Linux GPU Performance

CrzyP writes "AnandTech.com has benchmarked the most popular graphics cards from ATI and NVidia on the Linux OS (SuSE 9.1). It is interesting to see that they have also written a custom benchmarking tool which can also be downloaded from the article. Take a look at Kristopher Kubicki's "Linux 3D AGP GPU Roundup" to see how each of the mid to high end cards performed on the Penguin flavored system."

Comments

Kristopher Kubicki

[TheReckoning]

Get that man on the KDE team IMMEDIATELY! His parents obviously had a grasp of the KDE naming convention long ago.

It's funny. Laugh.

Penguin-flavored?

[Anonymous Coward]

Umm, you're supposed to be putting the Linux CDs into the CD drive, not taste-testing them... ;)

What does a penguin taste like, anyhow?

Re:Penguin-flavored?

[temojen]

Herring, probably.

Re:Penguin-flavored?

[kfg]

What does a penguin taste like, anyhow?

Fat chicken.

KFG

ATI vs nVidia

[Eeknay]

Wow, this is very surprising. One would expect with similar Windows benchmarks for the X800 to be matching or beating the 6800 Ultra (depending on drivers of course), so these low X800 scores in Linux really are quite a shock.

Re:ATI vs nVidia

[Tumbleweed]

If low-performing Linux ATI drivers are a shock to you, you haven't been paying attention for a few years. :(

Re:ATI vs nVidia

[tempmpi]

Not really, it is common knowledge that Nvidia's linux binary drivers are much better than Ati's. Not only the performance is better in Nvidia's drivers but they are also more compatible. People often had problems getting ATI's binary drivers working, while Nvidia's drivers are working without problem in most configurations and even problems like 4k stacks were fixed withhin a reasonable time.

...vs. same cards with Windows?

[TrollBridge]

Those numbers are all well and good, but I'd be interested in seeing them side-by-side with the same tests performed (on the same machines, of course) running Windows.

Re:...vs. same cards with Windows?

[DeckerEgo]

Here's what they noticed (more of a summary than a benchmarking):

Although this analysis did draw some pretty strong lines as to where each card stand, we were more interested in how each game performed compared to their Windows counterparts. We drew a lot of conclusions from one of our more recent video card analyses from July. Surprisingly, most of our NVIDIA video cards scaled very similarly. Wine games like Jedi Knight took a 10% to 15% hit in performance compared to the Windows tests that we did just a few weeks ago. Other games like Unreal Tournament 2004 actually showed mild signs of an increase in frame rate on the NVIDIA graphics cards. Wolfenstein: ET generally performed with similar average FPS to our video cards from 2003. However, keep in mind that the drivers used then were almost a year old.

My linux GPUbreakdown from independant tests:

[stratjakt]

NVidia: Sort-of

ATI: Kind-of

Better drivers and licensing please

[Handbrewer]

The situation right now is quite frustrating - all distributions should be able to ship the binary drivers for the vendor kernel. It would make it so much easier, than having to get the kernel source and headers before building the module on your own. Thats an unneccesary burden only placed on our shoulders because of some paperwork. 2nd, id like some better drivers please, the ATI drivers are terrible, please stop treating me as a second rank costumer. My money is as good as anybody elses. Thanks so far NVIDIA, now we just need a better license.

Re:Better drivers and licensing please

[danheskett]

Your money is not as good as everyone elses.

90%+ of the market is Windows. To support your 10% slice of the market is vastly more expensive per sale.

Look at this way. If developing, optimizing, and supporting the drivers for a new line of cards was equally as expensive for Linux as for Windows, at say, $1M USD, and the card was expected to sell 10M units, and 90% of the customers are Windows users or OEMS who sell only Windows, then you are looking at a per-unit cost to develop that Windows driver of about $0.11 USD. For the 10% of Linux users, the cost per unit-sold is $1 USD.

Those are assuming all things are equal. I'd wager that for any given company, developing the Windows drivers, including packaging, tweaking, etc is much easier. Add in that Microsoft through its Windows Harward Quality Labs (WHQL or whatever they call it now) basically will subsidize your development costs, I'd imagine that the cost ratio is not 1:1 but much leaner in favor of developing Windows drivers.

So the bottom line is that, in any company, you try to maximize your profitable customers and ditch your unprotifable customers. If the Linux drivers cost only $2 per unit in volume for each sale generated where as the Windows drivers cost only pennies on that dollar, you have a big problem. Over time ATI and nVIDIA sell an awful lot of chips to OEMs and other re-branders who pay very little for hardware. If you are selling chips at $40 (my guess is that this is a high number.. your typical 800-1000 laptop doesn't cost more than 500-600 to build, and most of that goes to Intel/AMD, Micron for RAM and Samsung for LCDs) a unit that $2 cost is suddenely a huge, huge problem.

Of course these numbers are made-up out of broad cloth. But the ratios are real. Linux drivers would have to cost 10 times less to develop to make you an "equal" customer as a Windows customer.

In fact, you are lucky to get anything from nVIDIA or ATI.

Linux Gaming, In Summary

[TrollBridge]

From TFA:

"High performance gaming on Linux certainly isn't for everyone. We spent weeks preparing for this analysis and we still ran into problems that we could not correct. So many times, we came to a solution for a problem only to find our Linux distribution had some files in a slightly different place or our file dependency tree was completely broken. These are the things that scare away people from Linux."

That is the 100% gospel truth. I couldn't have said it better myself. How then will the Linux community and game publishers overcome this (IMHO) enormous obstacle?

Re:Linux Gaming, In Summary

[danheskett]

Part of the problem is the flexibility that Linux affords.

A highly-componentized system is great for flexibility, but a nightmare for usability and developers.

Microsoft's big-man has a penchant for screaming "Developers Developers Developers", but's he right. Developing a graphics intensive app on the Microsoft platform is a different ball of wax completely from developing on a *nix machine. DirectX works well enough that the developer is able to focus on developing things that add value to their product. There is a lot of that coming from the SDL camp, but truth be told DirectX is a success in the Windows world because of the underlying architectural differences: DirectX being install is a boolean operation: installed or not. (Well, now actually its actually a matter of versions.. all Windows distros have it.. so its just a matter of what version you want to deal with).

Windows has a great success at offering games and graphics heavy packages because of the broad hardware support, the straightforward development model for drivers and the resources MS provides for brand-name developers, and the unified relentless promoting and development of DirectX as a platform.

It's all Mesa's fault!

[Anonymous Coward]

If it weren't for the spinless Mesa developer for switching away from LGPL licence just to please the incompetent xfree crew, we would have major companies contributing open codes, instead of having them taking stuff and releasing slower, buggy, proprietary drivers.

Remember ALSA? It sticks its gun to GPL right down to the driver, and Creative actually donate SBLive driver for it, when the company was already crushing everyone else (Aureal included) sound card market! This should be how Mesa license the code, not the lame, bogus, xfree licence.

Re:Indeed

[Anonymous Coward]

For Windows users not familiar with the process, the kernel must be completely recompiled for ATI or NVIDIA drivers to work.

I don't know about the ATI drivers, but this isn't true for the NVIDIA drivers. You can download an installer from NVIDIA that will create a kernel module for you and places it with the other modules. No need to recompile the kernel at all. Just load the module (if the installer doesn't do this for you) and restart your X server.

What about DRI?

[lspd]

You can find my benchmarks of DRI compatible cards here. They're a first attempt at benchmarking DRI and still need some tweaking.

Eric Anholt's benchmarks of DRI on FreeBSD are here.

Roland Scheidegger's comparison of the three drivers available for the Radeon 9000 (DRI, FGLRX, XIG) is here.

It's a bit surprising that the Radeon 8500 series is completly absent from this comparison. The 8500 and FireGL 8800 are still remarkable video cards.

Proprietary driver hell

[freelunch]

I have an Nforce2 based MB with built-in video..

For a few months I ran Nvidia's proprietary driver but found that their support was poor. Countless people would report the same problem and Nvidia would basically just shrug and not even reply to the postings on their website. Stuff like "not our problem". They were very slow to support 2.6.

And as a gentoo user, I hated the binary installation program.

I finally dumped their stuff and went to the OSS driver. It is much slower, even when just opening new browser windows or xterms. But not having to mess with nvidia installer hell each time I gen a new kernel (which is pretty rare, actually) makes it worth it.

This was a great article, however, because it shows just how much chance and luck there is in getting these drivers to work. Buying the latest and greatest MB and CPU for use with Linux is still a huge unknown for the novice and experienced Linux user alike. And then there is the very real fear of whether it will work after you upgrade your kernel, etc.

Sad to see that Nvidia is the most Linux friendly vendor??

Standardization

[acherrington]

In my opinion this is the best thing to happen to linux in some time. Any time that you can develop standards for an industry, you can finally give a target for competitors to aim at (e.g. each other). This will drive competition and really drive the market forward. I would consider this a first step forward.

After both ATI and nVidia clobber each other with better framerates and better overall performance, I think that a new competitive advantage will develop... perhaps this may be better graphics quality or easier installs.

The real question is

[IceFox]

Planning to plunk down some money soon and what I want to know is: What is the best video card you can get that works in Linux that *doesn't* require binary drivers? I don't perticularly care to be locked into one kernel if given the option. -Benjamin Meyer

Why Linux sucks

[Animats]

Enabling 3D acceleration (DRI) still needs to be done manually by editing the /etc/X11/XF86Config file after running the SaX2 utility. Enabling FSAA must be done by editing the XF86Config file by hand as well (see our AA/AF section for details). After a little more than 8 hours of playing with configurations, we hit paydirt.

Any questions?

This binary driver thing has got to go. As Linux gains desktop market share, pressure will increase to open up the hardware interface to the driver. It's not like hooking OpenGL to the card involves any technology that isn't well known in the industry.

Re:nvidia

[erikharrison]

Have you tried to run multiple kernels with the Nvidia drivers? Everytime I booted into a different kernel, I had to uninstall and reinstall the driver. And what about 4k stacks?

Besides, you didn't answer his question - he said "What's a good card with solid open source drivers?" You said "Nvidia has open source drivers but they suck, you shouldn't care about the binary only drivers."

I'd still advocate a Nvidia or ATI card. ATI makes regular code drops to the DRI and Mesa projects, and the open source drivers are of reasonable quality, and the nv drivers are high profile, with lots of work going into them. These cards are the most likely to see solid render acceleration in the future as XAA is replaced with a new acceleration architecture, so even with the Open Source drivers you'll see best performance with stuff like Composite (the basis of much of the X11 6.8.0 eye candy) with these cards.

Of the two, ATI and Nvidia, the open source drivers seem to be of roughly the same quality in my experience, but the Nvidia binary driver is far superiour to the ATI binary driver. ATI has got more bang for your buck, the GATOS project is working to support a lot of ATI's extra features, and ATI seems minimally more involved in the community with an eye to becoming moreso.

I think that pushes things solidly in ATI's favor if you're absolutely commited to the open source driver. If you're willing to use the binary driver, things become more even - it's ATI's price versus Nvidia's better support for the card under Linux/BSD

Open your wallet

[anti-NAT]

You mean like my Zoran TV tuner card that hasnt worked since the 2.2 series, despite their being public specs and sources for it? Drivers are only maintained as long as the developer is around. And unless you have the skills to write your own drivers (and most of us, including large numbers of application developers, *don't*), the having specs/source or not is irrelevant.

Why not offer to pay an open source developer to update it ? At least you have that option, independent of the manufacturer's support for doing so.

You going to have the same problem with Linux kernel version 4.0, when Nvidia don't provide a driver for their XYZ card in 2010. Problem is, at that time, you won't have access to the specs, so you won't even have the option of paying an open source developer to update the driver for you.