Official PS3 Thread

Status
Not open for further replies.
The optimist would think Sony will deliever MORE than 1000x PS2 performance

The utter fool would take the "1000x" at face value. The wise man would consider 10x to be pretty damn good, considering Murphy's law^Wrule-of-thumb in the timeframe.
 
I wouldnt know about that, I always thought the EE was a rather poor design in its initial iteration ... Im expecting ~30 myself.
 
AS I SAID, i expect much more.
just think about it. how much faster will PS3 be in long shaders programs than PS2? oh wait PS2 cant even do per pixel shading (i know it's a tad more complicated than that but just hold on for a sec)...

or, how much faster will ps3 be in displacement mapping?

see, all these things will add up to make a general figure of how much faster is PS3 compared to PS2. and i'm sure *ANYTHING* even out now, let alone 2006, is much faster than 10x PS2 in things such as the above mentioned features.

the Xbox is already many times faster than ps2 when it comes to pixel shading, if only because it supports it in hardware...

or take AA. let's say ps3 does 8x FSAA by default (EXAMPLE HERE!!). how much faster is that? i guess much more than 10x ps2's....

the 1000x figure is a *crowd wow-er*. but i'm sure that feature-by-feature ps3 will be much much more than 10x ps2....
 
EE = first design by the SCE + Toshiba group

SCE was not as experienced as they are now in high performance MPUs ( CPUs and GPUs ) design and manufacturing as they are now and they did not have as advanced manufacturing processes as the ones they will probably use for PlayStation 3's manufacturing.

1,000x != 1,000 * RAW specs maximum...

Going by the Theorethical maximum of 1 TFLOPS for PlayStation 3's Broadband Engine , I'd say ~161.29x of increase ( 1 TFLOPS might also include the FLOPS produced by the APUs in the Visualizer though... so the performance increase can go lower than 161.29x as far as Emotion Engine vs Broadband Engine is concerned ).

The comparison doesn't end here... if you add the resolution the PlayStation 3's GPU will be capable of rendering at while keeping a smooth 60 fps frame-rate ( we are talking about HDTV resolutions and/or nice AA ), the average per-frame texturing capabilities ( MB of textures per frame if we have to put it in a simple form ), texture filtering, complex vertex and pixel programs ( effects after effect... ), more complex AI ( Integer Heavvy code ), more advanced Physics, etc... you would end up with a set-up that would take a machine much closer to the 1,000x PlayStation 2 figure ( than what people expect ) to run the same set-up with the same smoothness and Image Quality.

The Broadband Engine has very high bandwidth e-DRAM and fast SRAM based Local Storages and plenty of registers to keep both Integer and FP Units nicely fed ( as much as possible ) while the Emotion Engine, as far as Integer processing is concerned, is quite limited by small Data Cache and Scratch Pad SRAM and the CPU to Main Memory latency.

PlayStation 3 has much bigger external RAM than PlayStation 2 ( I think they will have quite a bit more than 32 MB... on this strenght alone I might force a PlayStation 2 to throw its performance in the gutters by playing on the increased external RAM PlayStation 3 would have and have the PlayStation 2 keep going to the DVD drive or the HDD to retrieve data... with this in mind it should not be impossible to build games that would drive PlayStation 2 [limited to use the same effects, Image Quality and lighting and polygon count as the PlayStation 3 version] that run at 0.06 fps on PlayStation 2 and at 60 fps on PlayStation 3... even better would be showing a 500x jump.. from 0.12 fps to 60 fps ).

PlayStation 3 should have a more versatile Rasterizer and a faster optical medium as well as more of faster external RAM ( external == not embedded )...

What I have been trying to say ( a bit messily ) is that the trick which we can use to get closer to the 1,000x performance jump figure is by observing how much more efficient ( an EE overclocked to 1,000x its clock speed [rest of the subsystem stays the same] would perform nowhere near 1,000x faster ), counting also the effects and features PlayStation 3 without worries ( and in the case of the Rasterizer how much more flexible it is... increased AA, per-frame texturing, good pixel programs, etc... would not be particularly fast on PlayStation 2's Graphics Synthesizer ), should the PlayStation 3 be compared to PlayStation 2 and add that to the jump in theoretical performance...
 
Gunhead said:
The optimist would think Sony will deliever MORE than 1000x PS2 performance

The utter fool would take the "1000x" at face value. The wise man would consider 10x to be pretty damn good, considering Murphy's law^Wrule-of-thumb in the timeframe.

10x more Vertices/frame with the added AA, more complex Vertex Shaders Pixel Shaders ( and if we used a REYES-like renderer, how slow would PlayStation 2 be ? One key to show the increase in performance will also be the handling of the extra Integer power and the fact that performance has indeed increased but we also have more features like e-DRAM that help use stay closer to the theoretical maximums [the Emotion Engine would be slower on a clock by clock basis] ), physics, A.I. all at 60 fps ---- this looks something that would take a machine more than 10x faster than PlayStation 2 to process...
 
is there a slight possiblity that if the get the right number of fpu's,apus's and clockspeed the ps3 could do 5120 tflops per second that would be amazing if it can
 
qwerty if there were no limits on heat, die size, cost, etc, there could be a 16 TFLOPs CPU. or more. just combine enough PEs/APUs/FPUs with high clockspeed. but unfortunately that won't happen in a console.

however a workstation, ala GSCube, they will certainly surpass 16 TFLOPs.
 
10x more Vertices/frame with the added AA, more complex Vertex Shaders Pixel Shaders ( and if we used a REYES-like renderer, how slow would PlayStation 2 be ? One key to show the increase in performance will also be the handling of the extra Integer power and the fact that performance has indeed increased but we also have more features like e-DRAM that help use stay closer to the theoretical maximums [the Emotion Engine would be slower on a clock by clock basis] ), physics, A.I. all at 60 fps ---- this looks something that would take a machine more than 10x faster than PlayStation 2 to process...


yeah. I was thinking that if PS3 only achieved 6.2 GFLOPs but kept its features and efficiency, it would walk all over PS2. maybe achieving 50% of its peak performance while PS2 only achieved 1/6th. plus all those rendering features PS3 has that PS2 does not.
 
Going by the Theorethical maximum of 1 TFLOPS for PlayStation 3's Broadband Engine , I'd say ~161.29x of increase ( 1 TFLOPS might also include the FLOPS produced by the APUs in the Visualizer though... so the performance increase can go lower than 161.29x as far as Emotion Engine vs Broadband Engine is concerned ).

The comparison doesn't end here... if you add the resolution the PlayStation 3's GPU will be capable of rendering at while keeping a smooth 60 fps frame-rate ( we are talking about HDTV resolutions and/or nice AA ), the average per-frame texturing capabilities ( MB of textures per frame if we have to put it in a simple form ), texture filtering, complex vertex and pixel programs ( effects after effect... ), more complex AI ( Integer Heavvy code ), more advanced Physics, etc... you would end up with a set-up that would take a machine much closer to the 1,000x PlayStation 2 figure ( than what people expect ) to run the same set-up with the same smoothness and Image Quality.

The Broadband Engine has very high bandwidth e-DRAM and fast SRAM based Local Storages and plenty of registers to keep both Integer and FP Units nicely fed ( as much as possible ) while the Emotion Engine, as far as Integer processing is concerned, is quite limited by small Data Cache and Scratch Pad SRAM and the CPU to Main Memory latency.

PlayStation 3 has much bigger external RAM than PlayStation 2 ( I think they will have quite a bit more than 32 MB... on this strenght alone I might force a PlayStation 2 to throw its performance in the gutters by playing on the increased external RAM PlayStation 3 would have and have the PlayStation 2 keep going to the DVD drive or the HDD to retrieve data... with this in mind it should not be impossible to build games that would drive PlayStation 2 [limited to use the same effects, Image Quality and lighting and polygon count as the PlayStation 3 version] that run at 0.06 fps on PlayStation 2 and at 60 fps on PlayStation 3... even better would be showing a 500x jump.. from 0.12 fps to 60 fps ).

PlayStation 3 should have a more versatile Rasterizer and a faster optical medium as well as more of faster external RAM ( external == not embedded )...

What I have been trying to say ( a bit messily ) is that the trick which we can use to get closer to the 1,000x performance jump figure is by observing how much more efficient ( an EE overclocked to 1,000x its clock speed [rest of the subsystem stays the same] would perform nowhere near 1,000x faster ), counting also the effects and features PlayStation 3 without worries ( and in the case of the Rasterizer how much more flexible it is... increased AA, per-frame texturing, good pixel programs, etc... would not be particularly fast on PlayStation 2's Graphics Synthesizer ), should the PlayStation 3 be compared to PlayStation 2 and add that to the jump in theoretical performance...


very nice post, Panajev. you guys should read that at least twice as I did. it makes perfect sense. couldn't have said it that well myself. not by a longshot. the above is the ONLY way you can reasonably conclude that PS3 could be anywhere near 1000x the performance of PS2. since PS3 will almost certainly not reach 6.2 TFLOPS - perhaps 1~2 TFLOPs counting the FLOPs of the CPU and GPU.

As said many times already, it is the rendering features, pixel shaders, large CPU & GPU caches & eDRAMs (compared to PS2) better bus architechure, large external memory (512 MB perhaps) bandwidth, hopefully lower memory latency as well as faster media (small HDD would be even faster!) that will allow PS3 games to achieve the kind of megaleap Sony is promising. the same PS3 games on PS2 with the same geometry, lighting, features and quality would run at many times less than 1fps. the fact that PS3 games will be rendered/displayed at HDTV resolutions 1024i/1024p could alone be concidered a 4-5x leap from 640x480 or 640x240. add 8x FSAA and you are another "8x" there :)

Panajev wrote it better than i did.... but you should see now :)
 
There was a SCEI press conference today discussing PSP PSX and Cell.

Speaking of the PS3, SCEI's presentation also included information on its new semiconductor fabrication plants. One of the fabs is dedicated to manufacturing PS2, PSP, PSX, and CELL chips. CELL is a multi-arhitecture chip that's purported to be 1,000 more powerful than the PS2's processor. It will be used in the upcoming PS3 as well as other Sony electronics. SCEI claims that it's investing 200 billion Yen (around 1.67 billion USD) on the CELL portion of the plant. The total investment in the new fabs is 500 billion Yen (roughly 4.19 billion USD).

It's official, although not only from this press release that Cell is CONFIRMED to be in PS3. This will end any more of this BS.

http://gamespy.com/articles/june03/sony/

Oh and pictures :)

24.jpg


26.jpg


25.jpg
 
Yup, Paul. there should be no doubt that Cell is for PS3. even if it's some derivative of Cell or not even called Cell anymore (i.e. EE3/BBE)




Another thing I wanted to add about PS3 - if PS2 could only achieve say, 1/10th of its theoretical / peak performance due to horrible ineffiecencies, the PS3 could well get at least 25% of it's theoretical / peak performance. maybe as high as 30-50%, I don't know. perhaps 1T-SRAM will be used as the eDRAM or some other low-latency memory. (i know Rambus Yellowstone is the main external mem) because GAMECUBE gets much, much closer to its peak performance than PS2, thanks to low- latency memory as well as thoughtful bus layout.
 
megadrive0088 said:
Yup, Paul. there should be no doubt that Cell is for PS3. even if it's some derivative of Cell or not even called Cell anymore (i.e. EE3/BBE)




Another thing I wanted to add about PS3 - if PS2 could only achieve say, 1/10th of its theoretical / peak performance due to horrible ineffiecencies, the PS3 could well get at least 25% of it's theoretical / peak performance. maybe as high as 30-50%, I don't know. perhaps 1T-SRAM will be used as the eDRAM or some other low-latency memory. (i know Rambus Yellowstone is the main external mem) because GAMECUBE gets much, much closer to its peak performance than PS2, thanks to low- latency memory as well as thoughtful bus layout.

1-T SRAM has a much lower latency than Direct RDRAM yes, but the problem with integer performance for PlayStation 2 was the lower clock speed ( compared to Gekko ), the latency of the main RAM ( important yes ) and the lack of a big L2 cache ( that helps Gekko a lot )...

The Graphics Synthesizer's e-DRAM cannot be classified as being much less efficient than the e-DRAM on Flipper IMHO... the Graphics Synthesizer's e-DRAM architecture was well designed.

Yellowstone != Direct RDRAM

The architecture is quite different and IMHO moure promising than Direct RDRAM...

Busses are bi-directional ( all data busses are ), addresses are not multiplexed with data ( higher effective data transfer bandwidth as we have separate busses for addresses and data ) and it uses a faster signaling rate ( 3.2 GHz vs 800 MHz )...
 
Panajev,

couldn't Sony go for the 6.4 Ghz version of Yellowstone?
seems they will need all the bandwidth they can get.


Also, 48 GB/sec bandwidth for PS3's main memory would
be really nice and Yellowstone is said to be capable of
50-100 GB/sec bandwidth. I know I read that somewhere.

I know you've said PS3 should get 25+ GB/sec but that would
be the absolute minimum would it not?

2 channels get around 25 GB, right, while 4 channels would get
48-50 GB, and 8 channels get 96-100 GB/sec I remember that
being mentioned here on Beyond3D before... Say PS3 got 4
channels....that would be pretty decent.

512 MB memory w/ 50 GB sec bw is nice

256 MB memory w/ 25 GB sec bw is like the bare bones minimum
 
I think Pana should seriously work with IBM on Cell..

Oh and Pana, let's say for PS4. Would it be possible to just increase the APU floating point from 32GFLOPS to like 512 for each APU? And put more eDRAM on the whole thing, increase bus sizes and the likes.

Could it work like that?
 
25.6 GB/s uses 3.2 GHz signaling and a 64 bits Yellowstone memory controller...

Considering the e-DRAM, 25.6 GB/s is pretty fast... 8x faster than PlayStation 2's Direct RDRAM ( 3.2 GB/s ) and with lower latency... also, we know that off-chip RAM doesn't grow in speed ultra-fast... think the years that passed from PC100 to PC1066...
 
Oh and Pana, let's say for PS4. Would it be possible to just increase the APU floating point from 32GFLOPS to like 512 for each APU? And put more eDRAM on the whole thing, increase bus sizes and the likes.


yeah :)


Cell for PS3 is like lots and lots of VUs (the APUs) but with added interger power as well as FP.

PS4 should be lots and lots of Cell2s with APUs that do 512 GFLOPs or 1 TFLOPs each :)
 
Yea.. I imagine Sony will use the Cell architecture from here on out for future PS's. Just increase the APU gflops numbers, make the chip on a smaller micron process, add more edram and away you go. Of course there is more to it than that, but that's basicially the gist of it.
 
Yea.. I imagine Sony will use the Cell architecture from here on out for future PS's. Just increase the APU gflops numbers, make the chip on a smaller micron process, add more edram and away you go. Of course there is more to it than that, but that's basicially the gist of it.


yeah. at least for PS4. I doubt Sony will come up with a completely NEW concept for PS4. maybe for PS5 there will be totally new technology.

much like PS2 is based on PS1 technology in many ways.
 
I think Pana should seriously work with IBM on Cell..

I think Sony, IBM and Toshiba already have amongst the best engineers in the field today working on it, people that have earned the stripes for such a project ( while I haven't ) :) ( people with lots of years of practical experience with mciro-processor design )
 
Status
Not open for further replies.
Back
Top