Radeon RADV vs. NVIDIA Vulkan Performance For F1 2017 On Linux

[humbug]

Fantastic progress and kudos to all the contributors of RadV. Good to see the RX 580 beating the GTX 1060 as we move to less CPU bound scenarios. As the driver matures as expected we should hope to see more of these AMD wins given the typical performance of AMD hardware vs Nvidia in Vulkan and DX12 on windows.
i.e
HD 7970 > GTX 680
R9 290 > GTX 970
RX580 > GTX 1060
Vega56 > GTX 1070

BTW something I should add in your favour... there is nothing inherent about DX11-over-Vulkan that makes it *impossible* to write a game which would take full advantage of AMD HW; it's just that typical performance tuning work on DX11 will not lead you there.

I take it certain departments of AMD are actively trying to hammer that point home when you'll talk to game developers?

Raja made a point about a year ago where he said he was trying to increase developer mind share. i.e. make sure more developers are using AMD systems as that can only lead to good things.

[theriddick]

Looks like more optimization needed at 1080p, 4k results are almost on par.

[wdb974]

Nvidia has offered basically the same Windows performance to Linux users and what do they get in return? "You suck, because your driver is closed." Doesn't that sound at least a little bit awkward?

That comment is perfectly fine. An open source OS works best with open source apps and drivers. Closed source stuff simply doesn't integrate well. It's not just nvidia, by the way; stuff like VMware doesn't integrate well either. But, in nvidia's case, their efforts to collaborate with the open source driver team has led to old videocards not being able to run Plasma 5. That and having both Intel and nvidia in laptops will sometimes lead to an absence of GUI after an update (or even a simple reboot? Didn't get to test long enough).

[juno]

Because nvidia driver is closed source, nvidia can cheat in rendering algorithms for speed.

Real-time graphics is nothing but cheating. As long as it doesn't look different, cheating for more speed is a good thing.

open source drivers did render better looking graphics in Tomb Raider 2013 with wine-staging.

doubt


But BTT: nice looking perf for radv. Congrats and thanks to Dave and Bas for this awesome community work and for delivering what AMD can't or doesn't want to

[pracedru]

Very nice to see all these cards perform well. Something that most games didn't do until now. Hardware support is beginning to be ready for Linux as a gaming platform.

It is also nice to see Vulkan being used on Linux and DX11 on windows.

[alex79]

As nVidia 1070 owner all I can is well done for AMD open-source driver. Previously I had AMD 390 and than changed to 1070 because of no support from AMD. If open-source drive continue this growth my next graphic card will be AMD.

[nanonyme]

That comment is perfectly fine. An open source OS works best with open source apps and drivers. Closed source stuff simply doesn't integrate well. It's not just nvidia, by the way; stuff like VMware doesn't integrate well either. But, in nvidia's case, their efforts to collaborate with the open source driver team has led to old videocards not being able to run Plasma 5. That and having both Intel and nvidia in laptops will sometimes lead to an absence of GUI after an update (or even a simple reboot? Didn't get to test long enough).

Actually in many ways VMWare is better supported through OSS on Linux than nVidia. Guest tools are now available as opensource and most paravirtual drivers are part of Linux kernel

[bug77]

Thanks for suggesting that - the first thing I noticed is that there are not any graphs for "full workload" (4K, AA on) which is where I would expect our hardware to do best.

What I see is the following:

1. The NVidia driver hits what appear to be CPU limits at a higher FPS point than radv, so with light GPU workloads all the NVidia cards are faster than all the AMD cards

2. As the workload increases the AMD cards move up and start to interleave with the NVidia cards

3. We ran out of graphs before getting to full workload (4K, ultra-high or AA on or both). I'm not sure that the pattern of "moving up with increasing workload" would continue but based on results from other games it seems pretty likely

4. Michael included a few lower-end AMD cards (550, 560, 285) but no corresponding NVidia cards, so AMD appears generally lower overall than NVidia

BTW something I should add in your favour... there is nothing inherent about DX11-over-Vulkan that makes it *impossible* to write a game which would take full advantage of AMD HW; it's just that typical performance tuning work on DX11 will not lead you there.

Thanks, that's the picture I was getting, too.

[aht0]

just saying, an RX 580 beated at 1080p and 4k ultra an GTX1060 using a community backed driver. I would call that PRETTY DAMN F*** GOOD if you ask me.

I do expect an army of nVidia fanboys to hijack this thread tho(and soon) saying the game is trash and thats why the nVidia driver is not universally faster, bla bla bla but either way RADV pretty damn good job and congratulations, same to FERAL team, you rock

How to parse this, btw? It is unintelligible.

[wdb974]

Actually in many ways VMWare is better supported through OSS on Linux than nVidia. Guest tools are now available as opensource and most paravirtual drivers are part of Linux kernel

VMWare ain't as bad as nvidia. It was just another example of closed source that might give you headaches. I still remember having to jump through hoops just to uninstall it under Debian. And I thought having a .run installer file would make it easy to install and uninstall...