Sign in with
Sign up | Sign in
Your question
Solved

GPU benchmarks dependency on CPU

Last response: in CPUs
Share
June 4, 2010 5:11:36 AM

Hi everyone,

I am re-wording the language of this thread in an attempt to see if Someone remembers the review in question or if they have opinions backed up by benchmarks.

I am trying to find a thread which was posted here on Tom's on the topic of How Dependent the GPU is on the CPU in Benchmarking/Performance results.

The Review [I am almost sure] was here on Tom's and the results were Hotly Debated by members either praising the results or criticizing it because they felt the games the author chose actually favored some attributes of certain CPU's being tested. Several GPU's were matched up against several CPU's and benchmarks were given.

The authors conclusion was that GPU Benchmark results had Very Little depenency on which CPU was used i.e. saying 'it doesn't matter what CPU you have [ i.e. core 2 quad, i5, i7, phenom x4 etc...] and that gaming benchmarks are practically entirely determined by the GPU in use.

Does anybody remember this review? If not perhaps discussion can start with the hopes that either someone will stumble across it or a majority has an opinion.

Thanks!
Cog

Related resources
a b à CPUs
June 4, 2010 2:34:37 PM

I'm pretty sure the authors concluded that it was important to have a good CPU, otherwise you can't run dual-GPU setups or games that are CPU-intensive very well.
m
0
l
June 4, 2010 4:26:01 PM

Bluescreendeath said:
I'm pretty sure the authors concluded that it was important to have a good CPU, otherwise you can't run dual-GPU setups or games that are CPU-intensive very well.



NO ... that's exactly the point of my inquiry ... The Author concluded that: To a Great Degree the CPU had VERY LITTLE impact on the benchmark of the GPU. :o 

That's exactly why I am trying to find the article. The opinion you have is the common opinion but this article showed that current games are NOT dependant on 4 Monster CPU cores for the GPU to give great benchmarks; i.e. the same benchmarks are produced by the GPU, independant of the quad CPU in use.

Ah well .. I know I wasn't hallucinating because I read it twice. I thought I saved the link but apparetly, and obviously, I forgot to.
m
0
l
June 4, 2010 4:56:49 PM

That's exactly what i gave you, no?
Then do give me 1 when you find out.

I'd remember there's another review conclusion about no difference between a stock 2.66GHz and the overclocked 3.8GHz i5-750 to GPU bottleneck.
m
0
l
a b à CPUs
June 4, 2010 6:08:29 PM

cognitoo said:
NO ... that's exactly the point of my inquiry ... The Author concluded that: To a Great Degree the CPU had VERY LITTLE impact on the benchmark of the GPU. :o 

That's exactly why I am trying to find the article. The opinion you have is the common opinion but this article showed that current games are NOT dependant on 4 Monster CPU cores for the GPU to give great benchmarks; i.e. the same benchmarks are produced by the GPU, independant of the quad CPU in use.

Ah well .. I know I wasn't hallucinating because I read it twice. I thought I saved the link but apparetly, and obviously, I forgot to.


As bluescreen said, once you start using the monster GPU setups then the CPU begins to matter more. Even in a very GPU dependent game, once the frames are running fast enough the slower and even somewhat faster modern CPUs have trouble keeping up with the calculations.

However I will agree that with most single GPU cards, most CPUs can keep up at stock settings, however to say this always happens (even in GPU dependent games) would be silly. I'm guessing conclusion was made using a mid-high range single GPU.



See what happens when stronger GPUs are used (5870 crossfire). A lot of modern CPUs today have trouble keeping up with a 2.0 or 2.4ghz i7. Now see how much more is gained by scaling up to 4ghz in Crysis (which is GPU dependent)?

http://www.legionhardware.com/images/review/ATI_Radeon_...
m
0
l
June 4, 2010 8:46:02 PM

Raidur said:
As bluescreen said, once you start using the monster GPU setups then the CPU begins to matter more. Even in a very GPU dependent game, once the frames are running fast enough the slower and even somewhat faster modern CPUs have trouble keeping up with the calculations.

However I will agree that with most single GPU cards, most CPUs can keep up at stock settings, however to say this always happens (even in GPU dependent games) would be silly. I'm guessing conclusion was made using a mid-high range single GPU.

http://www.legionhardware.com/images/review/ATI_Radeon_HD_5870_Crossfire_CPU_Scaling_Performance_Part1/Crysis_01_Corei7vsPhenomIIX4.png

See what happens when stronger GPUs are used (5870 crossfire). A lot of modern CPUs today have trouble keeping up with a 2.0 or 2.4ghz i7. Now see how much more is gained by scaling up to 4ghz in Crysis (which is GPU dependent)?

http://www.legionhardware.com/images/review/ATI_Radeon_...




Thanks for that link Raidur ... However soooo many CPU's that aren't the "gtx 9xx (4.0 Ghz)" at stock can indeed be Overclocked to that level. Furthermore, that chart doesn't state what CPU's are being used ... only that they have increased GHz relative within that list. So it can be assumed by that chart that any CPU (in the 9xx series) that can be clocked to that level will produce those results. Further validating the results of the review in question?
m
0
l
a b à CPUs
June 4, 2010 9:29:02 PM

Yes, it is safe to assume all 9xx i7s will perform exactly the same clock4clock.

Here is some more CPUs using 5870 crossfire.

m
0
l
a b à CPUs
June 4, 2010 9:38:55 PM

I know these examples are somewhat extreme using 5870 crossfire, just pointing out that modern CPUs today matter in the super-high end, which means down the road another couple years, when the 5870 crossfire won't be super high-end, todays modern CPU's differences will matter.
m
0
l
June 4, 2010 10:00:18 PM

Excellent .. so GPU benchmarks DO NOT depend on the CPU ... only the point to which it is clocked. Hence my i7 860 overclocked to 3.89GHz will produce relatively the same benchmarks as the 9xx Extreme at 5 times the price will. For that matter someone's i5 750 will also achieve these benchmarks overclocked to ~4GHz.

http://www.corsair.com/systembuild/report.aspx?report_i...



Sooo, why do Gamers strive to pay $$$$$$ for Super Monster CPU's? :ouch: 

It apparently Is true that GPU benchmarks are relatively Not dependant on the CPU With Respect To the Quad core i5's and i7's. Otherwise I'm just not getting it. A few years down the road there will be i99 11xx GTX's and none of us will have any use for what we have now except for the few that went Super Monster and they will also be pining for the new i99 11xxGTX's and of course by then the rest of us will buy the i97 10xxGTX's and be able to produce the same results with overclocking...

m
0
l
June 4, 2010 11:09:42 PM

cognitoo said:
Sooo, why do Gamers strive to pay $$$$$$ for Super Monster CPU's? :ouch: 


1. People paying $1000 for a CPU are probably also willing to pay $1000 to SLI the fastest graphics cards they can find, so they can take advantage of it.
2. Upgrading a GPU is much easier than upgrading a CPU, with a much longer upgrade path. I was still able to upgrade the GPU in my old P4 machine to one that was about 4x faster a couple of years after Intel stopped making CPUs that were compatible with the motherboard; that allowed me to play the then-current games at medium to high settings as the CPU could still keep up.

That said, I still don't think that paying $1000 for a CPU on a gaming system makes sense unless gaming is your primary hobby.
m
0
l

Best solution

a b à CPUs
June 5, 2010 5:41:20 AM

cognitoo said:
Excellent .. so GPU benchmarks DO NOT depend on the CPU ... only the point to which it is clocked. Hence my i7 860 overclocked to 3.89GHz will produce relatively the same benchmarks as the 9xx Extreme at 5 times the price will. For that matter someone's i5 750 will also achieve these benchmarks overclocked to ~4GHz.

http://www.corsair.com/systembuild/report.aspx?report_i...



Sooo, why do Gamers strive to pay $$$$$$ for Super Monster CPU's? :ouch: 

It apparently Is true that GPU benchmarks are relatively Not dependant on the CPU With Respect To the Quad core i5's and i7's. Otherwise I'm just not getting it. A few years down the road there will be i99 11xx GTX's and none of us will have any use for what we have now except for the few that went Super Monster and they will also be pining for the new i99 11xxGTX's and of course by then the rest of us will buy the i97 10xxGTX's and be able to produce the same results with overclocking...


Different CPU archs still matter, notice the difference between the E8xxx series and i3, or i7 and Phenom II x4. Newer architectures have a higher IPC, meaning they are faster 'per clock (or per mhz)'.

The reason the i7 9xx extreme, 9xx, 8xx, and i5 750 all perform the same clock4clock is because they are virtually the same CPU. The only differences between these CPUs (besides turbo and stock frequencies/multipliers) are; The extremes have an unlocked multiplier, the i7-9xx are on the 1366 socket, the i7-8xx are on the 1156 socket, and the i5-750 has no HT. HT doesn't help in gaming, and the multiplier differences only help in overclocking headroom (all CPUs are capable of 4ghz however with a good mobo).
Share
June 6, 2010 6:31:45 AM

Best answer selected by cognitoo.
m
0
l
!