PS3's CPU, GPU, RAM and eDRAM configuration?

[j^aws]

Please choose the best CPU, GPU, RAM and eDRAM configuration you think would be optimal for the PS3 with CELL and NV5x tech from the options below,


A)

[CPU]<==>[GPU]---> Output
  |
[512MB]

B)

[CPU]<==>[GPU]---> Output
  |        |
[256MB]  [256MB]

C)

[CPU]<==>[GPU<>eDRAM]---> Output
  |
[512MB]

D)

[CPU]<==>[GPU<>eDRAM]---> Output
  |        |
[256MB]  [256MB]

Some basic assumption,

1. Expect die sizes to be constant for CPU and GPU for all cases. i.e. if you have eDRAM on the GPU, expect less shader cores.

2. I've chosen a total 512MB RAM here but you can chooses what you feel is expected but keep the same ratio between CPU and GPU as above. Also, if eDRAM is used, 32MB would be fine but you can assume what you think would be expected but keep it consistent.

So which option would be good for PS3, A, B, C or D

EDIT: As noted above, remember if you choose eDRAM, there will be LESS shader cores on the GPU...just making that point clearer!

 

[Junekwan]

E.

[CPU]<==>[GPU<>eDRAM]---> Output

└ ─[512MB]─ ┘

 

[London Geezer]

Well obviously D is the best, the more the better. But i don't think it will happen.

 

[Inane_Dork]

I voted for (C). The GPU won't need 256 MB of its own storage.

Actually, I rather doubt there will be 512 MB of RAM is any of the next gen machines. I still hope there will be, though.

 

[London Geezer]

i think there could be 128MB of VRAM and a separate pool of System RAM, probably 256MB. That would help things a lot.

Not sure about eDRAM, we've heard rumours but so far no word on it.

 

[Phil]

what about the configuration posted by Blachford's CELL Architecture explained?

2 dual PPE chips with each PPE having access to 64MB of RAM and one PE connected to the GPU with additional RAM (unknown size). I'm less interested in his proposition of 4 PPEs being feasable, but more about if 4 different sets of 64 MB memory pools for each PPEs is a good thing or not. Wouldn't 256MB of unified memory be better, which all PPEs could access?

Unless is betting that the bandwidth is so high between the Chips, that they could easily fetch data through the other chip?

 

[Gubbi]

what about the configuration posted by Blachford's CELL Architecture explained?

That guy is hilarious.

Cheers
Gubbi

 

[V3]

With regard to B and D, how good will the NV and Sony XDR implementation be for the this NVxx that's going to PS3 ?

 

[j^aws]

Well obviously D is the best, the more the better. But i don't think it will happen.

More what? Remember if you're having eDRAM, you're getting less Shader execution cores on the GPU...

 

[London Geezer]

what about the configuration posted by Blachford's CELL Architecture explained?

2 dual PPE chips with each PPE having access to 64MB of RAM and one PE connected to the GPU with additional RAM (unknown size). I'm less interested in his proposition of 4 PPEs being feasable, but more about if 4 different sets of 64 MB memory pools for each PPEs is a good thing or not. Wouldn't 256MB of unified memory be better, which all PPEs could access?

Unless is betting that the bandwidth is so high between the Chips, that they could easily fetch data through the other chip?

I would tend to think that one large pool of RAM for the CPU is the best thing. Separating RAM according to CPUs (unless all CPUs can access all pools of RAM at the same speed that is) doesn't sound so good.

So, RAM for CPU, maybe separate RAM for GPU, or one large pool for both, with the GPU being able to access RAM directly.

 

[j^aws]

what about the configuration posted by Blachford's CELL Architecture explained?

2 dual PPE chips with each PPE having access to 64MB of RAM and one PE connected to the GPU with additional RAM (unknown size). I'm less interested in his proposition of 4 PPEs being feasable, but more about if 4 different sets of 64 MB memory pools for each PPEs is a good thing or not. Wouldn't 256MB of unified memory be better, which all PPEs could access?

Unless is betting that the bandwidth is so high between the Chips, that they could easily fetch data through the other chip?

I you're thinking multiple CELLs, assume they are still equal to [CPU] above...i.e. [CPU] = multiple CELLs with memory hanging off it...

 

[Shinjisan]

I agreee with Junekwan.

E.

[CPU]<==>[GPU<>eDRAM]---> Output

└ ─[512MB]─ ┘


My dream configuration would be a 4 Cell CPU with 512MB of Ram (128MB per Cell) and 100GB/s bandwidth.
The GPU would have its embedded ram (64MB can be fitted in since the gpu won't have vertex shaders but will only do pixel processing) but also with the possibility to use the system memory if needed.
More realistically we can also have the same configuration with a 2 Cell CPU and 50GB/s bandwidth.

 

[London Geezer]

I agreee with Junekwan.

E.

[CPU]<==>[GPU<>eDRAM]---> Output

└ ─[512MB]─ ┘


My dream coniguration would be a 4 Cell CPU with 512MB of Ram (128MB per Cell) and 100GB/s bandwidth.
The GPU would have its embedded ram (64MB can be fitted in since the gpu won't have vertex shaders but will only do pixel processing) but also with the possibility to use the system memory if needed.
.

If anything, that should be a requirement!!! Textures and all the pre-processed graphics data would be in the 512MB pool of memory! EDRAM is not there to store textures and geometry

 

[j^aws]

I agreee with Junekwan.

E.

[CPU]<==>[GPU<>eDRAM]---> Output

? ?[512MB]? ?


My dream coniguration would be a 4 Cell CPU with 512MB of Ram (128MB per Cell) and 100GB/s bandwidth.
The GPU would have its embedded ram (64MB can be fitted in since the gpu won't have vertex shaders but will only do pixel processing) but also with the possibility to use the system memory if needed.
.

If anything, that should be a requirement!!! Textures and all the pre-processed graphics data would be in the 512MB pool of memory! EDRAM is not there to store textures and geometry

Is 'E." above even possible?

 

[London Geezer]

I agreee with Junekwan.

E.

[CPU]<==>[GPU<>eDRAM]---> Output

? ?[512MB]? ?


My dream coniguration would be a 4 Cell CPU with 512MB of Ram (128MB per Cell) and 100GB/s bandwidth.
The GPU would have its embedded ram (64MB can be fitted in since the gpu won't have vertex shaders but will only do pixel processing) but also with the possibility to use the system memory if needed.
.

If anything, that should be a requirement!!! Textures and all the pre-processed graphics data would be in the 512MB pool of memory! EDRAM is not there to store textures and geometry

Is 'E." above even possible?

Well, E would be the current vision of the PS3 among the rumourers, a big pool of RAM shared between CPU and GPU, both being able to access it directly without influencing each other, and some eDRAM on the GPU. Only there's 512MB of RAM instead of the rumoured 256MB.

 

[j^aws]

I agreee with Junekwan.

E.

[CPU]<==>[GPU<>eDRAM]---> Output

? ?[512MB]? ?


My dream coniguration would be a 4 Cell CPU with 512MB of Ram (128MB per Cell) and 100GB/s bandwidth.
The GPU would have its embedded ram (64MB can be fitted in since the gpu won't have vertex shaders but will only do pixel processing) but also with the possibility to use the system memory if needed.
.

If anything, that should be a requirement!!! Textures and all the pre-processed graphics data would be in the 512MB pool of memory! EDRAM is not there to store textures and geometry

Is 'E." above even possible?

Well, E would be the current vision of the PS3 among the rumourers, a big pool of RAM shared between CPU and GPU, both being able to access it directly without influencing each other, and some eDRAM on the GPU. Only there's 512MB of RAM instead of the rumoured 256MB.

What your describing is 'C.' if the GPU can DMA and have eDRAM...

 

[London Geezer]

What your describing is 'C.' if the GPU can DMA and have eDRAM...

No,
in your case, texture data has to go from the CPU to the GPU, whereas it should just go from the main memory to the GPU without having to bother the CPU.
Like the difference between the PS2 and the Xbox.

 

[j^aws]

What your describing is 'C.' if the GPU can DMA and have eDRAM...

No,
in your case, texture data has to go from the CPU to the GPU, whereas it should just go from the main memory to the GPU without having to bother the CPU.
Like the difference between the PS2 and the Xbox.

Then what your describing is 'D.' and not 'C.' then...this is what I don't get about 'E.'...it's like a hybrid 'C.' and 'D.' but I can't see how?

 

[London Geezer]

Then what your describing is 'D.' and not 'C.' then...this is what I don't get about 'E.'...it's like a hybrid 'C.' and 'D.' but I can't see how?

You said it was C!
It's not even D. It could be D if the developers decided to reserve 256MB of the main pool for the CPU and the rest for the GPU.
E is more flexible, u're not bound by D's 256MB limit for either processor. Want to use mroe than 256MB for graphics data? Just cut on the game data.

See the difference?

In the end, it's very unlikely we'll get 512MB so it will probably just be E but the large pool having 256MB instead of 512.

 

[version]

what you mean this?