how bound is the GPU to CPU performence?

notAFanB · Jul 28, 2003

seeing as the last thread was locked. when would we start to see diminshing returns on CPU speeds as regards to overall GFX performence?

do we really need 8GHz CPU if we are following the Xbox route or not? will there come a point where addition CPU PWR is superflouse in closed GFX systems?

the only long term benefit from such massive CPU ratings I can see is for Physics and AI and even then you run into the issue of storage space.

let's try to keep this one locking if we could.

DeadmeatGA · Jul 28, 2003

...

A GPU runs as fast as a CPU could keep feeding it with display list. You still need lots of CPU power in the next gen to keep the GPU busy since GPUs don't do physics nor vertex animation, they are still strictly the domain of CPU processing.

notAFanB · Jul 28, 2003

Re: ...

DeadmeatGA said:
A GPU runs as fast as a CPU could keep feeding it with display list. You still need lots of CPU power in the next gen to keep the GPU busy since GPUs don't do physics nor vertex animation, they are still strictly the domain of CPU processing.

hang on I was led to believe that vertex animation and rudimetry phyics (ie hacks) were being integrated into DX9+?

doesn't the above become problematic for CPU bound engines? any way to offload the setup stage of rendering (triangle stup, display lists etc..) to a different component?

lastly what's the likely ratio for optimal performance?

EDIT: for the sake of argument I'm assuming bt 'fast' you are referring to transforming vertices/performing vertex operations.

DeadmeatGA · Jul 28, 2003

...

hang on I was led to believe that vertex animation and rudimetry phyics (ie hacks) were being integrated into DX9+?

They are??? Microsoft is smarter than I thought they are.

lastly what's the likely ratio for optimal performance?

Varies depending on games. You would want to first maximize the CPU utilization, then add additional rendering passes and effects to keep the GPU busy if you have additional free cycles.

notAFanB · Jul 28, 2003

They are??? Microsoft is smarter than I thought they are.

I don't know, can anyone clarify?

Varies depending on games. You would want to first maximize the CPU utilization, then add additional rendering passes and effects to keep the GPU busy if you have additional free cycles.

like you said it varies. why not tone down CPU requirement if you are going to do a lot of multipurpose rendering passes on the GPU for a lot of titles?

Josiah · Jul 28, 2003

noone uses vertex animation anymore. it uses too much memory and is less flexible. skeletal animation can be done by the video card since DX7 I believe. in most cases (on a PC), a faster GPU makes a bigger difference than a faster CPU. here is a good example:

http://www6.tomshardware.com/graphic/20020819/radeon9700-25.html

jvd · Jul 28, 2003

Well I think more is allways better. But it depends on how much more. If Ms is sitting on a 4ghz chip and 512 megs of ram or a 3.8ghz chp and a gig of ram (throwing numbers out) Then i believe the ram will make much more of a diffrence .

Of course on my pc the higher i can clock my cpu (more to the point the fsb) the more performance I get from my pc. On both my 9700pro and my 9800pro.

Of course on a closed platform with a non bloated os system I think it will be less cpu bound .

jvd · Jul 28, 2003

http://www.beyond3d.com/forum/viewtopic.php?t=7092

dunno why u didn't post to this . But yea. Same question two diffrent ways.

London Geezer · Jul 28, 2003

well, i've been through this already, but at the end of the day, there's more to graphics when playing a game.
i mean i dont know about u but at the end of the day, the burden of "making a game realistic" is not only on graphics alone, but on what changes the things that are on the screen, and come next generation, phisics and animation will be paramount to a realistic game.
and at the end of the day, whether u put that on CPU or GPU, u still have a "chip" doing it. GPU's have been enormously more powerful than CPU so far for the simple fact that they had hardwired features which are ridiculously faster than software rendering.
but in years things will get more and more software-driven and only basic rendering features will be hardwired.
when everything will be software-driven (u know what i mean, when GPU's will only have basic hardwired features, leaving the rest to shaders) then, whether it's done by the CPU or the "GPU", the "best chip" will win.... i dont even think we will have a "CPU" and a "GPU" in years to come, i mean MANY years to come...

notAFanB · Jul 29, 2003

jvd said:
http://www.beyond3d.com/forum/viewtopic.php?t=7092

dunno why u didn't post to this . But yea. Same question two diffrent ways.

seemed like it belonged there. an I was kinda curiouse as to want binds GFX performence to the CPU.

at least using the PC paradigm.

jvd · Jul 29, 2003

notAFanB said:
jvd said:

http://www.beyond3d.com/forum/viewtopic.php?t=7092

dunno why u didn't post to this . But yea. Same question two diffrent ways.

Click to expand...

seemed like it belonged there. an I was kinda curiouse as to want binds GFX performence to the CPU.

at least using the PC paradigm.

no i meant why didn't u post a link to it in this thread. It would make sense for us to be able to refrence both posts would it not?

notAFanB · Jul 29, 2003

ah right didn't come to mind at the time.

jvd · Jul 29, 2003

notAFanB said:
ah right didn't come to mind at the time.

cool

how bound is the GPU to CPU performence?

Similar threads