my theory: how to improve GPU scaling...

[asim1999]

i have a great idea of how games can take full advantage of multi-gpu configs using intel CPUs with hyperthreading.

well basically the game can divide the threads between the GPUs for improved task scheduling, lower latency and better performance potential. 

for example: say u are running a 5930K, here is how the threads will divide between each graphics card:

1 GPU; 12 threads

2-way SLI/Crossfire: 6 threads per GPU

3-way SLI/Crossfire: 4 threads per GPU

4-way SLI/Crossfire: 3 threads per GPU

Please let me know what you think of my theory

thanks

 

[WereCat]

Doesent matter how good can be scaling done when only 1 or 2  of 10 games is optimized for PC and rest is just a console port locked for 30FPS 742p.

Also I dont think that scaling works like you described it.

[Enderman]

CPU threads do completely different processing tasks that have nothing to do with the GPUs.

 

I think you dont seem to understand the underlying electrical engineering that links the graphics card and CPU together...
maybe if you go to computer engineering at a university for five years then you might understand that what you just said is complete nonsense...

[Altecice]

Yes but in reality the other cards in SLI/Crossfire mirror the top card, So they are only doing the extra calculations and then passing it onto the top card. I think this solution seems overly simple and if DEVs could do this they would have done it already   

[-rascal-]

A GPU has 2000+ cores...so 2000+ threads if you look at it that way.

With today's multi-GPU technology, data is mirrored onto each card.

 

If you are talking about using all 12 threads on the CPU, then that's different. The game has to be CODED to detect all 12 threads and must be programmed to use them.

If the programmer only coded the game to use 4 threads, then, the remaining 8 will do nothing...since they are not "instructed" to do anything.

[asim1999]

meh, worth a try: i was thinking more in the context of CPU intensive games

[Johannes_Lazor]

The main problems that i can see without thinking for half a hour when im tired
is that they has to work Together
and another problem is that if the loads on the cores are different the time it takes for it to finish will be different
And therefore different parts of the game might become out of sync.
And it would be hard to make it work like that, Everything will have to be built around that instead of what is currently used.


It is a good theory,
But not enough rethinking.

[asim1999]

i just sound so stupid now lol: seemed like i had a brainwave earlier

[Johannes_Lazor]

i just sound so stupid now lol: seemed like i had a brainwave earlier

Meh, Most people skip thinking things through sometimes.

[themaniac]

well i see logic in this in theory if you can get a cpu's thread or cores to help process graphics information aswell as getting the gpu to help with processing the stuff the cpu does it would in theory lower the work load on both cpu and gpu and overall make everything perform better. thats what i think it would do