PowerBook G5... Not Yet...

[wrldwzrd89]

<snip>
Simply confused here. The FSB on the PPC970 is not HT. There is no HT bus on the PPC970. IBM doesn't use HT in any shipping PowerPC chip.

Not even AMD uses HT for the FSB on the Opterons and other x64 chips. HT has too much latency for a memory FSB - the Opterons have an on-die memory controller so that the FSB runs at up to 2.4 GHz and 128-bits wide.

A question for you, AidenShaw:

DDR2 and the forthcoming DDR3 have much more latency. Wouldn't these be suitable for HyperTransport if the net latency is the same? I'm thinking that the HT latency and the RAM latency happen at the same time, so the net additive latency is zero.

 

[matznentosh]

If the heat is the problem with the G5, then Apple should find a way to get the heat to recharge the battery, just like a steam engine... 😛

First new idea I've read in a long time. It's called a Sterling engine. Sounds like a good idea to me, like a self winding watch.

 

[AidenShaw]

A question for you, AidenShaw:

DDR2 and the forthcoming DDR3 have much more latency. Wouldn't these be suitable for HyperTransport if the net latency is the same? I'm thinking that the HT latency and the RAM latency happen at the same time, so the net additive latency is zero.

Latencies are almost always additive, especially for reading.

You have to pay the HT latency penalty first, before the request is even seen by the memory DIMM. Then you pay the DIMM penalty, and the HT penalty again for the return trip with the data.

You can get good bandwidth by overlapping latencies, so it is possible to have a high latency, high bandwidth memory system. It's not possible to combine a number of high latency components and get a low latency, high bandwidth memory system.

Of course, YMMV - some applications are more sensitive to bandwidth, and don't mind latency. Others are highly sensitive to latency. (This is one reason why multi-CPU Opteron makers usually highlight the bandwidth benchmarks, and downplay the latency benchmarks - the Opterons are NUMA systems where some memory must be accessed over the coherent HT inter-CPU links.)

For writing, the CPU can post the write and go on, and let the write "slowly" work its way down the busses to the DIMM. (Of course, we depend on cache-coherency and other techniques to ensure that a read of that location gets the new data from the pipe and not the old data from the DIMM.)

Also note that the "worse latency" of DDR2/DDR3 is in cycles - so a "CAS 2" 400 MHz DDR and a "CAS 3" 600 MHz DDR2 actually have the same latency.

Just ask yourself why none of the major HT proponents are using HT as the FSB???

 

[DVNIEL]

Couldn't apple just take the dual-core G4 processor and just call it a G5? =)

 

[wrldwzrd89]

Latencies are almost always additive, especially for reading.

You have to pay the HT latency penalty first, before the request is even seen by the memory DIMM. Then you pay the DIMM penalty, and the HT penalty again for the return trip with the data.

You can get good bandwidth by overlapping latencies, so it is possible to have a high latency, high bandwidth memory system. It's not possible to combine a number of high latency components and get a low latency, high bandwidth memory system.

Of course, YMMV - some applications are more sensitive to bandwidth, and don't mind latency. Others are highly sensitive to latency. (This is one reason why multi-CPU Opteron makers usually highlight the bandwidth benchmarks, and downplay the latency benchmarks - the Opterons are NUMA systems where some memory must be accessed over the coherent HT inter-CPU links.)

For writing, the CPU can post the write and go on, and let the write "slowly" work its way down the busses to the DIMM. (Of course, we depend on cache-coherency and other techniques to ensure that a read of that location gets the new data from the pipe and not the old data from the DIMM.)

Also note that the "worse latency" of DDR2/DDR3 is in cycles - so a "CAS 2" 400 MHz DDR and a "CAS 3" 600 MHz DDR2 actually have the same latency.

Just ask yourself why none of the major HT proponents are using HT as the FSB???

That makes sense, AidenShaw. I'm now enlightened as to why HT doesn't appear as a dedicated FSB.

 

[sinisterdesign]

whew!

i for one am glad that there's still no G5 PB, now my 400MHz TiMac won't seem outdated.

😛

 

[AidenShaw]

That makes sense, AidenShaw. I'm now enlightened as to why HT doesn't appear as a dedicated FSB.

Note also that on the multi-CPU Opterons the "remote" reads that go over the HT are cache-line operations, not individual reads.

Therefore a "remote read" buffers a number of memory accesses into a single HT transfer, thereby mitigating some of the HT latency.

For example, assuming 128-bit memory and 64-byte cache lines, a remote read is

- left CPU sends request for address X over the HT link to right CPU

- right CPU gets request, starts read
---- reads 16 bytes at X to buffer
---- reads 16 bytes at X+16 to buffer
---- reads 16 bytes at X+32 to buffer
---- reads 16 bytes at X+48 to buffer

- right CPU sends 64-byte buffer over HT to the left CPU

- left CPU gets data into the local cache, and resumes the thread which is stalled on the memory read​

Note that even though the 4 memory reads are local (to the right CPU) and can be overlapped to hide latencies, measurements show that a dual-CPU Opteron has about 50% higher latency for remote reads vs local reads.

 

[darkwing]

Just a word of advice...

Don't go making statements like these without backing them up.

For example, such a statement cannot be verified or discounted without knowing what version of the JVM was used on the Mac and on the PC.

And this matters because? Obviously any Mac user in this forum is going to be using the latest version of anything offered by Apple. Currently that's 1.4.2. Even if the Windows version is superawesomeomgwonderful I'm still using the best Apple gave me. It isn't good enough.

 

[darkwing]

Are you freeagin kidding me??? Not that much faster?? OK... a few points to consider on purchasing a PB G4 today, or a PB G5 Q2 or Q3 at the latest:

• A G4 is obsolete the second you buy it. Clearly 32 bit apps are not the future, and they will drop off in numbers incredibly in the coming 3-4 years as 64 bit G5s are plentiful, cheaper, and in large circulation. Why invest in a time bomb?
• G4s can address more memory than the PB can hold, that's not the issue.
•*The G5 bus isn't just merely faster, it's FOUR - SIX TIMES faster, *NOT* including the fact that it can pull 2x32bits.. 32 this way, 32 that way, at the same time... which definitely clears up the FSB rainbow swirls crunch!
•*Next G5 revisions get AltiVec 2. Don't expect any new improvements to the aging G4 architecture.. FreeScale already promised they aren't going to change the bus format (they prefer their slug to HyperTransport!)
• IBM already said it has system-on-a-G5-chip designs in testing... which means a massive reduction in hardware size, number of components, heat production, and an increase in overall reliability. Yes, they implied it was especially going to help Apple.. as in, it's for Macs. Watch and wait.. PBs may shrink to iMac Mini size plus a whale of a screen attached.. or become the back side of the screen, like an iMac G5 lite, to go... (likely if battery tech improves massively)..
• The lowest available clocked G5 is faster than the fastest available clocked G4. That's no joke, and it would produce far less heat to under clock a G5 than max out a G4... and it would be far more reliable in terms of MTBF! (Mean time between failures).

>... I can wait for the G5.. I aint buying anything 32bit again... and I don't want to be tethered to a desk either. (I shouldn't be surprised that I've seen a couple people lugging their iMacG5s around.. that 64-bit HyperTransport FSB is really hard to give up once you've tasted it.)

The G5 has a much wider pipeline than the g4. Remember that preesntation a couple years ago showing why the 867 (or whatever it was) g4 was faster than a 1.7 ghz P4 because everytime you had to do a pipeline flush, it would take more cycles on the P4? Well, the G5 has an even larger pipeline than the P4 which means that it takes that much longer. That's the reason you need all that bus bandwidth. The G5 requires extremely well scheduled and optimized code in order to stay really fast. I'm not saying that it's slow, but I'm saying that a big chunk of that faster FSB is for this reason.

 

[tdewey]

The G5 has a much wider pipeline than the g4. Remember that preesntation a couple years ago showing why the 867 (or whatever it was) g4 was faster than a 1.7 ghz P4 because everytime you had to do a pipeline flush, it would take more cycles on the P4? Well, the G5 has an even larger pipeline than the P4 which means that it takes that much longer. That's the reason you need all that bus bandwidth. The G5 requires extremely well scheduled and optimized code in order to stay really fast. I'm not saying that it's slow, but I'm saying that a big chunk of that faster FSB is for this reason.

I think a G5 PB would be great. But what I really want is a better GPU. I'd much rather have a G4 PB with a 128/256MB MR9800/X800 (or NVIDIA equivalent) than a G5 PB with an MR 9600 or NVIDIA Go 5200 or whatever.

The MR 9700 is nice, but really just an incremental upgrade to the 9600. I was hoping that Apple would ship the new PBs with a 9800 (given that the chip has been out six months) or better chip.

I use my computer for work & writing, surfing and WoW. So a G5 doesn't do all that much for me.

T

 

[Caiwyn]

Couldn't apple just take the dual-core G4 processor and just call it a G5? =)

Nope. One of the big selling points of the G5 is that it is a 64-bit processor. The dual-core G4 is not. There would be way too much confusion about system requirements if they did that.

 

[wrldwzrd89]

And this matters because? Obviously any Mac user in this forum is going to be using the latest version of anything offered by Apple. Currently that's 1.4.2. Even if the Windows version is superawesomeomgwonderful I'm still using the best Apple gave me. It isn't good enough.

I can't necessarily assume you're running Panther, though - since you wouldn't be running Java 1.4.2 in that case. Plus, some people have incentives NOT to upgrade. All this means I can't just assume you (or any other Mac user) is running Java 1.4.2.

Anyway...

I guess I'll have to do some testing and report back, since I happen to have a Windows machine in front of me (I'm typing this on my Mac).

 

[Slipperyskates]

Ok, so I keep reading all these posts about people whining about a lack of G5 PowerBook. Can someone honestly tell me WHY they think they *need* one? I mean, really, people. Are you rendering raytraced scenes for an upcoming Pixar film on your PowerBook and really need the extra speed? What would a G5 chip *really* give you over the current 1.67GhZ G4?

I'm serious, I really want to know. I'm planning on getting a G4 PowerBook soon, and I'm curious as to why people seem to think they need the G5. If you're doing serious crunching of numbers and heavy utilization of the processor, perhaps a laptop is not what you need.

Is it me? Am I missing something here? I'm being honest, here. I'm not trying to start a flame war or anything; I'm seriously curious. Thanks in advance for honest replies!

Theres just something very very cool about the characters G5. I mean, people are tired of looking at the G4s, and being able to say that you have a G5 powerbook really brings ou the "wow factor."

I know its totally irrational, but most marketing is.

 

[Surreal]

The G4 PBs are terrible DAW, they are fine to just run sequences that you've already optimized in 'the studio', and control something like a nord3 or whatever, but they can hardly run the demanding plugins and software combinations that most contemporary engineers/artists use. On the other hand, the P4 I auditioned at a friend's house (2.53, 533fsb) could run almost 25 instances of the Minimoog vsti within a sequencer host. My G4 with Logic6 can run 1-2max, and even then it often stutters or crashes. That's pathetic for a product marketed to the professional audio/video user. So, yes, there are a lot of professionals and even students out there waiting for a mobile power solution for Osx, -the way the G5 is clearly from the ground up optimized for one thing: running the most demanding Osx applications of now and the future. This simply hasn't happend yet on a mobile platform, the G4 was a nice step up back in Os9, (alti-vec optimized, Premiere!) but its not enough to be a mobile alternative to the PMG5, which the G4PB clearly was. Remember the old ads "A realtime mobile FCP workstation." You don't see those anymore for the PBs... More, its "look how sexy this is for buisness people/english majors!" but I digress, I merely meant to point out that this exact wierd ignorance of the very real professional mac user seems endemic of this whole category of thread. There ARE users that NEED a G5 or a power-equivalent similar solution, and they don't have a capable notebook just now.

if you have ever messed with Propellerhead's Reason, i need to know whether running logic and reason as a slave will be worthwhile on the updates.

i know that after a couple of virtual instruments are added on an emac 1.25 it isnt that responsive ...that is without reason running in the background and without much midi routing...

 

[Surreal]

8 * 24bit * 192000/sec --> 4.4MB/s

surely any FSB will eat that for breakfast 😉

anyway, i think you can get better clarity with 44.1kHz sessions if you just buy better preamps. and i'm not guessing here. take a focusrite octopre for example, which is not even very high-end unit, but it just kicks ass recorded at 44.1kHz compared to the motu at 192k, assuming that we are not comparing the +20kHz frequencies 😀

(i dislike logic, too. protools is my choice.)

no i mean i am going to buy logic pro 🙂

and will the pb run smoothly with all of that audio going in and out i mean.. it CAN get the audio in and out ... but i get slowdown with multiple virtual instruments... recording glitches would ruin a session.

 

[Surreal]

on-location recording and live front-of-house mixing with protools. that's what i do on a weekly basis. i run my G4 laptop about 85% cpu stress for at least four hours every gig and cannot afford a single hiccup. i would really appreciate a good speed boost for powerbooks, as it's really not feasible at all to carry desktops on the road.

what powerbook do you use and how many tracks on average? and do you think that the updates are purchase worthy or should i buy some used one and wait.

 

[OldTimey]

8 * 24bit * 192000/sec --> 4.4MB/s

surely any FSB will eat that for breakfast 😉

anyway, i think you can get better clarity with 44.1kHz sessions if you just buy better preamps. and i'm not guessing here. take a focusrite octopre for example, which is not even very high-end unit, but it just kicks ass recorded at 44.1kHz compared to the motu at 192k, assuming that we are not comparing the +20kHz frequencies 😀

Jfreak is right here. recording in 192khz without using a decent pre is overkill. the difference between 44.1k and 192k is very difficult for most people to hear. not like 16bit vs. 24bit. besides at 192k, without a UAD or HD accel processor card, many computers will choke once you start adding plugs.

(i dislike logic, too. protools is my choice.)

couldnt keep this to yourself huh? you just had to insult the dude's choice of daw, didnt you? not suprising coming from a slave to pro tools.

(death to digidesign, too. digital performer is my choice.)

 

[all-devourer]

Talking with a friend in Austin at the G4 fab, he said they got first silicon of what sounds like the single core 8641(first Freescale chips at 90nm) in December. Haven't heard anything concrete about the dual core yet, but he said they should have first run on those "soon".
I'm expecting the dual core G4s to be commercially available the latter part of this year to the first part of next year.

Z

I'm almost positive that the parts your friend told you about are 7448s. It's the first 90nm part coming out of Freescale, meant to sample in 1Q 2005.

 

[blizzard_nz]

Heat Problems

How about running some sort of cooled fluid pack over the G5 chip. Not sure how you'd cool the fluid but there must be something and someway to possible implement it into the powerbook.

PS: I'd quite happliy purchase a 1 and a half inch think g5 PB if thats what it takes.

 

[Furrybeagle]

Just to point something out...

Also, let us not forget that Doom 3 supposedly requires a G5 chip. :-(

Go to Asypr and check the Doom 3 Requirements. Or just look here:

Preliminary System Requirements (these may change before the game's release):
# Operating System: Mac OS X 10.3.6 or later
# CPU Processor: PowerPC G4/G5 or later
# CPU Speed: 1.5GHz or faster
# Memory: 384 MB or higher
# Hard Disk Space: 2.2GB free disk space
# Video Card (ATI): Radeon 8500 or better
# Video Card (NVidia): GeForce FX 5200 or better
# Video Memory (VRam): 64 MB or higher
# Media Required: DVD Drive

Which means that it will even get by on a Revision D 12" Al PB, along with the other PowerBooks. Not that it matters as I'm waiting for the new formfactor before I switch. *Stares at the Dell in front of him and thinks of broken back* 🙁

 

[darkwing]

I can't necessarily assume you're running Panther, though - since you wouldn't be running Java 1.4.2 in that case. Plus, some people have incentives NOT to upgrade. All this means I can't just assume you (or any other Mac user) is running Java 1.4.2.

Anyway...

I guess I'll have to do some testing and report back, since I happen to have a Windows machine in front of me (I'm typing this on my Mac).

I didn't realize that 1.4.2 was only available to Panther users. Sorry about that. 🙂 Yes I find that the JVM apple provides is extremly slow. My 1.5 ghz powerbook should run things much faster than a pentium 3 windows box. It should not be spinning 100% of the cpu and running jerky while the p3 chugs along at 10%.

Steven

 

[ShnikeJSB]

ATI MRX800...

Although preliminary tests have shown ATI's Mobility Radeon X800 (M28) to be friggin' fast as all heck, here is a little something I pulled from Beyond3D.com's preview of the card:

"Whilst the thermal conditions of Mobility Radeon 9700 saw it end up in a wide variety of solutions, right down to fairly thin notebooks, heat generated from the Mobility Radeon X800, and from the high performance CPUs that are also going to need to be supplied alongside such parts, are going to ensure that these will be in big chassis and, more often than not, need fairly impressive cooling for a mobile solution."

In other words, since Apple just HAS to keep the PowerBooks trim and light, I HIGHLY doubt we will be seeing the MRX800 in ANY powerbook. This pisses me off, as I TOO am waiting for better graphics in the PowerBook before I buy one (one of several reasons for waiting to replace my TiBook/667). We might have to settle on the 9700 for awhile, though they MAY be able to peruse the MRX700, which is due to be out soon. Just pointing this out to everyone like me who wants to actually play games on their 'book -JB

 

[JFreak]

what powerbook do you use and how many tracks on average? and do you think that the updates are purchase worthy or should i buy some used one and wait.

protools six is not so optimized, so in real-world single-G4@1.5 is faster than single-G5@1.6 HOWEVER, that may very well change with protools seven. number six is after all the first version on osx, so it's not yet as good as it gets. currently digital performer performs much better, if pure plug count is most important factor.

anyway, the new powerbook models look promising, but if you get a killer 2nd hand deal, just buy one used and spend the rest for something more useful.

i have a 1.25GHz powerbook, with 1GB ram and 7200rpm internal hard drive. i also use lacie firewire drives via dedicated firewire buss installed into pcmcia slot and couldn't be happier about hard drive performance. i have a digi 002R connected to the builtin firewire and with that unit i can only get 16 inputs plus one external reverb via spdif i/o. i however drive 8 outputs as well - one stereo pair is for FOH sound, one stereo pair is for wireless monitoring, and four mono outputs for regular monitors.

i use focusrite octopre and presonus digimax preamps. nowadays i tend to use waves plugs less if not at all, and favor urs and mcdsp stuff instead. pspaudioware also has some very nice plugs, especially their delays kick ass.

 

[Lacero]

I'm going to wait for Doom 3 Arena. Not too fond of single player games.

 

[zac4mac]

All-devourer - thanks, my friend is understandably vague about these things. I'm a little cautious too after the Moto "G5 fiasco" of several years past(2001 or 2002?). He told me of seeing this new chip run and it was awesomely fast but never came out to play due to "Errata". Then the whole project got killed.
When I asked directly about the dual core, the timeframe from him was late05-early06 availability.

Z