ATI: Linux vs. Windows Vista [Archive]

phoronix

10-30-2007, 01:10 PM

Phoronix: ATI: Linux vs. Windows Vista

Since AMD introduced their new Linux display driver last month, we have published a number of different articles looking at the Radeon performance across their different GPU product generations. This ATI/AMD Linux driver testing and exploration continued this month with the release of the 8.42 driver, which finally introduced AIGLX support for the fglrx driver. One area though we haven't yet analyzed is how their official Linux driver now compares to their much-optimized Windows Catalyst driver. Today, however, we will be looking just at that as we compare the ATI Radeon HD 2900XT 512MB performance under Linux and Microsoft Windows Vista.

http://www.phoronix.com/vr.php?view=11337

robin1979

10-30-2007, 01:12 PM

I must be dreaming.. This can't be true...

conholster

10-30-2007, 01:23 PM

A bit more details in the test would have been good, like what version of DirectX was running on Vista if 10 then I guess the game looked better on vista than on linux which comes at a cost of fps's. Dont want to come off as some windows fanboy but usually these kind of tests dont tend to hold up when examined with microscopes....but yes I thought I was dreaming too :D

Xipeos

10-30-2007, 01:46 PM

That huge performance difference in ET:QW may be caused by a faulty linux driver.

Performance in X² - The Threat increased greatly between 8.40 and 8.42 on my system, but now particles from engine jets render as white billboards (http://img221.imageshack.us/img221/7863/x2screen2of9.jpg) (before, particles were the main cause of slowdowns).
Also, I get an interlaced screen when I run it at 1280x1024.

Who knows what other things fglrx doesn't do, which results in increased performance/decreased visual quality.

But then again, 8.42 still seems geared towards r600, and I have an r500 (x1950pro) ...

Uchikoma

10-30-2007, 02:23 PM

As pessimistic as I'd love to be, I'm not going to be as pessimistic.

...white billboards, LOL. If those were real, you could use your exhaust as weaponry :)

Anyways, at this point I wouldn't say that they're "cheating", but just that they still need many more releases to improve and fix current (and future bugs). We do know they're on a new code base, so it is possible that it's a problem with a driver, not an intentional cheat.

...but time will tell.
[In any case, I'm happy to see Linux pwning Vista. Thanks, that's all from me]

Morlyn

10-30-2007, 02:29 PM

I'd really like to seen screen by screen comparisons of the same sittings to verify that everything looks exactly the same. As amazing as the results are, I don't think I'll buy just yet...

I don't know much about this game, but to keep the results fair, both instances of the game should have been using OpenGL. Although performance between D3D and OpenGL is in theory very similar, it really depends on the machine the two are running.

Svartalf

10-30-2007, 02:40 PM

There's probably some missing rendering operations going on there. In theory, the codebase is mostly the same (if you buy their PR...), so any differences will be due to better fits with one or the other- or missing functionality.

deneb

10-30-2007, 02:43 PM

Looking at the fps numbers, I'd say that rendering of ET:QW on Windows was synchronized to screen refresh (60 Hz), while on Linux that was not the case (is OpenGL vsync even supported in these drivers?). Make sure vsync is not enabled in Windows drivers or game settings and test again, please.

If vsync issues and other potential rendering problems are ruled out, the results seem very impressive indeed.

givemesugarr

10-30-2007, 05:25 PM

the results are pretty impressive.
though, i'm not really surprised. the enemy territory for what i've read around is based on a new engine that isn't optimized specifically for dx10, since they're limited to vista and there are still a lot of xp users that in that case wouldn't be able to play it. since the engine is new and studied to express the full potential of a graphic board it would interesting a benchmark of the same game and settings with the new nvidia driver under linux and windows and a confrontation between the 2 leading graphic producers. i'm prepared to bet a little sum on the nvidia linux driver also beating the vista one.
remember that vista is still a fresh system of microsoft and as that it's still in its beta test phase. only after the first sp maybe it will be a good one.

now, let's return to the bench themselves:
are there texture errors or tearing during normal use of the system with the new ati drivers?! if you didn't experienced issues, may this be caused by fact that the last releases of the driver are optimized to work with the r600 series?! this could really be a reason why the new driver performs so well with the hd board.
and as a test, it would be nice if an amd64 or ia64 version could be tested, since the number of linux users using a 64bit version distro is quite high. how do these games perform with a 64bit version distro?

Ole-Martin Broz

10-30-2007, 06:01 PM

VSYNC endabled on my system XP/Linux, dont lock to framerate, I CAN COMFIRN THE PERFORMANCE, NO SHITTY graphics on my 2900 XT.

Things goes faster running under linux. its soo true. i wrote topic or post about it, with not real evidence.

here was it. the one who got the particel thing fucked had 1950 PRO so the 2900 XT should be correct, btw got demo, might be that i dont see the particles at all, who knows, i havnt seen the game without yet, and dont got money to buy the damn game before 11TH.
performance wise is the 2900 XT good, but at a price for u US. guys, and prolly powerusage. WAIT FOR hd3XXX
ima gonna buy 4x hd3xxx if they come cheap =). hope CF and OC options are avaible in linux soon aswell

remm

10-30-2007, 06:09 PM

Based on those encouraging results, I think it would be time for comprehensive ATI vs NVidia Linux testing, comparing with Win numbers for reference. I've been buying NVidia for a while, but I think this will change in the future with the specification releases and good driver performance (while I've noticed NVidia isn't working very hard on Linux driver performance ...).

aoanla

10-30-2007, 06:13 PM

To be honest, I'm not sure how to interpret these results; either they show that Vista drivers are badly written (and we do know that there were certainly problems for both nVidia and ATI's first attempts at writing for Vista), or they show that the fglrx driver is really fast; or possibly they show both. Comparison with the XP drivers seems like the only way to be sure which is the case...

Ole-Martin Broz

10-30-2007, 06:13 PM

ETQW isnt for 64 bit ;)

Im running xp, and i can tell, it isnt the graphics driver, but all the crap start lagg and such, linux starts with greater fps while cpu usage/mem usage and hdd usage is high.

I dont know, i cant tell for sure, but my hd2900 XT in linux definetly beats my highly overclocked 2900 XT in windows =)

aoanla

10-30-2007, 06:25 PM

ETQW isnt for 64 bit ;)

Yes, but it runs fine in 64-bit systems with 32-bit compatibility libraries. There's a thread full of people running it in Ubuntu Gutsy AMD64 on the Ubuntu forums...

Ole-Martin Broz

10-30-2007, 06:30 PM

Yes, but it runs fine in 64-bit systems with 32-bit compatibility libraries. There's a thread full of people running it in Ubuntu Gutsy AMD64 on the Ubuntu forums...

yeah.. i run it in 64 bit, but since it run 32 bit libaries, means. no improvement, or ami wrong, sounds kind of straight forward in the name: 32 bit libaries.

aoanla

10-30-2007, 06:33 PM

yeah.. i run it in 64 bit, but since it run 32 bit libaries, means. no improvement, or ami wrong, sounds kind of straight forward in the name: 32 bit libaries.

It'll be using the 32-bit builds of libGL, yes, but the kernel fglrx will be 64-bit, and the underlying kernel itself will be 64-bit. Potentially, there is room for some speed up just from the increased kernel efficiency.

(In addition, since the 64-bit fglrx seems to be more likely to cause problems than the 32-bit compile, it'd be nice to see if there is any difference in their performance...)

givemesugarr

10-30-2007, 06:38 PM

ETQW isnt for 64 bit

a multilib amd64 system doesn't have any problems running a 32bit application. the prof is that you can actually see wmv/asf videos, or listen to wma, or run acroread without problems in a 64bit system.
you don't have optimization for 32bit apps, but a lot of apps like kde, gnome, xorg have a better result when running on 64bit instead of running 32bit on 64bit hw (on my system this is about 10-12% of increased speed).
my idea of testing this on 64bit arch is due to the fact that i've heard that the 64bit ati isn't quite as good as the 32bit counterpart, so that a 64bit test could be interesting, so that we can also try to see what changes when compared to 32bit.

eNTi

10-30-2007, 06:42 PM

why not take an nvidia card for comparision too? that would be really interesting.

Svartalf

10-30-2007, 06:43 PM

now, let's return to the bench themselves:
are there texture errors or tearing during normal use of the system with the new ati drivers?! if you didn't experienced issues, may this be caused by fact that the last releases of the driver are optimized to work with the r600 series?! this could really be a reason why the new driver performs so well with the hd board.
and as a test, it would be nice if an amd64 or ia64 version could be tested, since the number of linux users using a 64bit version distro is quite high. how do these games perform with a 64bit version distro?

Heh... This would be a given, considering that the 8.41 driver set was officially only R600- you were dead on your own if you used it with anything else and they actually blocked certain classes of chipsets because they really weren't sure the code would work right on the chip or they were dead certain it'd crash in some very embarrassing ways. Now, having said this, I want, somehow, to get to the bottom of this little interesting turn of events- is it because of missing pieces in the driver's support of OpenGL 2.X or is it because the implementation meshes MUCH better with Linux with those games?

Redeeman

10-30-2007, 09:27 PM

but even with faster drivers, how is compatibility for ati these days? the fglrx drivers a year ago certainly werent compatible with ALOT of stuff, like wine, or just the various opengl apps out there... have the ati provided libgl come to a point where its actually working for more than the ID games there are?

remm

10-30-2007, 09:57 PM

I expect if they do a round of testing, they would test with a variety of applications, including games running on Wine.

Ole-Martin Broz

10-31-2007, 03:17 AM

i cant seem to tell ive had any issues running apps, except some DX games, but they can be run in opengl. like wow warcraft 3, and to be honest, starcraft broodwar goes better on 2900 xt than 8800 GTS/GTX, which kind of suprised me.

well, you all might say, starcraft doesnt lagg on my whatever kind of videocard, but it doesnt lagg, till like 5 minutes in the game, the game start to become active and all that. like while playing with 200+ actions per minute, pro gaming.

well. i got some annoying bugs in like gnome, while using my workspace, thats the issue for ati users primarily i guess, for hd users probaly.

I think ati have gotten this driver nearly perfect for the HD series, just to give the same support all the way down to x700.

But with 64 bit fglrx gnome xorg and all that. it still means that we got a bottle neck >_< the GL! so i'd be expecting very simular results, i dont know, lets throw it in the party of benchmark, and see what happens

zappbrannigan

10-31-2007, 04:23 AM

What about Quake 4? On my 9800 Pro I can't compare the results on XP and Linux. On both plattforms the game is playable but on Linux is looks very ugly. I remember the situation with Doom 3 was the same.

Dragoran

10-31-2007, 06:09 AM

Well I only registered to say something to this comments...
First the game does not use DX10 on windows and even if nobody is stopping you creating this effects using opengl as long as the hardware and driver supports it (nvidia has opengl extensions for this; I am not sure about ATI).
The only two things that can affect perfomance in this case are either vsync which is already mentioned, or AA (enabling it in the game did not have any effect for me using the demo, I had to enable them using the drivers.) But the article should have mentntioned any use of AA/AF.

ivanovic

10-31-2007, 06:25 AM

Okay, I think there might be another thing that *could* cause quite some difference:
CPU-Limitation!
Yes, it does look unlikely when looking at the specs of the cpu used, but the very similar results at 1280x1024 and 1680x1050 do look like there is a cpu limitation. The difference between those resolutions shoud be more than only 0.2 fps (<1%). The CPU used does only have a rather low clockspeed (1.8Ghz) and maybe the 1.1 version of etqw is basically singlethreaded. In such a situation there *might* be some differences due to cpu handling from the side of the operating system, though the other 7 cores should be able to handle the os rather well (no idea if that really is the case on windows... ;) ).
Regarding the changelog of patch 1.2 for etqw there might be some significant changes with that version.

And to have more reliable results, there are two things that I would like to see:
1) Screenshots for image quality comparisons. Do both versions really look identical?
2) A comparison with WinXP as in "is the really bad vista driver at fault". I would not be suprised if it is just the vista driver misbehaving since both, AMD/ATI and NVidia did have quite some problems with getting speed out of their cards under Vista.

Never the less, the performance of the latest AMD/ATI driver seems to be really good. Looks like the guys really did a great job with the new opengl implementation.

Ole-Martin Broz

10-31-2007, 08:13 AM

mitcoes

10-31-2007, 02:00 PM

As there are no more websites that benchmarks linux i would like persistent benchmarking for drivers.

I also would like 2 game tests OpenGl and Directx.

The first one (opengl) with Xp, Vista, Linux native, Wine, XP virtualized on Xp, on Linux, and Vista, Vista virtualized on XP, Linux and Vista, and Linux vurtualized on XP, Linux and Vista

(Starting with Quake3, and one opengl game for each year)

The second one identical, but for directx games.

You can have Nvidia and Ati identical machines, and test them with latest drivers and a driver for quarter for example.

Of course Linux would have better curve than Windows, but it was so far, now, that drivers are closer repeat this tests after each upgrade would help a lot to choose better to play.

Last, but not least, the future of gaming, could be memory sticks (perhaps USB3) with live distributions, with linux and open source dirvers (identical to propietary ones I hope in a near future) that you can play in several different hardware, as PCs, MACs, PS3, NWii, iPhones and other game machines, including Xbox perhaps MS would reject this kind of distribution, but it would save a lot of mpney in development and it would work great even against piracy.

givemesugarr

10-31-2007, 02:49 PM

The first one (opengl) with Xp, Vista, Linux native, Wine, XP virtualized on Xp, on Linux, and Vista, Vista virtualized on XP, Linux and Vista, and Linux vurtualized on XP, Linux and Vista :rolleyes:

you've forgotten to mention solaris, solaris virtualized on vista, on xp, on linux, on solaris and project indiana, then freebsd and freebsd on unix that is virtualized on xp that is virtualized by vista, which is virtualized by kubuntu, which runs on ubuntu.... ;)

Of course Linux would have better curve than Windows, but it was so far, now, that drivers are closer repeat this tests after each upgrade would help a lot to choose better to play. :eek: wow, you've been able to say something without really saying something...

Last, but not least, the future of gaming, could be memory sticks (perhaps USB3) with live distributions, with linux and open source dirvers (identical to propietary ones I hope in a near future) that you can play in several different hardware, as PCs, MACs, PS3, NWii, iPhones and other game machines, including Xbox perhaps MS would reject this kind of distribution, but it would save a lot of mpney in development and it would work great even against piracy.

yes, you're right!!! i'm buying right now an iphone to run unreal tournament 3 on it. i've had it powered by an ati hd 2900 xt 1gb of ram in sli with an nvidia 8800gt 1gb...
;)

after all it seems that you haven't really understand what here at phoronix are benchmarking...

Uchikoma

10-31-2007, 04:41 PM

Okay lets be nice now...-_-

Firstly though mitcoes, I'm kinda curious how we'd do directx testing in linux, and openGL in windows. Linux games don't have support for DirectX for a rather obvious reason(s). [It's not there! *gasp*]

Secondly, virtualization just kills performance. I'm not even sure most gamers would like to run modern games in a virtualized environment especially when there are native clients available.

I don't get the remaining statements in your post though.

Anyhow, that's my thoughts on your proposal. Correct me if I'm wrong ^^

But the idea of persistent benchmarking like what is carried out in Windows systems, would be nice...further down the line when Linux becomes more of a gaming platform with more install base.

givemesugarr

10-31-2007, 05:19 PM

But the idea of persistent benchmarking like what is carried out in Windows systems, would be nice...further down the line when Linux becomes more of a gaming platform with more install base.

wow... say it right: you're the universal translator from star trek!!! it's difficult to think of that from that words...
the problem is always the time and the fact that linux is fully customizable, so there may be differences when running on newer/older versions of xorg, or on kde vs gnome vs evolution... the continuous benchmarks would be interesting but there's a really great work to do that and i don't know if the guys at phoronix have this manpower to do these benchmarks everytime.
maybe assigning them to a list of some willing users would help, but then there would be first the need to set out a policy on how to do the benchmarks before venturing in such a project.

Svartalf

10-31-2007, 05:22 PM

I expect if they do a round of testing, they would test with a variety of applications, including games running on Wine.

You would expect that, now wouldn't you?

Dandel

10-31-2007, 06:56 PM

These results are nice, but i see a few key games that should be tested missing... namely the ones as follows:

Quake 4
Unreal tournament 2004 ( might be good to use Unreal Tournament 3 when that comes out instead )
Serious Sam 2

Malikith

10-31-2007, 11:03 PM

Where are you guys getting that Quake Wars is using Directx 10? Its a opengl game.. Yes under Windows it uses Directx for input and stuff, but NOT for rendering. And besides, I think Michael would had turned off vsync for the benchmark (which I'm pretty sure he did). Hes aware of those things.

Just because Vista has directx 10 doesn't mean every game you run on it is in "Directx 10" it is not. The game has to be coded for Directx 10. Which, there aren't really any out there, maybe one or two, and we know how impressive those were. *sarcasm*

mitcoes

11-01-2007, 08:09 AM

:rolleyes:

:eek: wow, you've been able to say something without really saying something...

sorry for my bad english, I wanted to say that the improvement curve with linux drivers is better than windows becouse it was worse at the beginning

yes, you're right!!! i'm buying right now an iphone to run unreal tournament 3 on it. i've had it powered by an ati hd 2900 xt 1gb of ram in sli with an nvidia 8800gt 1gb...
;)
I wanted to say future Iphones, and smartphones or PDAs,

after all it seems that you haven't really understand what here at phoronix are benchmarking...

I would like to read benchmarks "pears vs pears", Opengl vs opengl, and directx vs directx, and as directx canbe run under linux with wine, cedega or under virtualization I would like to read this, but a real benchmarking of "pears vs pears" if you are running Virtualbox in a system it must be benchmarked too versus Virtualbox in that OS.

Sorry for the misunderstanding, but at least I made you laught

Ole-Martin Broz

11-01-2007, 08:16 AM

givemesugarr

11-01-2007, 09:23 AM

I would like to read benchmarks "pears vs pears", Opengl vs opengl, and directx vs directx, and as directx canbe run under linux with wine, cedega or under virtualization I would like to read this, but a real benchmarking of "pears vs pears" if you are running Virtualbox in a system it must be benchmarked too versus Virtualbox in that OS.

a virtual box on windows to run a windows game it's nonsense... you run virtualized games in wine e cedega since that it's the only way to make them run in linux, and it's obvious that you'd have some loss (not always anyways). running a virtualized windows game in a windows system is just futile. and i don't think that would users want to know about them. what's important is to have a comparison of the same game in windows and linux.
the part of opengl vs opengl is possibile with recent games as enemy territory for example, but not with older games which run on directx. and directx run only on cedega, for what i know, that is a proprietary stuff.
another thing is that you have to have some software good for benchmarks as newer games which, without a linux native client, don't run under non-proprietary software and often don't run on linux in the first one or two months when they're out.
for example the unreal 3 for windows won't run on cedega before the linux client to be ready...
and for a gamer i recommend to have installed both windows and linux in a paravirtualization method since the new processors from amd and intel support the paravirtualization without changes in the kernel. in this way they can run a windows box without closing the linux one when they need to play some particular game that doesn't have a linux client.

aoanla

11-01-2007, 10:48 AM

a virtual box on windows to run a windows game it's nonsense... you run virtualized games in wine e cedega since that it's the only way to make them run in linux, and it's obvious that you'd have some loss (not always anyways). running a virtualized windows game in a windows system is just futile. and i don't think that would users want to know about them. what's important is to have a comparison of the same game in windows and linux.
the part of opengl vs opengl is possibile with recent games as enemy territory for example, but not with older games which run on directx. and directx run only on cedega, for what i know, that is a proprietary stuff.
another thing is that you have to have some software good for benchmarks as newer games which, without a linux native client, don't run under non-proprietary software and often don't run on linux in the first one or two months when they're out.
for example the unreal 3 for windows won't run on cedega before the linux client to be ready...
and for a gamer i recommend to have installed both windows and linux in a paravirtualization method since the new processors from amd and intel support the paravirtualization without changes in the kernel. in this way they can run a windows box without closing the linux one when they need to play some particular game that doesn't have a linux client.

Erm, what? Wine isn't proprietary software, although it attempts to provide a translation layer between the (proprietary) DirectX API and the OpenGL API. Relatively old versions of Wine are okay with many recent games (they lack complete DirectX 9, and all DirectX10 support, but then, it's not like you're missing much in the latter case), including recent Source-engine games.
Now, Cedega is a proprietary fork of Wine, yes. That doesn't mean it works better than it - indeed, recently, a number of people have had better success with Wine than with Cedega, on a range of games.

(And what's this with the implication that "older games use DirectX"? Half-life (the original one) has an OpenGL renderer... it was only with later DirectX versions (8 and later) that people really started ignoring the OpenGL rendering path. And, in any case, "having an OpenGL renderer" != "runs on Linux"...)

In addition, it seems very silly for you to recommend paravirtualisation as a solution for gamers - it is only recently that any virtualisation solution has managed to support hardware 3d acceleration, and as far as I know it is still fairly unreliable. As far as I know, CPU-level paravirtualisation has the same problems - certainly, the last time I looked at the Xen mailing list, even using AMD paravirtualisation, they still couldn't provide hardware 3d, because of the issues in memory mapping between the virtual and real address spaces for DMA type stuff.

givemesugarr

11-01-2007, 11:53 AM

Wine

i wasn't talking about wine but about cedega, which is actually proprietary software, as it is the codeweavers's wine software.
and cedega DO HAS directx support unlike wine which doesn't!!!

In addition, it seems very silly for you to recommend paravirtualisation as a solution for gamers - it is only recently that any virtualisation solution has managed to support hardware 3d acceleration, and as far as I know it is still fairly unreliable. As far as I know, CPU-level paravirtualisation has the same problems - certainly, the last time I looked at the Xen mailing list, even using AMD paravirtualisation, they still couldn't provide hardware 3d, because of the issues in memory mapping between the virtual and real address spaces for DMA type stuff.

maybe i was too harsh on this. i don't know where the paravirt has arrived, but i do know that the newer processors have the virt capabilities and that it's possible to run both windows and linux on the same box without problems. i thought that with the announce made out by the xen people on the half of 2005 by now they would have provided good 3d virtualizazion so i don't know if the 3d is actualle provided or not (as with vmware and similar softwares), but this would also be an interesting point to find out:
how does a paravirtualized linux-windows box behaves. i don't know if someone has tested one yet. phoronix should have a big amount of data of tests on windows and linux and maybe testing out a parav box to see how behaves wouldn't be too difficult.

deneb

11-01-2007, 01:53 PM

i wasn't talking about wine but about cedega, which is actually proprietary software, as it is the codeweavers's wine software.
and cedega DO HAS directx support unlike wine which doesn't!!!
As aoanla said, Wine does have support for DirectX (<10), and it is probably about as complete as Cedega's support.

givemesugarr

11-01-2007, 04:45 PM

Wine does have support for DirectX (<10), and it is probably about as complete as Cedega's support.

wow, i've never known that... maybe it's due to the fact that i don't use wine.
see how many things you learn when talking to other people that know more than you on a certain matter. :)

Dandel

11-01-2007, 09:22 PM

okey theese are the games i wanna test for linux BOTH ATI NVIDIA AND INTEL cross OS.

lets see windows 2000 and 2003 server, if phoronix want the copy of 2003 server i got from the book i bought about win 2k3 they can have it, or just ask microsoft about it.

i used to use windows server, increased my performance in lotsa stuff, strange, but did.

Okey

UT yes the first, Ut 2004 and Ut3 when it comes.
Quake3 and Quake4
Quakewars

Wine :
Warcraft 3
WOW.

and we need dx wine games, any suggestions ? what does ppl game alot through wine thats DX?

How about adding the following titles then:

The Elder Scrolls 4: Oblivion ( has opengl and D3D if i am right )
The Elder Scrolls 3: Morrowind (opengl? )
Dark Messiah of Might and Magic
Far Cry
Gothic 3
Titan Quest ( With it's expansion. )

Malikith

11-02-2007, 07:45 AM

I don't think the list should be very long, maybe 3 games at most for Wine testing. WoW shouldn't be in the test bed because that game requires a monthly fee and I don't think Michael would be interested in doing testing on a game and have to spend 15$ just to do it. (I wouldn't either in a million years)

Just remember that the tests with Wine depend on Wine more importantly than just the graphic drivers. And that I don't think Wine should be the center of attention for testing. People want it, but its NOT native (In theory it is, but thats a different story).

All of the games like UT 2004 and such are all testable in Windows as well. Just remember to use the OpenGL renderer in Windows as well to give a more accurate test, since the Windows version defaults to the Direct3d renderer.

More games will be tested, just let Phoronix do their thing. Wine is not of any importance here. Thats a different can of worms.

aoanla

11-02-2007, 08:07 AM

Just remember that the tests with Wine depend on Wine more importantly than just the graphic drivers.

That, of course, depends on the game - some Wine optimisations for 2d games have been broken with fglrx for some time now, because the fglrx driver doesn't make available the full range of OpenGL extensions that it should. That's, arguably, not Wine's fault.

Of course, you're quite right that this is outside of the remit of the kind of testing Phoronix is thinking of - this kind of full-scale testing would also require an nVidia card as a control, to isolate Wine issues from fglrx issues, and I doubt they have the time to do that many benchmarks...