Support NeoGAF

[gofreak]

And yet sneaky, only a 3 month difference between Xenon and PS3 launch? The longer the gap is (and we all know it will be long), the more foregone the conclusion that the PS3 has a significant power advantage.

I'd say it'll be 6-8 months, tbh. I can see an Apr-June 2006 launch timeframe in japan.

Of course that means we have to wait till Nov 2006...or longer in europe.. (which would give MS a 12 month headstart).

 

[Panajev2001a]

And yet sneaky, only a 3 month difference between Xenon and PS3 launch? The longer the gap is (and we all know it will be long), the more foregone the conclusion that the PS3 has a significant power advantage.

Azih.... how can you paint such a mean figure of me ?

3 Months was the minimum amount of time for me to start showing a small, but noticeable difference in peak performance (the approach Xbox mainly followed) or manufacturing costs (the approach that GameCube mainly followed).

Of course it will not be anythign too huge as that would require far more time.

Still, even if they came out at the same time you would have each console excelling in some areas and doing worse than the competition in others as the hardware is not identical or bought off-the-shelves from the same producers.

 

[Azih]

Just Kidding Pana. . So what you been up to anyway?

 

[Panajev2001a]

Just Kidding Pana. . So what you been up to anyway?

School-work and PlayStation 2 programming... to take over the world Pinky .

What about you ?

 

[gofreak]

http://www.theregister.com/2004/11/29/ibm_sony_cell_debut/

The Register reports that the current Cell workstations pack 2Tflops of power. Not sure where they got this from?

Also, a guy on Slashdot claiming to work for a developer in Melbourne, Australia, says that Sony and IBM are organising a meeting soon (in Oz) to show off more of cell to developers there. He doesn't specify when, and I'm not sure how likely that all sounds..

 

[teiresias]

Well, obviously if Tecmo isn't invited than the meeting will have never actually happened.

 

[McFly]

http://www.theregister.com/2004/11/29/ibm_sony_cell_debut/

The Register reports that the current Cell workstations pack 2Tflops of power. Not sure where they got this from?

Damn, I could so use one of them.

Fredi

 

[WordofGod]

http://news.bbc.co.uk/2/hi/technology/4051983.stm

First look at PlayStation 3 chip

The successor to the PlayStation 2 will be launched in 2006
Some details of the chip inside Sony's PlayStation 3 have been revealed.

 

[Azih]

School-work and PlayStation 2 programming... to take over the world Pinky .

What about you ?

Graduated, got a job, dull stuff. Hey if you ever take over the world, offer me a job

 

[gofreak]

http://www.gamespot.com/news/2004/11/29/news_6114033.html

Gamespot have chimed in. They say PS3 itself is due to be unveiled in Japan in March.

That'd put it on a nice PS2-style schedule. (PS2 was unveiled 2 March 1999 - 2 March 2005 would be a nice unveiling date for PS3 )

 

[deadhorse32]

http://www.gamespot.com/news/2004/11/29/news_6114033.html

Gamespot have chimed in. They say PS3 itself is due to be unveiled in Japan in March.

That'd put it on a nice PS2-style schedule. (PS2 was unveiled 2 March 1999 - 2 March 2005 would be a nice unveiling date for PS3 )

Schedule is very PS2-like.

February 2005 : First technical info at ISSCC
March 2005 : First SCEJ detail with spec and early demo
May 2005 : More demos and games announcement.
March 2006 : release in Japan

 

[gofreak]

Schedule is very PS2-like.

February 2005 : First technical info at ISSCC
March 2005 : First SCEJ detail with spec and early demo
May 2005 : More demos and games announcement.
March 2006 : release in Japan

Hehe, indeed. I've a funny feeling, also, that these developers who have cell-based workstations now are actually more working on demos for March than games themselves :lol I can't wait!

 

[Midas]

Hehe, indeed. I've a funny feeling, also, that these developers who have cell-based workstations now are actually more working on demos for March than games themselves :lol I can't wait!

A demo featuring cars, made by PD would be interesting.

 

[Smiles and Cries]

time to finally buy a PlayStation does the PS3 play PS2 games?

 

[gofreak]

time to finally buy a PlayStation does the PS3 play PS2 games?

Not confirmed. Seems likely though.

A demo featuring cars, made by PD would be interesting.

I'd say that's a given. Who else? Maybe a Naughty Dog demo?

A square demo? It'd be cool if they took some current-gen FMV from one of their games and reproduced it in realtime (even if not perfectly), ala the ballroom dancing scene from FFVIII in the original PS2 announcement.

 

[Defensor]

I expect a tech demo from Namco, Konami, and Square-Enix in March.

 

[Midas]

Yeah probably those three. Sony will look pretty bad in the eyes of averge joe with just a couple of tech demos and specs though. Isn't MS expected to unviel Xbox 2 with games, final hardware (design) and everything else at CES in January?

 

[gofreak]

Yeah probably those three. Sony will look pretty bad in the eyes of averge joe with just a couple of tech demos and specs though. Isn't MS expected to unviel Xbox 2 with games, final hardware (design) and everything else at CES in January?

No, all we know is Bill Gates is keynoting. An Xbox2 announcement seems very likely, but what form that will take - who knows?

By that logic, in the eye's of the average joe, PS2 looked bad when DC was announced/launched...it's tricky to second-guess what the mass market is thinking!

 

[nitewulf]

so, each PE unit is like so,

PE
|
CPU + DMA controller + 8 SP
|_______|___________| <--- BUS (speed?)

and there are multiple inside one cell.
each SP contains 8 APUs (4 FP + 4 Int).
but there may or may not be 8 SPs per PE.
and an arbitrary number of PEs per cell.
ok, good.

 

[xexex]

sounds like:

*first experimental prototype cell-based workstation reaches 2 tflops

*future cell-based workstation will reach 16 tflops

*PS3 a fraction of a tflop


could be wrong.....

 

[gofreak]

sounds like:

*first experimental prototype cell-based workstation reaches 2 tflops

*future cell-based workstation will reach 16 tflops

*PS3 a fraction of a tflop


could be wrong.....

Prob not too far wrong.

PS3 seems likely to range anyware from ~0.25TFlop-1TFlop.

 

[Pimpwerx]

Yeah probably those three. Sony will look pretty bad in the eyes of averge joe with just a couple of tech demos and specs though. Isn't MS expected to unviel Xbox 2 with games, final hardware (design) and everything else at CES in January?

Well, they made sure to mention next-generation computing applications, knowing full well everyone's gonna know it's for the PS3. I think they're trying to pre-empt what's sure to be a big Xbox2 unveiling at CES. But I really don't think there's anything they can do to affect that though.



I look forward to the official unveiling of Cell in Feb. We'll find out a lot about the system then. I want to know about eDRAM. I thought I heard someone say 32-64MB back in the day? Am I wrong here? PEACE.

 

[Panajev2001a]

so, each PE unit is like so,

PE
|
CPU + DMA controller + 8 SP
|_______|___________| <--- BUS (speed?)

and there are multiple inside one cell.
each SP contains 8 APUs (4 FP + 4 Int).
but there may or may not be 8 SPs per PE.
and an arbitrary number of PEs per cell.
ok, good.

No, a single PE contains a PU with, according to latest patents, seems to have 2 levels of cache.

A PE also contains a variable number of APUs or SP units or whatever you call them these days: each APU contains 4 FPUs and 4 FXUs (likely the logic for them is shared and you won't be able to execute from all of them at the same time).

The APUs have 128x128 bits registers and ~128 KB of Local Storage and, again from latest patents, share a L1 cache together, but e-DRAM might have been taken off from the CPU as the cache would help with performance without enlarging the chip-size as much (expect the GPU to have quite a bit of e-DRAM IMHO). A shared L1 cache would help APUs' performance avoiding long latency main RAM accesses.

A PE also contains one DMA Controller with a TLB of some sort on chip allowing the DMAC to understand and work with Virtual Addresses instead of being limited to Physical Addresses like for example the PlayStation 2's DMAC in the EE is limited to.

The DMAC allows each APU to access external RAM through the DMAC without deferring the work to the PU (APUs can feed themselves and not rely exclusively on the PU to feed them like it is the case for the VUs in the EE).

If you read the patents I posted at the end of this post, which nAo found out about last week, we have a DMA engine per APU: an APU complex is an APU + a DMA engine or MFC (Memory Flow Controller).

http://appft1.uspto.gov/netacgi/nph...N/"International+Business+Machines"+AND+Kahle

http://appft1.uspto.gov/netacgi/nph...N/"International+Business+Machines"+AND+Kahle

 

[Amir0x]

I am fascinated with your technological knowledge, Panajev2001a.

Color me impressed.

 

[duderon]

You ain't seen nothin dood. Just wait till he makes a thread on CELL/PS3 that only a select few on GAF can comprehend. It will happen.

 

[Vortac]

No cell based dev kits have been released to any developers, including Square. Just theoretical work on PCs. Probably will stay that way until next summer.

Sony is calling Microsoft's bluff in a big way...with Western publishers PS3 is wait and see at this point, while Xenon full steam ahead.

Therefore, where Sony only has demos, MS will have an impressive array of launch games, since they already have launch dates and everything set in stone.

We'll see, but if Sony gets their technical support and documentation coordinated and up to speed with MS it will really be close next gen...as long as Western gamers are still interested in Japanese-player oriented games (interest is obviously decreasing yearly compared to PSX golden era).

Nice thing they will be far enough apart (at least a year) for most early adopter types to get both at launch.

 

[aaaaa0]

I'd just like to point out that as far as we know, the 16 tflops Workstation rack is NOT the PS3 dev kit. That unit is intended for rendering movies for Hollywood.

The actual PS3 dev kits will be different (and slower).

Frankly, I'm doubtful the 2 tflops Workstations being presented will really be what PS3 dev kits end up looking like in terms of performance.

 

[VNZ]

Frankly, I'm doubtful the 2 tflops Workstations being presented will really be what PS3 dev kits end up looking like in terms of performance.

It will be quite interesting to see how the PS3 TOOLs will be configured. Will SCE build them akin to the PS2 ones with identical computing power (plus a big amount of extra RAM) that hooks up to a more conventional computer (be it in the same box) -- or will it be 100% Cell-based, with the cores configured to facilitate both fully compatible PS3 functionality and some extra development workstation functionality on the same "Cell-array". That'd be brave!

 

[gofreak]

I'd be wary of these 2Tflop workstations. The Inquirer is reporting that, but I've no idea where they're getting that from..

Also, Vortac: hasn't Sony said in these press releases that cell-based workstations are in the hands of some game developers?

 

[Dynamite Shikoku]

sooo.... will the graphics be good?

 

[fart]

rumors rumors rumors. wait until something's published guys.

 

[Suerte]

rumors rumors rumors. wait until something's published guys.

You do remember where you're posting, right?

 

[Fafalada]

Frankly, I'm doubtful the 2 tflops Workstations being presented will really be what PS3 dev kits end up looking like in terms of performance.

Workstations, regardless of their power relative to PS3 being higher or lower, I'd expect to be basically Phase-1 devkits. Mainly there to give people some idea what the new architecture will be like (in a way similar to current Xenon devkits).

 

[McFly]

Thx to Vince @ B3D (added some small modifications)

1st generation Cell Processor, 90nm SOI.
4.6GHz global PE clock speed
1.3V
85deg C w/ heatsink
6.4Gbit/pin external communication
6.4GHz/link RaZer internal bus[/list]
64-bit Power-based PU
SMT, CMP
DMAC intensive, software driven RT arbitration of resources
S|APUs
upto 8/PU
4 FPUs
4 FXUs
128*128bit Registers
128KB of LS (SRAM) @ 4.8GHz
1.2->2.4TByte/sec bandwith*
Second generation, 65nm Cell Processor has already sampled
2TFlop/sec DevKits have been sent out to a few key developers
16TFlop/sec Racks for content creation will follow next year

The above was presented at ISCC and based off the 90nm silicon that's in existence. What is likely extrapolated from this:

2nd generation Cell Processor, 65nm SOI.
>4GHz clock
1-4 PE's per IC
8 S|APUs per PE
256GFlop -> 1.3TFlop/sec FP
1.2TByte/sec -> ~10TByte/sec bandwith to LS*
Fill in what I missed.

*Upper number dependent on single cycle read/write like in EE VU's

Fredi

 

[gofreak]

Asides from the Inquirer, where else did the 2TFlop workstation/dev-kit news come from?

edit - also, how accurate is the information in Wikipedia's Playstation3 entry?
http://en.wikipedia.org/wiki/PlayStation_3

They say Cadence Design has revealed involvement in a graphics synthesizer for PS3, but the process they're talking about (0.18) seems way off. If it's true, however, I guess it confirms that there won't just be a cpu in PS3..

 

[Panajev2001a]

No cell based dev kits have been released to any developers, including Square. Just theoretical work on PCs. Probably will stay that way until next summer.

I'll answer this with a .

The CELL workstation is coming full speed and soon: I won't speak of any time-tables, I'll just say "soon".



It is natural that Microsoft is at a more advanced stage in terms of software being developed for Xbox 2/Xenon: they are the one that decided to jump the gun, end their generation early and launch the successor to the Xbox sooner.

PlayStation 3 will be at E3 and will launch after Xbox 2/Xenon hence it would be horrible if Microsoft did not have software closer to being complete at E3 compared to Sony and Nintendo.

 

[Panajev2001a]

It will be quite interesting to see how the PS3 TOOLs will be configured. Will SCE build them akin to the PS2 ones with identical computing power (plus a big amount of extra RAM) that hooks up to a more conventional computer (be it in the same box) -- or will it be 100% Cell-based, with the cores configured to facilitate both fully compatible PS3 functionality and some extra development workstation functionality on the same "Cell-array". That'd be brave!

with the ability of CELL to "virtualize" (taking a term from POWER5) or to run multiple operating systems at the same time and packing IMHO more than enough power to run a decent OS and compiler tools, I'd expect the TOOL for PlayStation 3 o be a reduced version of the full CELL WorkStation: after-all, Sony said that the CELL WorkStation would be used for both game content creation as well as movie and other digital content creation tasks.

Cool things could be done to help developers: one of the first ideas that comes naturally to mind is distributed processing in terms of tasks such as compiling.

It is likely that all the CELL WorkStations that a game developer has would be connected on a fast LAN able to provide high bandwidth and low latency and there are some tasks such as code compilation that would receive quite a nice speed-up thanks to distributed processing across the CELL WorkStations (code compilation is not such a quick task, especially when the project grows big).

 

[gofreak]

Well, I tried to confirm the 2TFlop figure for current workstations with IBM, but they simply said they couldn't discuss any further details until the ISSCC.

Feb never seemed further away...Boohoo

 

[nitewulf]

No, a single PE contains a PU with, according to latest patents, seems to have 2 levels of cache.
...

I see.
So the streaming processors are the APUs. The wording confused me and I thought each SP would include upto 8 APUs, resulting in 64 APUs max per PE.
DMA engine per APU is fascinating, I thought they shared one engine. It'll increase hardware costs however. And the pre-fetch algorithm is genius.

 

[Phoenix]

So little information has actually been given out that people need to keep their pants on

 

[Phoenix]

It is likely that all the CELL WorkStations that a game developer has would be connected on a fast LAN able to provide high bandwidth and low latency and there are some tasks such as code compilation that would receive quite a nice speed-up thanks to distributed processing across the CELL WorkStations (code compilation is not such a quick task, especially when the project grows big).

Except that we've had distributed build environments for almost 2 years now. Even that piece of crap that dares to call itself a development environment (Visual Studio) has Incredibuild

 

[nitewulf]

So little information has actually been given out that people need to keep their pants on

exactly, w/o knowing how many PEs will be on the PS3 cell(s), and how many SPs each will contain, not much can be ascertained about the PS3's power from this recent information.

 

[Panajev2001a]

Except that we've had distributed build environments for almost 2 years now. Even that piece of crap that dares to call itself a development environment (Visual Studio) has Incredibuild

Yes, you had it also on GCC with distcc... but I am thinking about a top-bottom solution, fully optimized, fully tested and debugged thought for distributed processing to quicken game development.

 

[Kai Dracon]

It seems obvious to me that Tecmo is only biding their time and appearing disinterested. Soon their master ninjas (selected, of course, from the Master Ninja Tournament) will strike and steal the CELL design documents for Microsoft.

 

[mrklaw]

I wonder if the layout is flexible. i.e Could sony adjust the numbers of units/cell and cells per PS3 late in the day.

If so, I hope MS's announcment of 3 dual core processors make Sony up the ante even more.

 

[gofreak]

A question regarding manufacturing processes:

It seems they're a little behind their original schedule re. Cell manufacturing? I'm reading some older stuff regarding the project, and apparently Kutaragi said back in May that they had started sampling at 65nm? Since Cell has now emerged on 90nm, can we assume there have been problems at 65nm?

PS3 is looking likely to start out on 90nm now? If so, what does this mean for the amount of power they can pack in there?

 

[McFly]

PS3 is looking likely to start out on 90nm now? If so, what does this mean for the amount of power they can pack in there?

From all I've read, no, only the workstations will start with 90nm cell chips. While it's most likely true that they have 65nm samples for some time, there's a big diference between samples and market ready chips, there's a lot of optimization needed before a sampled chip goes into finalized production. If they would have 65nm market ready chips, then they would be at least one year earlier ready (if not two) then anyone else in the industry.

Fredi

 

[Panajev2001a]

A question regarding manufacturing processes:

It seems they're a little behind their original schedule re. Cell manufacturing? I'm reading some older stuff regarding the project, and apparently Kutaragi said back in May that they had started sampling at 65nm? Since Cell has now emerged on 90nm, can we assume there have been problems at 65nm?

Sampling in65 nm does not require finished or near finished manufacturing lines apparently: which is not yet the situation they have in all their fabs especially Nagasaki #2.

Toshiba is probably ahead in terms of 65 nm manufacturing in the Toshiba-SCE co-owned Oita #2 fab.

Yes, I have heard rumors of issues with their 65 nm process, but at this point in time it is still normal.

PS3 is looking likely to start out on 90nm now? If so, what does this mean for the amount of power they can pack in there?

As a 2006 product PlayStation 3 should not have major problem launching with 65 nm technology which should be online even in Nagasaki #2 by mid to late 2005 (and this plant is only one of three plants [one at an IBM facility that has been financed by SCE and then we have Oita #2 co-owned with Toshiba]).

 

[gofreak]

From all I've read, no, only the workstations will start with 90nm cell chips. While it's most likely true that they have 65nm samples for some time, there's a big diference between samples and market ready chips, there's a lot of optimization needed before a sampled chip goes into finalized production. If they would have 65nm market ready chips, then they would be at least one year earlier ready (if not two) then anyone else in the industry.

Fredi

They're talking about starting "sample production" of 90nm cell chips early next year, though, right? It doesn't sound to me like that'll be mass commercial production. If they had 65nm samples, why wouldn't they use them instead in these ISSCC papers?

 

[Panajev2001a]

If they had 65nm samples, why wouldn't they use them instead in these ISSCC papers?

Because they want a fully working WorkStation proto-type that they can use in games (and other software) development very SOON .