CPU Core Scaling Results With GP104

[Sweepr]

Not a lot of CPU tests after Geforce GTX 1080 / GTX 1070 launch out there. They tested CPU core/clock performance scaling with Haswell-E and the new GPUs.

http://www.jeuxvideo.com/dossier/49...-seconde-carte-pascal-du-fabricant/498969.htm

 

[CHADBOGA]

These sorts of results is why I am convinced I will get 10 years out of my i5 3570K @4.0Ghz, and be able to play any game I want and not have my CPU be an unreasonable bottleneck.

 

[happy medium]

looking at those numbers it seems my 2+2 core at 3.8 will not bottleneck a gtx1070?
At least not that much? ANd if I overclock my cpu to 4.4, I should be good.
They are using a 2+2 core @ 3.0.

am I correct?

 

[Sweepr]

looking at those numbers it seems my 2+2 core at 3.8 will not bottleneck a gtx1070?
At least not that much? ANd if I overclock my cpu to 4.4, I should be good.
They are using a 2+2 core @ 3.0.

am I correct?

Yes. Pretty sure 3.8 GHz 2C/4T Skylake-S would do better.

 

[myocardia]

...

 

[LTC8K6]

Didn't we just have a thread in which AOTS absolutely used/needed more cores?

Why does it look like a 3.8 i3 will probably be just fine?

 

[ShintaiDK]

These sorts of results is why I am convinced I will get 10 years out of my i5 3570K @4.0Ghz, and be able to play any game I want and not have my CPU be an unreasonable bottleneck.

6700K is where the action is!

But ye, it will last quite a while.

 

[frozentundra123456]

One critical data is missing. They should have done 6 + 6 overclocked to 3.6 with the 1080.


Edit: or maybe the 6 + 6 with the 1080 is mislabeled. They say 3.0 on the 1080 and 3.6 on the 1070.

No frametimes either.

 

[poofyhairguy]

Didn't we just have a thread in which AOTS absolutely used/needed more cores?

Why does it look like a 3.8 i3 will probably be just fine?

In what universe is a sub-60fps framerate "fine" on this forum? More like "burn the witch, I need eight cores!"

 

[TheELF]

These sorts of results is why I am convinced I will get 10 years out of my i5 3570K @4.0Ghz, and be able to play any game I want and not have my CPU be an unreasonable bottleneck.

Yup testing only in GPU limited scenes(not to mention in game benchmarks) will do this.

Didn't we just have a thread in which AOTS absolutely used/needed more cores?

Why does it look like a 3.8 i3 will probably be just fine?

Same thing, ultra settings at a resolution that the (any) card just can't handle.

 

[LTC8K6]

In what universe is a sub-60fps framerate "fine" on this forum? More like "burn the witch, I need eight cores!"

You are looking at the score of the i3 equivalent at just 3.0 though.

And all were sub-60, so we'd obviously need to drop down from 4K.

A 3.7/3.8/3.9 i3, which is what would be used in reality, is bound to be much better than a 3.0 i3, I would think.

I think that 2C/4T looks like it's holding it's own in AOTS.

 

[AtenRa]

I think that 2C/4T looks like it's holding it's own in AOTS.

Perhaps because of the lack of Async Compute on GTX 1080/70

 

[coercitiv]

looking at those numbers it seems my 2+2 core at 3.8 will not bottleneck a gtx1070?
At least not that much? ANd if I overclock my cpu to 4.4, I should be good.
They are using a 2+2 core @ 3.0.

The i5 is in a decent position to do that, not the i3. Your i3 @ stock would likely still bottleneck these benchmarks, let alone actual gameplay.

Also, the 15% CPU overclock you're entering into the equation does not take into account the overclocked performance of the 1070, since in that scenario one would overclock both components, right?

Yes. Pretty sure 3.8 GHz 2C/4T Skylake-S would do better.

Correct me if I'm wrong, but when disabling cores on the 5820K, the 15MB cache is still used in it's entirety, is it not? I also assume the memory bandwidth was not gimped in any way.

The way I see it, the Skylake i3 @ 3.8Ghz would likely fail to come close to the CPU 4C @ 3.6Ghz benchmark results, meaning it would still bottleneck the 1070 even in some of the benchmarks.

 

[zlejedi]

This test is worthless - if they see no change from 3GHz to 3,6 Ghz then they tested in place limited by gpu not by cpu.

 

[know of fence]

These sorts of results is why I am convinced I will get 10 years out of my i5 3570K @4.0Ghz, and be able to play any game I want and not have my CPU be an unreasonable bottleneck.

Hmm, all games except that EA developed game on frostbite engine show a performance bump from 6/12 even @3 GHz.
Even though honestly I'm not in the market for either 700$ GPUs or 140W CPUs. The CPU bottleneck only gets worse as GPUs get more powerful.
The 8 Cores >> OC turning point is approaching, ever since AMD put 8 slow Jaguar cores into the consoles.
Maybe there really is a sweet spot for 8 slow cores, that are faster than intel's 4/8, isn't that the strategy after all.

This test is worthless - if they see no change from 3GHz to 3,6 Ghz then they tested in place limited by gpu not by cpu.

Measuring CPU bottlenecks with AVG-FPS isn't exactly easy.
There is clearly a change (second to third blue bar from the top) +0.6 GHz adds 5 FPS. For the 1070 it's 1 FPS difference.
If anything these Benchmarks show that raising frequency a little (20%) can't really compensate for not having enough cores or threads.

 

[zentan]

Yes. Pretty sure 3.8 GHz 2C/4T Skylake-S would do better.

Yeah,it should certainly improve the situation significantly

 

[Deaks2]

These sorts of results is why I am convinced I will get 10 years out of my i5 3570K @4.0Ghz, and be able to play any game I want and not have my CPU be an unreasonable bottleneck.

I completely agree. I figure my i5 3570k at 4.4 GHz is equivalent to the listed 4C 3.6 GHz CPU.

I can't justify a platform upgrade for a 10% performance improvement in games.

 

[frozentundra123456]

Hmm, all games except that EA developed game on frostbite engine show a performance bump from 6/12 even @3 GHz.
Even though honestly I'm not in the market for either 700$ GPUs or 140W CPUs. The CPU bottleneck only gets worse as GPUs get more powerful.
The 8 Cores >> OC turning point is approaching, ever since AMD put 8 slow Jaguar cores into the consoles.
Maybe there really is a sweet spot for 8 slow cores, that are faster than intel's 4/8, isn't that the strategy after all.



Measuring CPU bottlenecks with AVG-FPS isn't exactly easy.
There is clearly a change (second to third blue bar from the top) +0.6 GHz adds 5 FPS. For the 1070 it's 1 FPS difference.
If anything these Benchmarks show that raising frequency a little (20%) can't really compensate for not having enough cores or threads.

Only two games of the six show even a 5% increase. The rest are pretty much in the margin of error. The question is whether the six core is running at 3.0 or 3.6, because in the graph for thr 1070 it says 3.6, and it makes no sense to run it slower with the more powerful card.

 

[know of fence]

The question is whether the six core is running at 3.0 or 3.6, because in the graph for thr 1070 it says 3.6, and it makes no sense to run it slower with the more powerful card.

Interesting observation. Jeuxvideo looks like a dodgy site anyway.

Curiously Fallout the game that shows the biggest difference in the posted OP pics shows no difference in their previous 1080 dossier.

 

[CHADBOGA]

Hmm, all games except that EA developed game on frostbite engine show a performance bump from 6/12 even @3 GHz.
Even though honestly I'm not in the market for either 700$ GPUs or 140W CPUs. The CPU bottleneck only gets worse as GPUs get more powerful.

The performance bump is small and doesn't make the 4/4 core CPU's unplayable.

 

[know of fence]

It's important to understand that the amount of the relative increase is besides the point, the results are drowned out by the average. CPU bottlenecks are temporary and intermittent, but in some GPU/CPU pairings they don't occur at all, which should be the goal, so that performance is only limited by the GPU.

On second thought, overall we are in luck: with powerful hardware CPU bottlenecks only occur at high frame rates, at which skipping a frame or two still isn't noticable. Maybe you get occasional long frames that go over 16 ms, at a framerate of 120. But that's far from the perceivable 50ms or whatever the threshold is...
At high settings and resolution frame rate takes a dive, but CPU bottlenecks disappear.

 

[cusideabelincoln]

looking at those numbers it seems my 2+2 core at 3.8 will not bottleneck a gtx1070?
At least not that much? ANd if I overclock my cpu to 4.4, I should be good.
They are using a 2+2 core @ 3.0.

am I correct?

Can't really use these results to extrapolate what would happen on different processors. Even with the cores disabled in their tests, the active cores still have access to a huge amount of L3 cache that'll give a bump in gaming performance, much like we see how the L4 cache gives a bump.

 

[LTC8K6]

I doubt that the cache is making a big difference. Maybe 5% faster?

 

[alcoholbob]

I doubt that the cache is making a big difference. Maybe 5% faster?

It's not.

It's because 4 cores are reaching 100% utilization. This will happen with any modern DX11 game in a city with large numbers of AI models, like in Assassins Creed Unity.

I've seen it happen on Witcher 3 with a 4790K. When I switched to a 5820K I've still seen utilization hit over 70% on a hex core in Novigrad.

Basically what's happening is CPU utilization hits 100% on a quad core, whereas on a hex core due to Windows load balancing you still have performance left on tap so the game isn't bottling necking Windows background or other apps, and you get maybe 1-2 fps more.

 

[Sweepr]

www.expreview.com/47866-4.html