Valve's vision of Linux as a game platform stumbles as benchmarks are released

If you were hoping to move to Linux for good because of the idea that it was finally becoming a good platform for gaming (as seen with Valve’s homegrown distro), then you’re going to have to stick with Windows for now. Basically, a handful of games saw major performance drops compared to the Win10 versions (with initial comments to the article suggesting that these are not isolated cases).

In fact, Valve has actually removed the Linux builds of some games because of poor performance, and hopefully these issues are something that Linus and co. are willing to take seriously (if they want their year of Linux, then this is something that they absolutely cannot afford to ignore).

Now it might be true that Linux allows for better general performance for applications like Blender, but if you are trying to get away from Windows and want to move your game library as well, you may need to wait.

Comments…

is it a linux problem, or a linux driver problem?

Much of the OpenGL support in Linux (including the drivers) is developed by the community and not by Nvidia and AMD.

There’s also proprietary drivers and people report the graphics situation on Linux is improving, but last I read they still have a ways to go.

Then it sounds like Intel/Amd/ATi/Nvidia problem.

They should be encouraging linux gaming and 3d content development.
AFAIK Linux has a much smaller memory footprint and resource cost then
windows.

Very nice results for SteamOS considering that this movement of Steam on Linux is very new. I see that the results are quite competitive.
In Linux, Nvidia is performing well, AMD/ATI needs to improve drivers. I guess this is also something new for game developers and maybe they can optimize their games for Linux soon.

I think that’s wrong for proprietary drivers. Proprietary Drivers (ATI, Nvidia) have their own OpenGL implementations each one.

I just watched the new episode of The Tek and they are very suspicious of that report. When Wendell smells something fishy it’s probably going to flounder.

That system that the tests were done on was ancient to the point of no longer being relevant. That can cause a whole host of issues especially with fully featured Linux. It may have been a completely different outcome with a light desktop version. These people just really didn’t know what they were doing.

It was a dual core system running the new Ubuntu Unity. What could possibly go wrong right?

The open source drivers used by most Linux distributions is Mesa which is its own standalone open source project available for a variety of platforms including Windows. Mesa is not maintained by the Linux community and, in fact, has throughout its history had a number of corporate contributors including, in no small part, Intel.

In its inception Mesa was written with the blessings of SGI’s, developer of OpenGL, legal department and although the project was not tied to SGI Mesa ultimately became the OpenGL implementation of choice by SGI.

http://www.mesa3d.org/intro.html

Apparently when version 9 was released only Mesa and Intel drivers supported the latest OpenGL 3.1 specification.

Well, on the other hand there’s this report: http://blogs.valvesoftware.com/linux/faster-zombies/ (from 2012).
This new benchmark was only done on a single machine, so I wouldn’t over-interpret it.
From my own experience, games that run on both Win and Linux are really comparable, while running Win-only games on Linux (using Wine) usually slows them own by ~20%.
Of course, if you use the open source drivers, performance will be worse, but I’d like to see a Windows benchmark without installed GPU drivers :smiley:

So Half-Life 2:Episode 3(2030),Half-Life 2:Episode 4 Return to The Ravenholm(2040) or Half-Life 3(3007) confirmed?

For those interested in the subject here are some performance benchmarks comparing AMD, Nvidia and Intel graphics sets under Ubuntu and Windows.

AMD

The Radeon cards running on Ubuntu using the open source Mesa drivers version 11.1 ran roughly 18-27% slower than the same card using ATI proprietary drivers under Windows 10.

Intel

The Intel HD chipsets running the open source Mesa drivers version 11.1 ran around 5-28% slower than the same cards running Windows 10 and proprietary drivers.

Nvidia

The Nvidia test was a little different as the Ubuntu test system was running Nvidias proprietary Linux drivers rather than the open source Mesa drivers. Under Ubuntu with proprietary drivers the Nvidia cards ran about the same as they did under Windows 10, occaisionally a smidge better and occaisionally a smidge slower.

It should be noted that under all of the tests Xonotic performed significantly better under Windows including with the Nvidia test. It would seem there is something specific to Xonotic causing slowdowns under Linux.

AMD Update

Apparently ATI recently released PowerPlay patches for cards running open source drivers in Linux and these patches have been shown to boost open source Radeon performance by 200-600%! These patches will be available for mainstream Linux users with the release of Linux kernel 4.5 likely a few months from now.

Up to now, the newer AMD graphics cards on the open-source AMD Linux driver have been limited to whatever (low) frequencies the core and memory clocks are initialized to at boot time. With PowerPlay, they can finally (and dynamically) ramp up when to their rated specifications.

This actually has me kind of optimistic for what linux might be in 5 years.