Nvidia DX12 driver holding back Ryzen

[fs123]

It looks like AMD Ryzen's poor performance in some DX12 games is due to Nvidia's DX12 driver. The interesting bit starts at 10:37.

TL : DW
AdoredTv tested a Ryzen and i7 with a 1070 and RX480 mGPU and made a potentially shocking discovery.
The 1800X virtually catches up to the i7 7700K when dual RX480's are used in DX12 mode but significantly lags behind it when a 1070 is used. Since nearly every Ryzen review used a Nvidia gpu, this revelation could invalidate a lot of DX12 benchmarks.


To further validate the above findings, another youtuber did a similar test in The Division and also found that Ryzen ran better with an AMD RX470 compared to a Nvidia 1060 in DX12.

 

[Panos]

Wouldn't be surprised if we find some dodgy code in the drivers

 

[r7slayer]

Well ive noticed that AMD cards seem to get a noticable boost in DX12 where as my findings with my 980ti and 1080ti is that you get a boost but its not noticeable. So there is little gain and sometimes negatives when under DX12 which makes me thing its better for nvidia users to stick to DX11 which intel cpus will most likley dominate. Under DX12 Ryzen will dominate but paired with a AMD card it will shine.

 

[BigDaddyKong]

I am glad this came out. I am running 2 Saphire Fury cards, and I have been scratching my head when looking at reviews. I did not think I was crazy when looking at my results vs theirs.

 

[shankly1985]

VEGA plus Ryzen Plus DX12/Vulkan could be a killer setup! a lot of Bench results will need to be looked at again.

 

[SiDeards73]

VEGA plus Ryzen Plus DX12/Vulkan could be a killer setup! a lot of Bench results will need to be looked at again.

They won't be though, reviewers stand to lose too much if they rebench showing Nvidia DX12 is garbage.

 

[Freddie1980]

Ironically AMD consistently showcased Ryzen with a Geforce 1080 at all the press events (Battlefield 4, Sniper Elite 4) leading up to the launch. You would like to think they would have been aware of this and have mention this in there review note to the press.

 

[PapaLazaru]

They had to show it off the the best GPU of the moment, it would have looked a bit stupid showing it off with a 480.

 

[DarrenM343]

LOL, come on guys. If this is the case it'll not be intentional, NV will want their hardware performing best in all rigs, regardless of the CPU. Maybe it's down to AMD's own poor optimisation at the moment, ie, cpu code, not working well with NV cards/drivers, nobody knows the real picture here . Or maybe even AMD is fuelling this fire to persuade Ryzen owners to dump their NV cards and buy a Vega because they want people to believe it will perform best with their CPU's. WHo knows?
The AMD fans will want to believe one story, the NV lot the other - it's all sad, just geddon and enjoy gaming not read too much into benchmarks as they're flawed too...or believe everything you read.

 

[SiDeards73]

LOL, come on guys. If this is the case it'll not be intentional, NV will want their hardware performing best in all rigs, regardless of the CPU. Maybe it's down to AMD's own poor optimisation at the moment, ie, cpu code, not working well with NV cards/drivers, nobody knows the real picture here . Or maybe even AMD is fuelling this fire to persuade Ryzen owners to dump their NV cards and buy a Vega because they want people to believe it will perform best with their CPU's. WHo knows?
The AMD fans will want to believe one story, the NV lot the other - it's all sad, just geddon and enjoy gaming not read too much into benchmarks as they're flawed too...or believe everything you read.

While I do agree with the point of your comment, the fact remains this points to either Nvidia DX12 drivers being garbage, or Nvidia DX12 support at a hardware level being garbage. It brings us back to the suspicion that Pascal is a beefed up Maxwell.

It's a non issue for us current Nvidia users as we can just use DX11 performance for the time being as AMD still has nothing to compete coupled with the amount of rubbish DX12 implementation currently on the market.

The thing to keep an eye on is how Vulkan scales with both manufacturers, as i feel this is where the real future battle will be.

 

[smogsy]

AdoreTV has very very weird findings. he often igrones certain benchmarks. like in this FX830 video he igrones Starcraft 2 results because it didnt show what he wanted. "his reason was it was a terrible coded game"

 

[Brazo]

Please no god not another excuse for Ryzens performance. How many has there been now I've lost track!

 

[PapaLazaru]

Please no god not another excuse for Ryzens performance. How many has there been now I've lost track!

Chip has been out less than a month, this is what one should expect.

 

[fs123]

Well ive noticed that AMD cards seem to get a noticable boost in DX12 where as my findings with my 980ti and 1080ti is that you get a boost but its not noticeable. So there is little gain and sometimes negatives when under DX12 which makes me thing its better for nvidia users to stick to DX11 which intel cpus will most likley dominate. Under DX12 Ryzen will dominate but paired with a AMD card it will shine.

If you watch the AdoredTV video, even the i7 7700K gets the same fps in DX11 and DX12 when the 1070 is used but actually gains 7% more fps when the 480 mGPU is used in DX12. It proves that Nvidia's DX12 driver is not doing much at all.

AdoreTV has very very weird findings. he often igrones certain benchmarks. like in this FX830 video he igrones Starcraft 2 results because it didnt show what he wanted. "his reason was it was a terrible coded game"

AdoredTV provided the evidence that an AMD gpu gives Ryzen a significant boost in DX12. Nothing weird about that.
The other video also backs up his claim using a different game so pretty hard to deny the visible in-game evidence.

Another bit of evidence that has come up is this PCworld Ryzen review from 2nd March that used a FuryX as the gpu:

http://www.pcworld.com/article/3175...-fire-breathing-apex-of-amd-power.html?page=3

Here the 1800X beats the 5960X in Farcry Primal which is contrary to every other review which shows either Ryzen losing badly or just barely matching Intel in this game. FC Primal is not DX12 so the outlier seems to be the gpu vendor being used and not just the api being used.

 

[Nexus18]

I thought ryzen was damn good especially when compared to the far more expensive intel counterpart?

From what I've seen, despite the max fps not being as high as intel, the actual gaming experience is supposedly better/smoother due to higher min fps etc.? If so, to me, that is far more important as it is the min fps. that you notice the most, at least for myself anyway.

While I do agree with the point of your comment, the fact remains this points to either Nvidia DX12 drivers being garbage, or Nvidia DX12 support at a hardware level being garbage. It brings us back to the suspicion that Pascal is a beefed up Maxwell.

It's a non issue for us current Nvidia users as we can just use DX11 performance for the time being as AMD still has nothing to compete coupled with the amount of rubbish DX12 implementation currently on the market.

The thing to keep an eye on is how Vulkan scales with both manufacturers, as i feel this is where the real future battle will be.

+1

It should be well known by now that dx 12 is pointless for nvidia, at least everything pre-10xx anyway, as they seem to be the only cards that show some gain...

 

[smogsy]

what i think people are missing here is one thing

AMD has most likely tuned AMD drivers to work better with AMD CPU's

not saying Nvidia has good optimized code here & the results are wrong,

but we all know due to the AMD CCx's that Game require different tuning to get the most out of Ryzen.

ashes gained 28% performance boost by changing the way CPU jobs are handled for Ryzen

what if AMD has put some CPU interpretation into AMD GPU driver which is helping performance?

either way good Win for AMD

 

[Rroff]

what i think people are missing here is one thing

AMD has most likely tuned AMD drivers to work better with AMD CPU's

not saying Nvidia has good optimized code here & the results are wrong,

but we all know due to the AMD CCx's that Game require different tuning to get the most out of Ryzen.

ashes gained 28% performance boost by changing the way CPU jobs are handled for Ryzen

what if AMD has put some CPU interpretation into AMD GPU driver which is helping performance?

either way good Win for AMD

nVidia uses a lot of software assistance to do stuff like driver command lists, other forced threading, etc. in the drivers - one of the reasons they pulled up dx11 performance - its possible this is falling foul of SMT on Ryzen. Also from what I understand the "controlling" thread for this is spawned separately onto either the 2nd or last "core" of the CPU which had to be optimised to avoid slowdown with hyperthreading and benefits from high IPC.

I'd assume the AMD driver optimisation could be tested for using older drivers although you'd have to also eliminate application specific performance optimisations between driver revisions also.

 

[Panos]

what i think people are missing here is one thing

AMD has most likely tuned AMD drivers to work better with AMD CPU's

not saying Nvidia has good optimized code here & the results are wrong,

but we all know due to the AMD CCx's that Game require different tuning to get the most out of Ryzen.

ashes gained 28% performance boost by changing the way CPU jobs are handled for Ryzen

what if AMD has put some CPU interpretation into AMD GPU driver which is helping performance?

either way good Win for AMD

If you bothered to watch the video, you will see the Intel with RX480 was better also on DX12 in relative numbers.

What cannot be explained the huge jump on DX12 with Ryzen compared to the DX11.
DX12 supposed to relieve the CPU from much of the work.

Same applies to the overclocked TXP against the the RX480s CF @ 1300, the video shows near the end. The 480s shouldn't be nowhere near the TXP, however they beat it with a big margin.
And where you see TXP, you can put the GTX1080Ti also. And 2 480s cost £400 these days, while the TXP £1180 and 1080Ti £700+

 

[smogsy]

nVidia uses a lot of software assistance to do stuff like driver command lists, other forced threading, etc. in the drivers - one of the reasons they pulled up dx11 performance - its possible this is falling foul of SMT on Ryzen. Also from what I understand the "controlling" thread for this is spawned separately onto either the 2nd or last "core" of the CPU which had to be optimised to avoid slowdown with hyperthreading and benefits from high IPC.

very true,

my point was the oppsite could be in play to AMD has tuned their drivers for Ryzen

 

[Rroff]

I added a bit more to my post on that note as well.