Slashdot Mirror
S3 Linux Driver Outperforms Its Windows Twin In Nexuiz
An anonymous reader writes "Chrome Center has done some benchmarks with the proprietary S3 Chrome 400/500 Driver on Linux and Windows. They compared Nexuiz frame rates on a Phenom II system with a S3 430 GT — the surprising result: The Linux driver outperforms its Windows equivalent, offering frame rates about twice as high on average. The question now: Is the Linux driver that good or the Windows driver that bad?"
What is curious is that only the minimum framerate seems to change (which bumps up the average). The max remains the same, which may indicate that the benchmark is CPU bound.
BBH
Oh FP?
We were all warned a long time ago that MS products sucked, remember the Magic 8 Ball said, "Outlook not so good"
Perhaps the windows driver has some graphic features enabled and the linux one does not (like trlinear filtering, shaders, etc). Not enough tech output to make a good conclusion.
way back in Quake 3 days I thought Linux was running Quake3 faster than Windows on my nvidia card, only to realize the linux driver did not have one of the graphics features turned on, which caused it to run faster with the same in game settings.
There should be benchmarks for how other cards perform as well. It could just be Nexuiz isn't performant under load on windows.
I note that the tests were done using windows vista. I wonder if this could have anything to do with the encrypted video path.
Having never heard of this game I checked out the webpage. I see it's based on the Darkplaces engine, which is based on the Quake engine open sourced by ID Software back in the day.
The game's screenshots reminds me more of Unreal Tournament than Quake though. Has anyone here played it? What does it play like?
How is this at all surprising?
Anyone who's done any serious graphics or GPGPU programming on Linux, and Windows (excluding Vista/7 with their fucked WDDM - and significantly slower api calls) - knows Linux practically always performs as good or better (and admittedly, in some cases slower too) than Windows (true for Intel, nVidia, and S3), assuming you're running on a rather bulky X11 server (eg: xorg), and with custom lightweight X11, things only get better.
And in this case we are entitled to it.
Apparently, the drivers are supposed to be GPL, but no actual source was released.
Member of the 7 Digit UID Club
we know that bitch is a fuck up of an editor. he's a waste of human life. i'm sure his parents hang their heads in shame. fucking shitball.
yet another meaningless statistic of "Program X runs better on System Y because driver Z is faster on said system"..
portfolio
The drivers were Vista not XP. This may be a potential reason why there is a performance difference in the drivers. Personally I would like to see an XP benchmark.
In my day, we had to make our square with only a line segment; We then went outside to punch ourselves silly until that line segment started drifting appart into a mirror image of itself to square. And we liked it when kdawson brought a pack of Zimmas and Rammstein boombox afterwards! (*rubs nipple*)
You mean it's 2fps instead of 1fps?
This is S3 we're talking about right?
yes, just like the Open Source radeon driver outperforms the fglrx at miles! oh, wait..
What does wind river sucking have to do with drivers?
Someway this reminds me to today's xkcd..
That's just a matter of less people working on the open source driver than the closed source one. If we had all of those people who are working on the proprietary driver work on the open source driver instead, I'm sure that the open source driver would be better than the proprietary one currently is. Likely, the main reason they aren't doing this is to avoid letting NVidia in on all their implementation secrets. Even if they GPL it, NVidia could still grab some ideas without actually stealing the source.
The responsibility falls squarely on the shoulders of the FOSS community to convince ATI that GPLing proprietary driver source code is a smart decision in the long run (and actually releasing that source code, unlike S3). In addition to their current software engineers, they'll receive free help from FOSS enthusiasts in the community - finding and fixing driver bugs should become a lot faster. Plus, they'll have a lot of people switch to their graphics cards out of support. This should outweigh any intelligence NVidia could gain on them from the source.
So the main reason the open source drivers suck is that they have barely any people working on them, compared to the proprietary drivers. ATI's making a slightly larger step than NVidia, but they're not exactly making a lot of effort if seen in absolute terms.
By analogy, if S3 releases their source like it promised by saying their drivers are GPL, it should also do better.
Member of the 7 Digit UID Club
driver probably just claims that to be able to use otherwise forbidden GPL_ONLY symbols - those may be the reason that make it that fast.
But I could understand S3. EXPORT_SYMBOL_GPL is spreading all over the kernel, making it harder for developers of closed source kernel modules to actually use interfaces to do their job.
That's the reason, fglrx isn't working on PREEMT kernels (like Sidux and RT Kernels) - some tool writers got the idea of patching the fglrx to tell lies (not Ati) - which was finally 'solved' by adding
if (strcmp(mod->name, "fglrx") == 0)
add__taint_module(mod, TAINT_PROPRIETARY_MODULE);
to modules.c
These statistics are useless. You can't just test one game and draw your conclusions to be that it's the driver.
At least test with more then one game you cherry picked because it gives faster speeds. That's the kind of shit Microsoft does.
I hope this author gets banned from slashdot because I fucking hate these kind of tactics.
I wish the world was like that one that you described, however looking at the long history of open source drivers in linux kernel there are few times where I see a benefit.
I will give you an example from my experience with open source drivers. You call me a troll and mod me down, as have many forums that I tried to get answers out of.
I bought a cheap little gigabit card for a home built NAS. Just to explain, $15 is within my budget and $100 is not, of course if I could afford an Intel 1gb card than most likely I wouldn't have a problem. Than again I would have a gigabit controller and no motherboard. I use to run the NAS in Windows XP. It ran without any problems, running CIFS/SMB and even FTP. I had solid performance, reasonable speed (35MB/sec) however stable meaning it would sustain itself at 35MB without going down.
I make the decision to move to an open source platform, as I wanted to get a LAMP server running on the box and put it to some good use. I installed my favorite distribution (Ubuntu 7.10) and off the bat no gigabit for my controller, no big deal, there were two drivers I could use, however one delivered 12MB/sec and the other 30MB/sec (easy solution, right?). The problem is that the second driver was also unstable. So here are my options.
Driver A:
- Low Speed
- Consistent
- No Sig Faults
Driver B:
- High Speed
- Lots of up and down
- Sig Faults or connection drops
The problem is that there is a known issue with driver b. The name of it escapes me, has 1000 in the name though. And the only answer I received was that I should have spent money I didn't have and if I didn't like it I should either use 100mbps or stop using Linux. Keep in mind that I went through SuSE and Gentoo. With mixed results on both platforms. But no holy grail.
No offense, however the last thing I want to deal with is a panel of self-loving, nothing-is-broken crowd, which will deny that there is even a problem with the driver. There is no email address, the forums aren't always an answer, there is no incentive to release the drivers on time. Once you put 20 developers in a room, you will get a bureaucracy that makes the communists seem like un-organized anarchists. I do not want to be in a position where I have to wait on a driver for the latest card, because of soap-opera drama going down with the group managing the driver. Not to mention the 2 forks you will get. And lastly the last thing I want to hear from a forum is another disappointment in FOSS where I am told that my computer is too cheap to run well, whereas my old copy of XP doesn't complain. What ever happened to "Linux can run on anything"? So far, anything is as long as it isn't "just anything".
Open source works very well on software, however I am still not convinced when it comes to drivers. If its something old (old SCSI driver or tape drive) I'll give you that, but when it comes to recent hardware, good luck. I, like most people, do not care if a driver is open source. An open source version of the ATI driver still won't make it good. And an open source version of the nVidia driver still won't make it any better. How about we clean our own back yard before chastising the world for being proprietary. How about we make Firefox faster, as the version on a closed source platform seem to run better than on my linux box.
--
Regards,
Poor Anonymous Coward
S3 actually cares about Linux now? When did that happen?
The S3 drivers for Windows and Linux are like a Ferrari and Lamborghini. Pretty close by themselves. But the Ferrari (Windows driver) is hitched to a trailer loaded with a backhoe (Windows).
Have gnu, will travel.
It's funny you mention Nvidia since the open source version of the NFORCE drivers were so good that Nvidia tells people to use the in-kernel drivers.
While I don't particularly doubt the test as it it's not like it hasn't happened before, however, they tested one single game. How do we know it's not something with the game that caused it? Just because a game is cross-platform doesn't mean it performs equally in generally (not driver dependent) in different operating systems. They could have done several games or compared it to different graphics cards running the same game in different operating systems if they wanted to make an announcement with better proof.
Question is: was Linux running compiz ?
Cauuse running in 3D mode slows down everything a lot ?
Why should that surprise anyone? They are two OSes. Of course one is going to have better drivers in some cases, worse in other cases. Don't get me wrong, I don't think it's intelligent to overlook fundamental differences that can contribute to those differences though, and certainly in comparing the whole OSes against each other, Linux is leaner (yaya it's a kernel I know, shyaddup), so better performance here shouldn't be shocking to anyone. Even some Windows games seem faster in Wine on Linux, but I don't have any frame rate evidence to back that up, plus you have to make sure the game looks identical (same game feature set on both systems).
Promote true freedom - support standards and interoperability.
Quake style bunny jumping is so off putting
Would it be as off-putting if the character models looked like cartoon rabbits or hares?
The question now: Is the Linux driver that good or the Windows driver that bad?
Wrong question. In fact, that's not even a real question; it's just two sides of the same coin. The real question is whether the Linux driver performs better because it's coded better than the Windows driver, or whether it's because of some deficiencies in the Windows driver architecture, Windows graphics stack or the Windows OS itself. In other words, is this truly a case where Linux is better than Windows? Or is this just a case of one driver being better than another. If it's the latter (as the "question" above implies), they could just write a better Windows driver. Not all that interesting in that case.
It checks every so often (or not so often) that it hasn't been told *at the beginning* "this is A-OK".
And each test takes time where it cannot let the data through just in case it's going to leak information on DRM'd content.
Or do you have the source code???