2 but hey! do you think that the g5 can make it into the powerbooks before fall 2004?

yes it will be in PowerBook by Fall 2004, for the 2004 education market in September

 

[DeepIn2U]

differences

Tortoise,

I'm just curious of the cpu speeds of the Opterons and the PowerMac G5s that you use. Are they comparable, in the sense that your Opteron's are all Dual 2.2 cpu systems using EEC 333Mhz memory, and the G5's are dual 2.0Ghz?

It was my understanding that a memory intensive app, like say MS Office (I say this because speeds between a P3 1.2Ghz and a P4 1.6Ghz showed no significant increase in performance, actually it was reduced when the P4 debutted) would need faster ram to put/pull instructions in and out of the memory. Hence even if the on core memory controller on AMD Opterons and Athlon64s ran at cpu speed, the latency would still be inherent because the Error Correction of bytes and the access to the actual memory would reduce this reduce latency you speak of on the memory controller.

I see this as a runner still making faster steps on the track, but not getting anywhere faster than the next runner taking longer strides; like runner A (Opteron) takes strides at 1/2 a meter will actually take 2 steps per meter (on board memory controller), yet is no faster than runner B (PPC970), because runner B takes strides at 1 meter (1step per meter). If runner A takes steps twice as fast as runner B, then whom will win a race at say 50/100meters? Both will because the pace will be the same, and this is the reason why the debate is still ongoing by many in the know.
yes a sad analogy but I tried.

cheers

 

[tortoise]

Re: differences

Originally posted by Prom1
I'm just curious of the cpu speeds of the Opterons and the PowerMac G5s that you use. Are they comparable, in the sense that your Opteron's are all Dual 2.2 cpu systems using EEC 333Mhz memory, and the G5's are dual 2.0Ghz?

They are all dual 2.0GHz systems. The RAM in the G5 is theoretically faster, though I'm not sure if this is the case in practice.

Your analogy doesn't work. What typically happens is that the CPU spends a lot of its time doing nothing while waiting for a memory request to be fulfilled. These clock cycles are essentially wasted but the processor is still "busy". The number of clock cycles an Opteron wastes while waiting for a memory request to be fulfilled is much less than a G5 on average. For memory intensive apps that make a lot of memory requests, fewer wasted cycles per request means more work is done and the processor will appear to be faster even though the cores have roughly identical throughput in theory. As is often the case with modern systems, the CPU is starved much of the time and only lives up to a fraction of its theoretical potential outside of ridiculously narrow cases. It isn't that the Opteron has an intrinsically faster core, it is that it wastes fewer clock cycles waiting for work to do. There isn't a computer made today that doesn't have a bottleneck in its memory systems, and performance in the general case is often defined by the depth and width of this bottleneck.

For most applications, there won't be a significant difference between an Opteron and a PPC970, so choose whichever suits you. For DSP applications, the PPC970 will be significantly faster. For large memory intensive applications, the Opteron will be significantly faster. For workstations, there are usually more important factors than these as to which is better. I don't do DSP at work, but I still prefer OSX workstations.

 

[w0nt0n]

Was'nt a version of Window's NT written for loads of other processors sometime ago before the project got canned.

I'm sure Microsoft had versions of NT ported to run of PowerPC Risc processors aswell as MIPs and DEC's Alpha processor.

Because of this i'm sure theres probably alot of code that could easily be updated to run Xbox 2.

I'm presuming that Xbox 1 is running some version of Window's, but I don't really know alot about the console.

 

[cspace]

Originally posted by w0nt0n
Was'nt a version of Window's NT written for loads of other processors sometime ago before the project got canned.

I'm sure Microsoft had versions of NT ported to run of PowerPC Risc processors aswell as MIPs and DEC's Alpha processor.

not only did they have these ports, they actually sold these versions of NT (v3.5)

 

[eclipse525]

Doesn't really matter....I believe 5 years Apple will move on to chips made from carbon Nanotubes! http://news.zdnet.co.uk/hardware/emergingtech/0,39020357,39118949,00.htm

I believe....I believe....OR shall I dare say, laser, crystal and hologram processors.....don't blink, they'll be here before you know it.

Which one do you think will actually become a reality through the industry?


~e