• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Question about Die Size and Mhz ceilings.

Keysplayr

Keysplayr

Elite Member
Why are GPU Mhz ceilings soooooo much lower that CPU Mhz ceilings at the same die size?

Example:

PIII 1.13Ghz @ 18u
P4 2.0Ghz @ 18u
P4 3.4Ghz @ 13u
GFFX @ 13u ceiling hovers at around 500Mhz.

WTF?

I know they are VERY different architectures, but when it all comes down to it, arent they all still just transistors? Or are they different transistors?

Anyone have a clue? I dont.. 🙂

Thanks,

Keys
 
A GPU does a crap load more work than a CPU. All the work that it has to do generates a lot of heat.

Maybe it's easier to compare the load of work with the number of pipelines and how many stages there are on a CPU or GPU.

A CPU has one pipeline with either 10 stages, AMD, or 20-31 stages, Intel.

A GPU has multiple pipelines up to eight and possibly more and for each one of those pipelines, they have hundreds of stages.
 
i dunno "how much" you understand or not, but some article on Ace's Hardware explain much to me about how to get higher clock speed, by extending pipelines, doing more op per clock, increase the clock, and such, maybe it'll help you a little ( if you aren't already knew those stuff).
Just trying to help.

regards
 
I know they are VERY different architectures, but when it all comes down to it, arent they all still just transistors?
Click to expand...

The quantity of the transistors is a major factor. Current CPUs throw things off a bit as they use so much on die cache which packs nice and dense while chewing up a lot in terms of transistor count, but GPUs have considerably more 'functional' transistors then a processor, and this rate is likely to accelerate for a while yet. Expect the NV5x/R5x0 to be packing close to a quarter of a billion transistors, while we likely still be looking at relatively close to current transistor counts for processors.

Another issue is that processors are hand designed, GPU designes are actually generated by special built computers. Hand designing a processor(be it GPU or CPU) with today's levels of complexity takes years, and the vid market moves way too fast for that. Since processors evolve incredibly slowly they can take time to build them by hand and tweak out every transistor for optimal performance. This may change a bit around 'DirectX 11'(DXNext+1?) level hardware(as at that point we will be pretty much feature complete, more speed will still be needed for a long time however).
 
CPU's are also hand-designed (per Anands article on Prescot) while GPU's are largely computer-designed to meet the 6 month product cycle.
 
Ok that makes a lot of sense. Many more transisitors that are never idle and always working would generate tons more heat. So CPU's are much "less complex?" than GPU's. 2 to 8 more piplines, no cache like cpu's. Interesting. Thanks guys. Its a bit clearer now.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top