PS3 = 512Mbit XDR-DRAM (and Nintendo to adopt XDR-DRAM?)

[London Geezer]

What would u choose, 512MB (instead of 256) or 4PEs (instead of 2)? (assuming they're mutually exclusive)

In english:

512MB and 2PEs (16 SPEs)
or
256MB and 4PEs (32 SPEs)
???

That probably should read:

512MB and 1PEs (8 SPEs)
or
256MB and 2PEs (16 SPEs)



Tommy McClain

Just play the game bi*ch!

 

[AzBat]

Just play the game bi*ch!

ROFL

In the immortal words of Homie the Clown:

"I don't think so... Homie don't play that!" j/k

Personally at the performance levels the cell is already at I'd rather have more memory. Same for Xbox 2.

Tommy McClain

 

[MrFloopy]

Personally at the performance levels the cell is already at I'd rather have more memory. Same for Xbox 2.

May not be possible:

http://www.realworldtech.com/page.cfm?ArticleID=RWT021005084318

The obvious advantage of the XDR memory system is the bandwidth that it
provides to the CELL processor. The less obvious drawback is that it is
a memory system designed specifically for a high performance embedded
system. Particularly, the maximum of 4 DRAM devices means that the CELL
processor is limited to 256 MB of memory, given that the highest
capacity XDR DRAM device is currently 512 Mbits.

Are 1Gb XDR devices on horizon? Timeframe?

 

[YeuEmMaiMai]

512MB for a console is a lot of ram especially if the GPU has a dedicated pool of 64MB to 128MB ram. Don't forget that a lot of texture compression is available with S3TC...

 

[passerby]

In summary:

XDR interface: 25Gb/s connected to 256 MB
Flexi IO: 45Gb/s outbound, 32 Gb/s inbound

We can make a safe assumption that the FIO is responsible for connection to the GPU. Interesting part is that both in/out bandwidth is actually higher than the XDR's. I'm sure this lends evidence to the proposed programming model of data easily flowing to and fro between CPU and GPU. (So hardware-clipping is almost a certainty )

I'm curious about the possibility of the GPU part actually having a pool of memory where certain graphics primitives gets loaded in instead of in main memory, and the CPU request for these from the GPU instead of from main memory.