Concerned by Nvidia's current state of DX12 and Vulkan performance

Curlyriff · 7 Oct 2016 at 13:42

Yeah I think if you placed an i5 or i7 at stock clock that most people would be running with a 1070/1080 then you won't see this anyways and Nvidia will still be all good.

What it does show is that really the AMD CPU isn't good enough because you are still losing a number of FPS due to the CPU no matter how you cut it. Yeah performance is better but I am not sure anyone will see that in real world scenarios unless we are suggesting people start buying i3 processors with the latest AMD cards?

Edit: Try checking the tables for 1440p direct next time as well rather than 1080p as as soon as you do that Nvidia with the FX is ahead again. Rather silly to compare to different resolutions to claim its the CPU. All that says is it's clear that Nvidia haven't optimised the drive for the 1080 to run at 1080p as they believe people will use it at 1440p and 4k accoridngly. Tie that in because the fact that it gains nothing over the 980Ti at only 1080p res but has significant increase in performance at the other resolution would bolster that argument/train of thought.

humbug · 7 Oct 2016 at 13:43

Gregster said:
I wonder how many NVidia users run with a crappy AMD CPU?

Even with the 6700K @ 1440P that GTX 1080 is showing signes of slowing compared with the Fury-X, there is a bigger gap at 4K and with faster CPU's.
That means even with the 6700K the GTX 1080 is bottlenecked in some way.

Turn the res down to 1080P and i bet they are just about neck and neck with the 6700K.

Rroff · 7 Oct 2016 at 13:47

humbug said:
Even with the 6700K @ 1440P that GTX 1080 is showing signes of slowing compared with the Fury-X, there is a bigger gap at 4K and with faster CPU's.
That means even with the 6700K the GTX 1080 is bottlenecked in some way.

Turn the res down to 1080P and i bet they are just about neck and neck with the 6700K.

Would be interesting to see the results with a 6c/12t CPU.

Even at 1080p the 1080 is 40% faster than the FX on the 6700K.

Curlyriff · 7 Oct 2016 at 13:50

humbug said:
Even with the 6700K @ 1440P that GTX 1080 is showing signes of slowing compared with the Fury-X, there is a bigger gap at 4K and with faster CPU's.
That means even with the 6700K the GTX 1080 is bottlenecked in some way.

Turn the res down to 1080P and i bet they are just about neck and neck with the 6700K.

Have you not seen the drop down menu that is there? The 1080p for the 6700k has no issues. What are you smoking?

KillBoY_UK · 7 Oct 2016 at 14:02

Gregster said:
I wonder how many NVidia users run with a crappy AMD CPU?

I was thinking the same thing, someone with a 1070/80/980ti/Fury X is 9 times out of 10 going to be useing a Haswell/Skylake (or even Ivybridge) i5 or i7

KillBoY_UK · 7 Oct 2016 at 14:06

humbug said:
Even with the 6700K @ 1440P that GTX 1080 is showing signes of slowing compared with the Fury-X, there is a bigger gap at 4K and with faster CPU's.
That means even with the 6700K the GTX 1080 is bottlenecked in some way.

Turn the res down to 1080P and i bet they are just about neck and neck with the 6700K.

lower the resolution the more the CPU becomes the bottle neck is so it makes sence performance normalises even with a FX cpu. Just look at CPU heavy games like Total war and a like

humbug · 7 Oct 2016 at 14:07

Well no #### Sherlock ^^^^ yet the GTX 1080 is lagging way behind the Fury-X.

Rroff said:
Would be interesting to see the results with a 6c/12t CPU.

Even at 1080p the 1080 is 40% faster than the FX on the 6700K.

Yeah, the thing is tho DX12 still has problem stored up for the future, Draw Call threads are limited to six, thats better than two of DX11 but in years down the line may yet again be a cause for bottlenecks. we see that already in AoTS, at 6 cores the performance stops scaling, both for Intel and AMD.

Vulkan and Mantle are much more future proof with 32 Call threads.

TheRealDeal · 7 Oct 2016 at 14:08

Curlyriff said:
Have you not seen the drop down menu that is there? The 1080p for the 6700k has no issues. What are you smoking?

These results are inline with most dx11 games and not 12. Fury X always seems to be at it's slowest at 1080p and gains ground as the resoltion goes up. At 1080p the gtx1080 has it's biggest lead over the Fury X suggesting there is no Cpu bottleneck. At 1440p the Fury X is much closer and at 4k the Fury X gains on the Gtx1080 again. This is with the 6700k.

On the Amd cpu it's a clear case of a Cpu bottleneck which Fury X handles better than the gtx1080 at 1080p.

Fury X is probably just gaining ground at 1440p/4k due to 1080p being it's weakest resolution in most games and stretching it's legs with it's superior Bandwidth in the 6700k tests. You can see the same with the RX480/390 which are faster than the 1060 at 1440p and slower at 1080p using the 6700k.

Gregster · 7 Oct 2016 at 14:08

KillBoY_UK said:
lower the resolution the more the CPU becomes the bottle neck is so it makes sence performance normalises even with a FX cpu. Just look at CPU heavy games like Total war and a like

Spot on. The CPU becomes the bottleneck and the GPU isn't working as it should and the more the CPU bottlenecks, the closer all GPUs will get. Basics really.

humbug · 7 Oct 2016 at 14:15

Gregster said:
Spot on. The CPU becomes the bottleneck and the GPU isn't working as it should and the more the CPU bottlenecks, the closer all GPUs will get. Basics really.

Closer? the Fury-X overtakes the GTX 1080 and then jumps up and down on it for fun.

Gregster · 7 Oct 2016 at 14:24

humbug said:
Closer? the Fury-X overtakes the GTX 1080 and then jumps up and down on it for fun.

Probably down to the memory bandwidth but you should know that a bottlenecked GPU will nowhere near perform as it should and with a big enough bottleneck, a Titan X will perform roughly the same as a 670. Come on Humbug, get off the defence and use logic.

humbug · 7 Oct 2016 at 14:33

Gregster said:
Probably down to the memory bandwidth but you should know that a bottlenecked GPU will nowhere near perform as it should and with a big enough bottleneck, a Titan X will perform roughly the same as a 670. Come on Humbug, get off the defence and use logic.

Your talking about Nvidia's problem in this context and applying it to both when in fact my point is its just Nvidia's problem.

This is not something elementary to everyone, this is a case where Nvidia's architecture is far more bottlenecked with these low level API's than AMD are.

TheRealDeal · 7 Oct 2016 at 14:33

Gregster said:
Probably down to the memory bandwidth but you should know that a bottlenecked GPU will nowhere near perform as it should and with a big enough bottleneck, a Titan X will perform roughly the same as a 670. Come on Humbug, get off the defence and use logic.

Yea but Nvidia are running into a bigger bottleneck full stop no matter which gpu you pick out on this test. A gtx1060 is losing more performance than an RX480 and they require way less cpu grunt to power compared to an gtx1080. Sure the gtx1080 will suffer the most but the point is in Dx12 at least in this game AMD are ahead in the cpu side of things. Its the reverse of dx11.

humbug · 7 Oct 2016 at 14:36

"the reverse of dx11" right ^^^^^ the point that keeps getting glossed over

Kaapstad · 7 Oct 2016 at 15:44

Gregster said:
Spot on. The CPU becomes the bottleneck and the GPU isn't working as it should and the more the CPU bottlenecks, the closer all GPUs will get. Basics really.

Worse still the more powerful the GPU is the worse it can perform when bottlenecked. What I mean by that is a Pascal Titan could perform worse than a 1070 when both are bottlenecked.

Griffildur · 7 Oct 2016 at 15:48

to be fair I wouldn't expect anyone with a 1070 or 1080 to skimp on the CPU so bottlenecks shouldn't really be an issue. It might be a "what if" scenario but highly unlikely one.

Kaapstad · 7 Oct 2016 at 15:54

Great example of GPUs being bottlenecked in this game.

970 SLI beating 980 SLI at low resolutions.

KillBoY_UK · 7 Oct 2016 at 16:03

Griffildur said:
to be fair I wouldn't expect anyone with a 1070 or 1080 to skimp on the CPU so bottlenecks shouldn't really be an issue. It might be a "what if" scenario but highly unlikely one.

When I was a poor teenager from a poor family I did buy a Ti4600 (from a Geforce 1 DDR), and ran it on my PC for a few months with a then lowly P200 MMX or was it a K6-3 400 (it was almost 15 years ago), as i saved to build the rest of the PC.

Anyway is was very very bottle-necked lol until a saved up for a new MB and Ram + a hand me down Thunderbird 800mhz, pencilled tricked that to over 1.2ghz

humbug · 7 Oct 2016 at 17:05

Kaapstad said:
Great example of GPUs being bottlenecked in this game.

970 SLI beating 980 SLI at low resolutions.

Grid 2 has a very inconsistent benchmark.

@ 1080 970 SLI looks like it has a clear lead and yet at 900P they are all but identical.

Run the again and you would get different results.

Gregster · 7 Oct 2016 at 17:24

humbug said:
Grid 2 has a very inconsistent benchmark.

@ 1080 970 SLI looks like it has a clear lead and yet at 900P they are all but identical.

Run the again and you would get different results.

No it doesn't. I ran that benchmark time and again on both NVidia and AMD and I thought there was something up with it when running on AMD, as it was so far behind, I ran it several times with varying tweaks.

You are also not using any common sense and ignoring a CPU bottleneck to try and justify your lack of knowledge. I can't explain it any simpler than I have done, so if you can't grasp it, not much I can add. You can lead a horse to water but you can't make it drink!

Kaap gets it and understands what is what.