So, 1 PE after-all or is this just for GDC 2005 ?

[kyetech]

Hey! You guys are on an ameeeeeer-i-can board, so start speekin' ameeeeerican!

It's british actually. More or less.

But eveyone knows Britain is just the 52nd state.

<not on topic>

well, we may not be a state, But G W Bush certainly does have a large portion of his right arm firmly up the ass of Tony blaire, Bush controls blaires little mouth by moving his index finger... And takes the notion of vantrilaquism to new levels. We are basically controlled by the most poorly articulated / twangy accented puppet master of our times.

</not on topic>

 

[Megadrive1988]

Panajev, I would like to remind you of the 8-PE, 72 processor CPU described by Mercury News in 2003.

With the PS 3, Sony will apparently put 72 processors on a single
chip: eight PowerPC microprocessors, each of which controls eight
auxiliary processors.

I would be utterly shocked if PS3 CPU consisted of only 1 PE. that is, one PU CPU core and eight S|APUs. i mean what the hell? that would PISS me off. after all the talks over the last two years.

2-4 PEs seems most likely.

i'm not pissed at you Panajev, don't get me wrong, i'm just pissed at the thought of one PE. but oh well, it's just a friggin videogame console. nothing to get too upset over right? just disappointing if Sony goes with a single PE.

worst case senario. Sony goes for a single PE CPU that provides 256 GFLOPs peak performance. the Nvidia GPU is just a modified NV5x with several times the performance of NV40 / GeForce 6800.

that *might* open up an oppertunity (only an oppertunity) for Microsoft to nearly rival PS3 performance with Xenon. especially in light of the fact that Xenon CPU is scalable & modular. like maybe Microsoft will decide to double the reported 3-core CPU configuration to 6. providing ~84 * 2 GFLOPs performance

 

[McFly]

like maybe Microsoft will decide to double the reported 3-core CPU configuration to 6. providing ~84 * 2 GFLOPs performance

They already talk about hundreds of watts and multiple fans ... there's no way they could just double the cores, that would not be that easy.

Fredi

 

[Megadrive1988]

yeah they did, but they did not say how many processors that implementation of Xenon was using. maybe they have already been using twin CPUs with 3 cores each. or fit all 6 cores onto one die. (not too likely but possible)

Yet on the other hand, it is also reasonable to think that Xenon CPU is only 2 or 3 cores.

 

[Brimstone]

Panajev, I would like to remind you of the 8-PE, 72 processor CPU described by Mercury News in 2003.

With the PS 3, Sony will apparently put 72 processors on a single
chip: eight PowerPC microprocessors, each of which controls eight
auxiliary processors.

I would be utterly shocked if PS3 CPU consisted of only 1 PE. that is, one PU CPU core and eight S|APUs. i mean what the hell? that would PISS me off. after all the talks over the last two years.

2-4 PEs seems most likely.

i'm not pissed at you Panajev, don't get me wrong, i'm just pissed at the thought of one PE. but oh well, it's just a friggin videogame console. nothing to get too upset over right? just disappointing if Sony goes with a single PE.

worst case senario. Sony goes for a single PE CPU that provides 256 GFLOPs peak performance. the Nvidia GPU is just a modified NV5x with several times the performance of NV40 / GeForce 6800.

that *might* open up an oppertunity (only an oppertunity) for Microsoft to nearly rival PS3 performance with Xenon. especially in light of the fact that Xenon CPU is scalable & modular. like maybe Microsoft will decide to double the reported 3-core CPU configuration to 6. providing ~84 * 2 GFLOPs performance

Deadmeat was predicting 1 PU with a 128 Gflop rating a while back.

 

[Megadrive1988]

I think Deadmeat predicted 128 Gflops for 2 PEs, and then 64 Gflops for 1 PE, if i recall. something to that effect.

 

[Brimstone]

I think Deadmeat predicted 128 Gflops for 2 PEs, and then 64 Gflops for 1 PE, if i recall. something to that effect.

I thought it was 1 running at a clock speed of 2.0 Ghz or so.

 

[Inane_Dork]

that *might* open up an oppertunity (only an oppertunity) for Microsoft to nearly rival PS3 performance with Xenon.

Might? Opportunity? Rival?

Please. Don't patronize Sony so much. The two will be very similar in performance whatever Sony goes with in the PS3. Sure, if you look at theoretical specs, the PS3 should be the victor. But effectively, all your qualifications (you only need 1, BTW) are not needed.

like maybe Microsoft will decide to double the reported 3-core CPU configuration to 6. providing ~84 * 2 GFLOPs performance

I can't think of a good reason for MS to suddenly double the number of CPUs in the X2. That's so much more money per unit it would give the accounting wing a heart attack.

 

[Mythos]

worst case senario. Sony goes for a single PE CPU that provides 256 GFLOPs peak performance. the Nvidia GPU is just a modified NV5x with several times the performance of NV40 / GeForce 6800.

IMO, will hit 1Tflop with CPU/GPU configuration. The work is there from all members in STI on the silicon front to reduce heat output and leakage. [/quote]

 

[Dave Baumann]

Sorry for the OT from the thread but IMO people are possibly focusing on the MS "cost" issue too much - everything they have done so far screams conigurability to make cost savings over time. While I have no more knowledge about MS's plans than the next man I'm not entirely sure that this will be any lower performance than can reasonably be expected for its launch time and they won't go for something fairly high cost in order to scale it back later (the configuration of the CPU patent and reference to 200W+ consumption are hints in my mind).

 

[London Geezer]

Hey! You guys are on an ameeeeeer-i-can board, so start speekin' ameeeeerican!

It's british actually. More or less.

But eveyone knows Britain is just the 52nd state.

54th thankyouverymuch.
52nd is Israel.
53rd is Iraq.

 

[Jacob]

I think Deadmeat predicted 128 Gflops for 2 PEs, and then 64 Gflops for 1 PE, if i recall. something to that effect.

Actually, in the period of a few months he predicted that PS3's theoretical floating point performance would be (off the top of my head, and switching from one to the other from week to week):

256 Gflops

128 Gflops

70/80 something Gflops

64 Gflops

32 Gflops

4 Gflops

1 Gflop

0.5 Gflop

All bases covered.

Hype aside, all next gen systems should be pretty close when it comes to performance. Sony, Microsoft and Nintendo have invested a lot of money on these machines, and have been working with some of the most talented experts in the field. Thinking that one of the three systems will leave the other two in the dust tech-wise is nonsense, be it PS3, Xenon or *sigh* Revolution (I hate this name).

 

[Mythos]

Hype aside, all next gen systems should be pretty close when it comes to performance. Sony, Microsoft and Nintendo have invested a lot of money on these machines, and have been working with some of the most talented experts in the field. Thinking that one of the three systems will leave the other two in the dust tech-wise is nonsense, be it PS3, Xenon or *sigh* Revolution (I hate this name).

Hype aside..I'm sure all console makers have put in considerable work but the work put in by STI in Cell has a great probablity to surpass the CPU capablities in both Nintendo's and Microsofts consoles. Here's some concrete examples of what to look out for.....

The first-generation multicore SOC CELL processor, to be described by IBM, Sony, and Toshiba, combines eight streaming processors on a chip providing a high-performance platform for multimedia and streaming workloads. These processors are designed with features specifically targeted for certain applications, saving power and area by this narrower application focus. Implemented in a 90-nm SOI process, the chip incorporates extensive power- and thermal-management techniques...

Developers from Rambus and Stanford University will address the transfer of data on and off chip at the high intrinsic speed of advanced processors. They will present an I/O interface with a total data rate of more than 0.5 terabits/second (that is, 500 billion bits per second)! Each pin operates at a data rate of 6.4 Gb/s, consuming an energy of only 20 pJ for every transferred bit. To achieve an accurate timing relationship between the I/O pins, the presenters will describe the on-chip clock-distribution technique that delivers the reference clock across the width of the I/O interface with a clock-tracking architecture that allows the use of low-cost clock sources. These circuits can maintain high analog performance even in an SOI CMOS process.[/quote]

 

[nAo]

What are you quoting?

 

[Mythos]

http://www.ieee.org

 

[function]

Sorry for the OT from the thread but IMO people are possibly focusing on the MS "cost" issue too much - everything they have done so far screams conigurability to make cost savings over time. While I have no more knowledge about MS's plans than the next man I'm not entirely sure that this will be any lower performance than can reasonably be expected for its launch time and they won't go for something fairly high cost in order to scale it back later (the configuration of the CPU patent and reference to 200W+ consumption are hints in my mind).

Well put.

Anyway, just a general thought going round my head is this: if the PS3's CPU were to be several times more powerful than Xenon's CPU, where would this power come form?

Both the PS3 and Xenon CPU's are currently rumoured to be new designs, share the same process size, and run at similar clock speeds. If this is case, where is a 16, 8 or even 4 times increase in performance for the PS3 CPU (over Xenon) going to come from? Wouldn't it take a vastly superior CPU design, with a process advantage and a much large die size to make this the case?

Someone here - I think it was Vince - said that "process is everything" (please excuse me if I'm quoting out of context). In the absence of such of advantage, where would an automatic guarantee of vast superiority come from? Heck, even with it ...

In terms of suitability of design, I again don't see where the huge automatic advantage for Sony is. Xenon's CPU is a clean sheet design, specifically for a games console by the very experienced IBM and MS themselves. I don't know much about Cell, other than that its use in PS3 ([edit]: with a suitable configuration, of course [/edit]) is the first step in Sony recouping some of its tremendous investment in processor technology which they hope to use to challenge other processor lines. Like I said, I don't automatically see an advantage for the PS3 in this.

 

[Dave Baumann]

Xenon's CPU is a clean sheet design, specifically for a games console by the very experienced IBM and MS themselves. I don't know much about Cell, other than that its use in PS3 ([edit]: with a suitable configuration, of course [/edit]) is the first step in Sony recouping some of its tremendous investment in processor technology which they hope to use to challenge other processor lines. Like I said, I don't automatically see an advantage for the PS3 in this.

Highlighted elements are also pertinent to this case. If half the stuff that we read about here is actually the case for Cell then it is going to be used in a multitude of other devices and its design should logically need to cater for those other devices, which may or may not actually aid it in a game console application.

 

[MfA]

Function, a little thought experiment. Replace the VUs on the EE with 2 superscalar PowerPC cores with 4 way SIMD, same peak performance but do you think it could be produced at the same size and cost? I could see Xenon competing with Cell, only because I can see Microsoft spending all the money they saved on the harddisk on extra processor power

 

[nAo]

I could see Xenon competing with Cell, only because I can see Microsoft spending all the money they saved on the harddisk on extra processor power

Damn..I was about to write the same thing! Anyway in the end we should consider the full system (CPU+GPU) and there are too many unknown variables out there about PS3 GPU to even start to make some educated guess. If PS3 CPU does cost significantly more than Xenon CPU maybe Sony will save money on the GPU..

ciao,
Marco

 

[function]

Highlighted elements are also pertinent to this case. If half the stuff that we read about here is actually the case for Cell then it is going to be used in a multitude of other devices and its design should logically need to cater for those other devices, which may or may not actually aid it in a game console application.

Yeah. It's hard to see it benefiting it any more than making a custom chip for that purpose though, at best.

Function, a little thought experiment. Replace the VUs on the EE with 2 superscalar PowerPC cores with 4 way SIMD, same peak performance but do you think it could be produced at the same size and cost? I could see Xenon competing with Cell, only because I can see Microsoft spending all the money they saved on the harddisk on extra processor power

This is a pretty new area for me, so excuse me if I sound like the village idiot. I'm assuming that replacing the VU's with PowerPC cores would make the whole thing bigger, more expensive, and due to it's increased complexity possibly harder to make run at higher clock speeds...

So in terms of pushing the high GFlop numbers and transforming vertices (for example), you're better of putting more vector units on the chip rather than a smaller number of PowerPC cores (with 4 way SIMD). However, in terms of a central processing unit for a console, how do you balance general purpose processing power and vector processing power?

Almost all of of the talk around CPU power seems to be based around GFlops, with little to no consideration going to integer processing, thread handling and switching, size and effectiveness of caches or any of the other things I obviously don't really understand the impact of.