The Relative Windows vs. Linux Performance For NVIDIA, Intel & AMD

[stiiixy]

Well, since Linux has basically no market share in the gaming industry, they release as cheap as they can and hope for a(ny kind of) RoI.

Which exacerbates the issue. It's a bit of a chicken and egg situation. Low sales for AAA tiles, "look! low numbers and no real uptake. We'll can future linux projects".

They chose to simply use a middleware on a game that quite likely is already using middleware. That comes with inherent limitations unto itself. One's you can not fix if any of the middleware companies don't feel like it. I'm sure there are other reasons, some benign, some innocent and others sheer laziness. And yes, drivers are an issue, but I guess they're riding and hedging their bets on the back of that improving driver situation, take the glory when the drivers are at fault, and fob the blame on to the rivers in the meantime.

At least it's not as bad as hardware companies pro-actively locking people out of attempting to create working software for their hardware.

[mitcoes]

Without WINE (staged) KVM and Xen with PCI pass through tests I think the benchmark is not complete, as it does not offers if the reason are the drivers, the libraries or the OS.
And it is rare, that Vulkan that in theory uses more bare metal direct access to be more OS agnostic makes those numbers, perhaps it is the earlier stage or better MS WOS drivers at this moment.

[nranger]

Thanks for this test. It does completely back Bridgman's statement from the Intel test:



It's not right to always blame the drivers and hardware vendors. Linux ports are mostly bad and it's up to the publishers to get that right.

Agreed +1. I'm hoping that Linux/SteamOS continues to gain visibility in the wider PC gaming community and as part of that, people call out sh*tty ports for what they are. I didn't hear Windows gamers blaming Microsoft for example when Batman Arkham Knight was a disaster after its console port. Especially on Nvidia hardware, Linux as a platform is not the issue. And AMD is even catching up, to the point where I'm thinking my next laptop should use one of their APUs for the best combination of Linux support, battery life, and performance.

[ermo]

@Michael:

Is there a way for you to lock the reference in the graphs to Windows instead of to whichever is faster? IMHO, it would make the graphs look more consistent.

[eydee]

Shouldn't expect any different from games using some shitty wrapper

Isn't it interesting that games using a "shitty wrapper", (eon, source etc.) usually perform great and native ports like shadow of mordor and tomb raider are pathetic? Software racism has to stop!

[bridgman]

Very nice article. It really shows off the power of what you have built with OpenBenchmarking.

I have to admit though, my first thought was not "gee these are bad ports" but "hey we actually have games running on Linux at more or less the same time as Windows". That's a pretty awesome change from a few years ago, and IMO the developers/porters need to be encouraged and supported not beaten.

[rhysk]

Great article. I prefer to see this as motivation for upcoming porters/native developers on Linux to drive to equivalent performance compared to Windows. From these charts it is clear that it can be done.

[Herem]

Well i dunno whether DOTA2(Reborn) should be defined a port or not, since Valve's staff rewrote the engine to support multiple platforms.

You mean they rewrote their engine so they could easily port games to multiple platforms?

[Herem]

Very nice article. It really shows off the power of what you have built with OpenBenchmarking.

I have to admit though, my first thought was not "gee these are bad ports" but "hey we actually have games running on Linux at more or less the same time as Windows". That's a pretty awesome change from a few years ago, and IMO the developers/porters need to be encouraged and supported not beaten.

Do you have any thoughts on the DOTA 2 results? It seems like the OpenGL results demonstrate that Linux just as capable as Windows when some time and effort is spent on optimisations; however, the Linux Vulkan results are lagging (a bit) behind.

I thought AMD's Vulkan drivers were ported straight from the Windows drivers? Are the Windows Vulkan drivers on a different version or maybe just not working as well on Linux yet or do you think it's more likely some difference in Valves Vulkan implementation on Linux?

[bridgman]

Do you have any thoughts on the DOTA 2 results? It seems like the OpenGL results demonstrate that Linux just as capable as Windows when some time and effort is spent on optimisations; however, the Linux Vulkan results are lagging (a bit) behind.

Remember that these charts only show the relative performance between Linux & WIndows. IIRC the absolute performance of DOTA on Vulkan is higher than on OpenGL, so we may be running into some new bottlenecks that need attention. The Vulkan driver is also relatively newer on Linux than it is on Windows, since we were in the middle of developing the new amdgpu stack at the same time Vulkan was coming up.

I thought AMD's Vulkan drivers were ported straight from the Windows drivers? Are the Windows Vulkan drivers on a different version or maybe just not working as well on Linux yet or do you think it's more likely some difference in Valves Vulkan implementation on Linux?

Probably versions are different, although I don't expect that to make a lot of difference. Bigger issue is that everything below the Vulkan/OpenGL driver is all new, as are the Linux-specific bits of those drivers, so there's going to be some more work required to catch up with 10 years of Linux Catalyst tweaking.

It could also be a slight difference between Valve's Windows & Linux implementations, but going any further than guesses requires a lot of developer time.