Nvidia drivers are handicapping Ryzens performance, says new analysis by AdoredTV

[Ursus]

So what does this mean?
What seems a Ryzen CPU bottleneck, is actually a Nvidia driver bottleneck in this particular game.
Nvidias drivers lower AMD Ryzens performance, by not giving enough threads to the Ryzen CPU, unlike AMD's drivers.
What do you all think, on purpose?

[AliG]

Meh more like laziness. What prior incentive did Nvidia have to optimize AMD CPU performance?

Now that Ryzen seems to be competitive with Intel, you may see that change (just to prevent people from going AMD gfx). But no, I don't think there's too much to this.

[Charles Wirth]

Good watch, makes more sense now.

[vario]

Yeah, it is pretty enlightening.
However it seems that it should help all more than 4 core cpu`s and only in dx12.
Even bulldozer should get a healthy boost :-D.
Someone should do dx12 480 cf ryzen review.Or at the very least, when vega comes out do a comparison.

[zanzabar]

Yeah, it is pretty enlightening.
However it seems that it should help all more than 4 core cpu`s and only in dx12.
Even bulldozer should get a healthy boost :-D.
Someone should do dx12 480 cf ryzen review.Or at the very least, when vega comes out do a comparison.

where are the BD chips with more than 4 cores, are people gaming on g34. i dont see 4 threaded float work getting a boost on BD since if it boosted int threaded work they would have done it to help HT parts. lets just forget BD existed and move on so we dont taint the ryzen.

[Particle]

Oh look, someone wants to argue about the definition of a core.

[vario]

where are the BD chips with more than 4 cores, are people gaming on g34. i dont see 4 threaded float work getting a boost on BD since if it boosted int threaded work they would have done it to help HT parts. lets just forget BD existed and move on so we dont taint the ryzen.

Well, maybe i should word it differently, bulldozers neither do have 4 cores nor 8. They have 4 modules.
TBH i dont know if they would get a boost, but there is a possibility.Anyhow, any intel "morecores" cpus should get a boost and maybe even phenom II x6 ;-)
And im gonna bet you there is someone out there gaming on a g34 :P

[zanzabar]

Oh look, someone wants to argue about the definition of a core.

in this case it has to do with multiple floating point threads so it does matter. this was also brought up last year with socket 2011 not using all of the cores on NV cards while amd cards did.

Well, maybe i should word it differently, bulldozers neither do have 4 cores nor 8. They have 4 modules.
TBH i dont know if they would get a boost, but there is a possibility.Anyhow, any intel "morecores" cpus should get a boost and maybe even phenom II x6 ;-)
And im gonna bet you there is someone out there gaming on a g34 :P

4 cores with SMT is rather spot on for what BD was. I also was not disagreeing with the more cores since lazy threads were a problem on socket 2011 with NV and dx12.


back when g34 came out we were trying to get amd to sell an unlocked one, but they never did. i dont even think people like Dave even had engineering samples that were unlocked. lets home SR will come in a reasonable spec for desktop use.

[xlink]

This has been seen for a while


AMD GPUs performed nearly equally on AMD and Intel set ups (when they were the bottleneck)
nVidia GPUs performed better on Intel by a non-negligible amount.

[Ursus]

This has been seen for a while


AMD GPUs performed nearly equally on AMD and Intel set ups (when they were the bottleneck)
nVidia GPUs performed better on Intel by a non-negligible amount.

What is shady to me is how cpu usage on a Geforce system is that much higher with an intel cpu than with a Ryzen cpu. Nvidia drivers basicly ignore half the cores.

It is an effective way to make ryzen gaming look worse than it really is, what review websites did not use nvidia hardware?

[BenchZowner]

Thread title is misleading, performance is "hindered" on Intel as well as shown in the vid.

What I don't see is overclocking the 7700K + RX480 SLI setup to see if there's a GPU bottleneck instead of a driver like claimed.

[zanzabar]

Thread title is misleading, performance is "hindered" on Intel as well as shown in the vid.

What I don't see is overclocking the 7700K + RX480 SLI setup to see if there's a GPU bottleneck instead of a driver like claimed.

it would be better to test with a socket 2011 and then disable cores.

[StAndrew]

What is shady to me is how cpu usage on a Geforce system is that much higher with an intel cpu than with a Ryzen cpu. Nvidia drivers basicly ignore half the cores.

It is an effective way to make ryzen gaming look worse than it really is, what review websites did not use nvidia hardware?

Because 4 core, 8 thread Intel vs 8 core 16 thead AMD CPU.

I think people use NVIDIA primarily as they offer the best GPUs. No conspiracy here. It's pretty well known NVIDIA drivers don't do well with maximizing thread count in DX12. Compounded with the new Ryzen architecture, looks like NVIDIA has some work to do.

Be interesting to see how quickly they try to fix drivers for a competitors CPU.

[Sparky]

Should be "Currently offer the best GPUs."

It isn't always the case. Just saying

It does seem to me that nvidia's drivers need work. Something's just messed up with them. Shows worse on Ryzen, but still shows some on the Intel systems too.

[Ursus]

Thread title is misleading, performance is "hindered" on Intel as well as shown in the vid.

Watch again. The video shows how going from Radeon to Geforce decreases Ryzen performance way more than it does Intel.

Nvidia drivers are "handicapping" Ryzen performance, handicapping either because of incompetence or something else.

I guess after years of nvidia optimizing their drivers for Intel platforms this was to be expected. Hope this issuehas some priority for them.

[BenchZowner]

There are plenty of variables that you and adoredtv are ignoring.
Starting from the very fact that the comparison starts with a dual-gpu setup vs a single-gpu setup.

[vario]

There are plenty of variables that you and adoredtv are ignoring.
Starting from the very fact that the comparison starts with a dual-gpu setup vs a single-gpu setup.

I was thinking about that too, someone should check SLI 1070`s .More gpus=more core usage ?
It could be.

[Ursus]

There are plenty of variables that you and adoredtv are ignoring.
Starting from the very fact that the comparison starts with a dual-gpu setup vs a single-gpu setup.

I don't think that matters for the point they are making, here's how I see the logic.

1. On a 1070 GTX using 1080P, in what SEEMS to be a CPU limited scenario, the Ryzen scores about 30% lower fps than the 7700K.
2. On a Radeon Xfire 480, the 7700K score increases by about 10% (meaning the Geforce 1070 score on the Intel system was actually GPU limited), but the Ryzen score increases by more than 35%, making the Ryzen system about 3-4% slower than the 7700K system.

The point is, the limiting factor are the Nvidia drivers, not the Ryzen CPU.
Now if this problem does not exist for nvidia in SLI, it still means the low Ryzen score is not a cpu bottleneck, because the cpu is able to offer better performance in a different scenario.
To draw final conclusions someone should test Ryzen with SLI, or Ryzen vs Intel on a single 480RX vs a Geforce 1060 on an extremely low resolution to ensure the bottleneck is the CPU.

[BenchZowner]

Since youtube took over, everybody became a... very educated reviewer...

Most of the vids on youtube are very technically lacking, and the tests are most of the times inconsistent and inconclusive because they always miss something.

[kromosto]

nvidia doing this intentionally is just what i expect from nvidia. it will be surprise if they dont do something like this imo.

[drmrlordx]

nvidia doing this intentionally is just what i expect from nvidia. it will be surprise if they dont do something like this imo.

But why? Nvidia is not competing with AMD in the x86 marketspace. They compete in the dGPU arena. Making it harder for Ryzen owners to run Nvidia graphics cards makes absolutely no sense . . . unless Intel is giving them kickbacks or something.

It's more likely that Nvidia spends more time optimizing for Intel systems since AMD's installed base is comparatively small.

[kromosto]

I am a member of PC community since 1996 (before a proud Amiga user ) and being attracted to hardware since 2002. From all those years I remember things like Nvidia cheating on drivers to increase score on 3dmark or sparing physx for itself only or Nvidia not implementing a directx version completely and/or because of this forces game companies to disable features about that directx version or that stupid gsync while there is something like freesync etc. etc. They just like to play dirty and nonsense so I expect anything from them.

[HTC]

I don't think that matters for the point they are making, here's how I see the logic.

1. On a 1070 GTX using 1080P, in what SEEMS to be a CPU limited scenario, the Ryzen scores about 30% lower fps than the 7700K.
2. On a Radeon Xfire 480, the 7700K score increases by about 10% (meaning the Geforce 1070 score on the Intel system was actually GPU limited), but the Ryzen score increases by more than 35%, making the Ryzen system about 3-4% slower than the 7700K system.

The point is, the limiting factor are the Nvidia drivers, not the Ryzen CPU.
Now if this problem does not exist for nvidia in SLI, it still means the low Ryzen score is not a cpu bottleneck, because the cpu is able to offer better performance in a different scenario.
To draw final conclusions someone should test Ryzen with SLI, or Ryzen vs Intel on a single 480RX vs a Geforce 1060 on an extremely low resolution to ensure the bottleneck is the CPU.

That's what i got from it: it seems nVidia drivers are a big variable when testing CPU's bottlenecking, which totally defeats the purpose of the tests.

No need for SLI or crossfire since it introduces other variables, like scaling, IMO: just test with a good card from each manufacturer.

[Ursus]

Since youtube took over, everybody became a... very educated reviewer...

Most of the vids on youtube are very technically lacking, and the tests are most of the times inconsistent and inconclusive because they always miss something.

I do not understand you.
You don't agree with the word "handicapped" you claim me and others are ignoring "variables" and when I explain myself you come with this vague story about everyone on youtube being wrong?

Please point out where the reviewer is wrong, or where my logic is wrong. I really am open to other opinions but with all the info I see, best case scenario is that nvidia failed to implement true cpu scalability in their drivers, worst case scenario is that nvidia is sabotaging AMD.

I lean towards the first, but history teaches us that nvidia is capable of orchestrating the second possibility as well.

[HTC]

This can even be proved using just ONE CPU (but one card from each manufacturer): just run RotTR using DX11 and DX12 with settings you're absolutely sure will bottleneck the CPU with both cards @ stock and then with the highest overclock on the cards you can get.

If the CPU is bottlenecked "properly", then going from a stock nVidia card to an OCed one should yield margin of error differences and the same should be true for stock AMD card VS OCed one but, if comparisons between manufacturers are allot higher then margin of error, then you'll have your proof right there.

Even if this turns out to be true, it could still be game specific, in which case, more games tested are required for a better "sampling" of this behaviour.