Can this SONY patent be the PixelEngine in the PS3's GPU?

I believe they are asking you about XDR DRAM.
As in off processor memory.

*Smile*
It's interesting. Some of the terms on your list seem familiar.

It's better to learn what they mean. Not just group them into a list.

16x FSAA.. heh, I'm curious how much memory you will say.
 
Panajev said:
but while I see the ratio of Texture Fetches vs Math Ops decreasing, I still do not see a decrease in use of Texture Ops from what we do now, instead I still see an increase ( which seems to be matched by a possible increase in Texture Fetch latency ) just one that is not as fast as the increase in Math Ops usage.
The balance between the ops is dictated by the hardware to programmer, not the other way around. Generally speaking, if lots of texture makes sense on your target platform, you will use them, if it doesn't, you will not. There's nothing inherently good on using one approach or the other just for the sake of it.

As food for thought, can you please expand on this point please ?
Well if you were switching primitive contexts, each switch would effectively stall the pipeline inside the APU, which I think would be much more of a performance issue than the cost of memory copy for the context switch.
Anyway if during fragment shading APU would mimick a pixel shader exactly (not really know info about context of incoming pixels), you could probably just send them data from different primitives at the same time...
 
Fafalada said:
Panajev said:
but while I see the ratio of Texture Fetches vs Math Ops decreasing, I still do not see a decrease in use of Texture Ops from what we do now, instead I still see an increase ( which seems to be matched by a possible increase in Texture Fetch latency ) just one that is not as fast as the increase in Math Ops usage.
The balance between the ops is dictated by the hardware to programmer, not the other way around. Generally speaking, if lots of texture makes sense on your target platform, you will use them, if it doesn't, you will not. There's nothing inherently good on using one approach or the other just for the sake of it.

You are preaching to the converted Fafalada, I am not discussing the ratio of Math Ops and Texture Fetches ( the balance between the ops ).

I am saying that I see, compared to what you see on PlayStation 2, Xbox and GameCube, more textures being used in next-generation games ( not as lok-up tables per-se, just to add detail and variety to the scene ) as well as better texture filtering ( Anisotropic Filtering and Tri-linear ).

AFx4-16 and Tri-linear being default ( hopefully ) would increase the load on the Texture Filtering Hardware as we would increase the number of samples it has to take and blend which might mean longer latency for each Texture Fetch.

Do you really see graphics engines in the next-generation using even less texture variety than what you see in PlayStation 2 games ?
 
Panajev said:
Do you really see graphics engines in the next-generation using even less texture variety than what you see in PlayStation 2 games?
I wasn't really referring to console stuff that much as the shaders there are barely worth their name compared to what's coming up. So while you will likely do a couple more texture lookups relative to those too, the total shader ops will increase exponentially...

Btw, texture variety isn't exactly relevant to number of lookups. If you have tons of memory you can have a different texture for every rendered object - only using a single texture lookup :p
 
IRC in Nov. the original GeForce was out. Didn't it have 4 pipes?

IST, yes, in November 1999, the GeForce1/GeForce256 (NV10) came out with 4 pipes, and it was announced in August or September of that year.

I was merely mentioning what was out at the time of PS2's announcement and PS2's release in Japan, I didn't note when exactly those PC chips came out.
 
David_South#1 said:
I believe they are asking you about XDR DRAM.
As in off processor memory.

*Smile*
It's interesting. Some of the terms on your list seem familiar.

It's better to learn what they mean. Not just group them into a list.

16x FSAA.. heh, I'm curious how much memory you will say.

Let me guess... a bucket loads! :LOL:

Maybe he meant to have typed 16x [pipeline & 4x] FSAA??
 
ERP said:
Wouldn't want to put a number on it at this point. Besides I'm not working on a racing game at the moment.

But I will predict that the real limit will be memory, depending on what you think memory sizes will be it'll give you a solid ballpark number.

The real big incrases this time will be in pixel processing that's where you'll see the huge leaps in performance.

Five year later...., thanks for the anwser ERP (and PC-Engine also).
 
Back
Top