forums.ps2dev.org

[JesusXP]

http://games.slashdot.org/games/06/06/05/0933239.shtml

Breakdown:

[Arjan]

according to my rumour-mill:
"Main memory" refers to the 256MB memory of the CPU. "Local memory" refers to the 256MB memory of the GPU. The CPU can read with ~16MB/s from this local memory, write bandwidth is ~4GB/s. So if you really need to read data from local memory, it'd be better to have the GPU render it to main memory and have the CPU read it from there instead of reading it directly from local memory.

[Saotome]

I have to agree with Arjan.
It's the usual inquirer bullshit.
The 16MB/s is the bandwidth for Cell reading directly from RSX local memory (as opposed to something like a DMA transfer from RSX). The article makes it sound as if its the bandwidth to Cells main XDR memory.
_________________
infj

[Benny Blanco]

Yeah The Inquirer article is so wrong it's laughable. I thought it was really odd too and with a little further investigation (something The Inquirer should try sometime) I came across this, which mirrors what Arjan basically said:

"It isn't at all unusual for the video memory to have incredible write speeds and painfully slow read speeds (back to the CPU that is). The reason is that in 3d graphics the video card does the actual rendering. Therefore you simply tell it "I want a blue triangle at the coordinates X,Y,Z (x3) with T texture applied". The card renders it and applies the texture from texture memory and then displays it onto the screen. You never need to read the (texture) memory, because the data contained in it is throw away (why would you need to read the texture in that you sent to the card?)

So it is perfectly normal for texture memory to be nearly write-only. As long as writing to it is extremely fast (which it is in this case according to the PP slide), that isn't a problem."

[yostane]

Thanks guys , I was going to believe it, but you saved me.

[Viper8896]

i thought it was local mem is the onboard mem on each spe and main mem is the one the available through the mic.

[Son_k]

[GregLee]

[Kojima]

I'm no expert on ps3(or anything else for that matter) but from what I gather, the bandwidth between cell and gpu read wise is painfully slow, BUT the bandwidth between gpu and system ram is very high.
So couldn't you simply dma from the gpu memory to cpu memory, THEN let the cell just read the memory copy.
it's been a while but if I'm right you'd be able to read at over 8gb a sec.

Just an idea, no idea if it's practical/workable. But I'd say there must be a way, cos too many ps3 games supposedly do post fx on the cell and the only way to do post fx on rsx rendered graphics would be some way of dma'ing rsx mem to cell mem. (I think.)

[Arwin]

[Kojima]

[cheriff]

Doesn't the RSX support pixel shaders? I thought on modern pc cards post fx were done by loading up certain pixel shaders and 'rendering' with that. It already has full access to the framebuffer (since runs ON the gfx chip - no need to transfer back to cpu) and various parameters can be passed via textures on which the shader can do lookups.
_________________
Damn, I need a decent signature!

[Kojima]

Yes, standard practice on pcs is to do multi-pass postfx usually involving some shader language or other(I.e cg/glsl/hlsl) entirely on the gpu. (Some involve cpu assistece like creating and uploading a wave map texture for say a vertex shader) but there has been quite a bit of talk about post fx done on the Cell on ps3.
And if that is true, it invalidates the slashdot article completely. (Not that it has a slow read rate directly, that appears to be true, but that the ps3 is broken. My bet is the slashdot writer was just a bit too dense to understand you could circumvent the slow read speed by taking an alternate route from the rsx ram to cell ram)

My guess anyway.