Hello everybody, new benchmark-test at phoronix.com: http://www.phoronix.com/scan.php?page=article&item=882&num=1
speedator wrote:
Hello everybody, new benchmark-test at phoronix.com: http://www.phoronix.com/scan.php?page=article&item=882&num=1
Wait, they tested 3D Mark 2001 and 3D Mark 2003...
Weren't those the exact same benchmarks NVidia was found to be cheating with by including hacks into their drivers on Windows? No wonder they found Windows to be faster in the graphics tests!
Can anyone find a real (ie, not rigged on Windows) benchmark to use?
Thanks, Scott Ritchie
On Saturday October 20 2007 08:40, Scott Ritchie wrote:
speedator wrote:
Hello everybody, new benchmark-test at phoronix.com: http://www.phoronix.com/scan.php?page=article&item=882&num=1
Wait, they tested 3D Mark 2001 and 3D Mark 2003...
Weren't those the exact same benchmarks NVidia was found to be cheating with by including hacks into their drivers on Windows? No wonder they found Windows to be faster in the graphics tests!
Can anyone find a real (ie, not rigged on Windows) benchmark to use?
Thanks, Scott Ritchie
I think that best real benchmarks is real-world applications. For example Unreal Tournament can show performance difference between OpenGL and DirectX in Linux and this can be compared with results on Windows. Of course there is a lot of other good games which can be used as benchmarks. It would be great if someone who have both Linux and Windows on same computer will run few games (preferably both old and new ones in order to test performance of DirectX 7, 8 and 9) at same conditions so we can see what is the real difference in FPS between Linux (WINE) and Windows. Unfortunately personally I havn't Windows installed (only in VMWare for running Autodesk applications) because dual-boot is unacceptable option for me and actually I havn't need in Windows at all, it just isn't technically suitable for my daily tasks. But I think there is some people with dual boot configuration and modern PC who has some games to test and enough free time to compare the results between Windows and Linux.
On Sat, 2007-10-20 at 09:21 +0000, L. Rahyen wrote:
On Saturday October 20 2007 08:40, Scott Ritchie wrote:
speedator wrote:
Hello everybody, new benchmark-test at phoronix.com: http://www.phoronix.com/scan.php?page=article&item=882&num=1
Wait, they tested 3D Mark 2001 and 3D Mark 2003...
Weren't those the exact same benchmarks NVidia was found to be cheating with by including hacks into their drivers on Windows? No wonder they found Windows to be faster in the graphics tests!
Can anyone find a real (ie, not rigged on Windows) benchmark to use?
Thanks, Scott Ritchie
I think that best real benchmarks is real-world applications. For example Unreal Tournament can show performance difference between OpenGL and DirectX in Linux and this can be compared with results on Windows. Of course there is a lot of other good games which can be used as benchmarks. It would be great if someone who have both Linux and Windows on same computer will run few games (preferably both old and new ones in order to test performance of DirectX 7, 8 and 9) at same conditions so we can see what is the real difference in FPS between Linux (WINE) and Windows. Unfortunately personally I havn't Windows installed (only in VMWare for running Autodesk applications) because dual-boot is unacceptable option for me and actually I havn't need in Windows at all, it just isn't technically suitable for my daily tasks. But I think there is some people with dual boot configuration and modern PC who has some games to test and enough free time to compare the results between Windows and Linux.
I can go ahead and try to install Supreme Commander and see if I can get it updated to the latest version. The auto-updater uses the .Net Framework so that makes it a little difficult.
But, it has a performance test / benchmark mode I could run to do some comparisons.
Stephan
On Saturday October 20 2007 09:28, Stephan Rose wrote:
On Sat, 2007-10-20 at 09:21 +0000, L. Rahyen wrote:
But I think there is some people with dual boot configuration and modern PC who has some games to test and enough free time to compare the results between Windows and Linux.
I can go ahead and try to install Supreme Commander and see if I can get it updated to the latest version. The auto-updater uses the .Net Framework so that makes it a little difficult.
For .net support in WINE you can run "winetricks mono12" (you can get winetricks at http://kegel.com/wine/winetricks ). Anyway, testing the latest version of the game isn't very important, only important thing is to test same version of the game on Windows and WINE.
But, it has a performance test / benchmark mode I could run to do some comparisons.
Stephan
It would be great. Don't forget to provide information about hardware configuration used for testing, version of WINE and video drivers along with FPS on Linux and Windows.
On Sat, 2007-10-20 at 09:51 +0000, L. Rahyen wrote:
On Saturday October 20 2007 09:28, Stephan Rose wrote:
On Sat, 2007-10-20 at 09:21 +0000, L. Rahyen wrote:
But I think there is some people with dual boot configuration and modern PC who has some games to test and enough free time to compare the results between Windows and Linux.
I can go ahead and try to install Supreme Commander and see if I can get it updated to the latest version. The auto-updater uses the .Net Framework so that makes it a little difficult.
For .net support in WINE you can run "winetricks mono12" (you can get winetricks at http://kegel.com/wine/winetricks ). Anyway, testing the latest version of the game isn't very important, only important thing is to test same version of the game on Windows and WINE.
But, it has a performance test / benchmark mode I could run to do some comparisons.
Stephan
It would be great. Don't forget to provide information about hardware configuration used for testing, version of WINE and video drivers along with FPS on Linux and Windows.
Hardware:
Intel Core 2 Duo 2.4 GHz 4gb 800MHz DDR2 4-4-4-12 Wine Version: 0.9.47 Video: nVidia 8800 GTX nVidia Driver Version Linux: 100.14.19 nVidia Driver Version XP: 162.18
Windows XP SP2 32-Bit Score: 16901 Ubuntu 7.10 64-Bit Score: 15996
Both performance tests were done with equal graphics settings at 1024x768.
On that note, I noticed that ReplaceFileW still isn't implemented as it's not possible to safe a profile under Supreme Commander. Wasn't there a patch for this? I distinctly remember that because I was going to implement that function until I saw someone else had already submitted a patch with an implementation. What happened with that?
Stephan
On 20/10/2007, Stephan Rose kermos@somrek.net wrote:
On that note, I noticed that ReplaceFileW still isn't implemented as it's not possible to safe a profile under Supreme Commander. Wasn't there a patch for this? I distinctly remember that because I was going to implement that function until I saw someone else had already submitted a patch with an implementation. What happened with that?
The patch was rejected by Alexandre because the patch was implementing that command using Windows API calls, when it should be implemented natively as POSIX/GNU calls.
Feel free to implement that function. The best way to get it accepted is to have another patch that provides a set of corresponding tests, so the behaviour can be checked against Windows. This then ensures that the API is doing the correct thing.
Test cases worth covering are: * using NULL values in all permutations of source, target and backup file; * covering all premutations of file existing and not existing for the source, target and backup file; * how the replace flags affect the behaviour; * caller not having write access to the file being replaced; * different ACL combinations on the source file.
Things worth testing for: * return value from the function; * GetLastError status; * file content of source, target and backup files (and presence thereof); * preservation of the attributes the API says are preserved (creation time, DACLs, etc).
This is a lot to cover, but a simplified implementation with a basic set of tests that cover at least: * using NULL values in all permutations of source, target and backup file; * covering all premutations of file existing and not existing for the source, target and backup file; and checks at least: * return value from the function; * GetLastError status; * file content of source, target and backup files (and presence thereof); * a subset (or none) of the attributes being preserved.
There should also be notes covering what functionality is missing.
NOTE: I am not sure all of the above is possible without using Windows API calls, but that is where tests will help convince Alexandre.
- Reece
Am Samstag, 20. Oktober 2007 15:05:00 schrieb Stephan Rose:
Intel Core 2 Duo 2.4 GHz 4gb 800MHz DDR2 4-4-4-12 Wine Version: 0.9.47 Video: nVidia 8800 GTX nVidia Driver Version Linux: 100.14.19 nVidia Driver Version XP: 162.18
Windows XP SP2 32-Bit Score: 16901 Ubuntu 7.10 64-Bit Score: 15996
Wow, that's almost 95% of Windows performance, and that on a geforce 8800 card which has comparably poor performance. (Although he performance issues seem to be fixed mostly, appart of World of Warcraft).
Am Samstag, 20. Oktober 2007 10:40:33 schrieb Scott Ritchie:
speedator wrote:
Hello everybody, new benchmark-test at phoronix.com: http://www.phoronix.com/scan.php?page=article&item=882&num=1
Wait, they tested 3D Mark 2001 and 3D Mark 2003...
Weren't those the exact same benchmarks NVidia was found to be cheating with by including hacks into their drivers on Windows? No wonder they found Windows to be faster in the graphics tests!
Can anyone find a real (ie, not rigged on Windows) benchmark to use?
Yes, our own (nonexistant) benchmark. I'm afraid that is the only way.
That's a bit much of an effort, but we can also tests where we have the source code available and can modify and recompile it, for example the sdk samples. Those aren't benchmarks, but they give us a rough understanding where our problems are. Just recompile them with gcc to break potential checksumming by the driver, change the filename and run them on windows and wine.
SDK samples aren't suitable for general game benchmarking, and they don't make themselves as good for advertisement, but they are helpful for development.
On 20/10/2007, Scott Ritchie scott@open-vote.org wrote:
Wait, they tested 3D Mark 2001 and 3D Mark 2003...
Weren't those the exact same benchmarks NVidia was found to be cheating with by including hacks into their drivers on Windows? No wonder they found Windows to be faster in the graphics tests!
Can anyone find a real (ie, not rigged on Windows) benchmark to use?
Thanks, Scott Ritchie
Actually, the results look pretty consistent with my personal experience of getting about ~70% of Windows performance on average, although the results vary quite a bit between different applications. What I think is a bit of a shame is that they didn't test with GLSL or FBOs, although to be fair those are non-default registry settings.
H. Verbeet wrote:
On 20/10/2007, Scott Ritchie scott@open-vote.org wrote:
Wait, they tested 3D Mark 2001 and 3D Mark 2003...
Weren't those the exact same benchmarks NVidia was found to be cheating with by including hacks into their drivers on Windows? No wonder they found Windows to be faster in the graphics tests!
Can anyone find a real (ie, not rigged on Windows) benchmark to use?
Thanks, Scott Ritchie
Actually, the results look pretty consistent with my personal experience of getting about ~70% of Windows performance on average, although the results vary quite a bit between different applications. What I think is a bit of a shame is that they didn't test with GLSL or FBOs, although to be fair those are non-default registry settings.
Speaking of which, why do we still not have a GUI option for them in winecfg?
And, more to the point, are they nearly stable enough to be default yet?
Thanks, Scott Ritchie
On 21/10/2007, Scott Ritchie scott@open-vote.org wrote:
Speaking of which, why do we still not have a GUI option for them in winecfg?
And, more to the point, are they nearly stable enough to be default yet?
They should probably be default these days.
H. Verbeet wrote:
On 21/10/2007, Scott Ritchie scott@open-vote.org wrote:
Speaking of which, why do we still not have a GUI option for them in winecfg?
And, more to the point, are they nearly stable enough to be default yet?
They should probably be default these days.
GLSL maybe, but not FBO. It causes major slowdown in most source games. And I'm talking about drop to 0.1 FPS. If that can be fixed then I don't see major reasons why FBO can't be the default. If anything it does fix a lot of lighting and geometry related problems in number of other games.
Vitaliy
On 21/10/2007, Vitaliy Margolen wine-devel@kievinfo.com wrote:
GLSL maybe, but not FBO. It causes major slowdown in most source games. And I'm talking about drop to 0.1 FPS. If that can be fixed then I don't see major reasons why FBO can't be the default. If anything it does fix a lot of lighting and geometry related problems in number of other games.
Is that a regression? At least for HL2 & CSS it never did that for me. (And at least a while ago CSS stress test actually was faster with FBOs)
H. Verbeet wrote:
On 21/10/2007, Vitaliy Margolen wine-devel@kievinfo.com wrote:
GLSL maybe, but not FBO. It causes major slowdown in most source games. And I'm talking about drop to 0.1 FPS. If that can be fixed then I don't see major reasons why FBO can't be the default. If anything it does fix a lot of lighting and geometry related problems in number of other games.
Is that a regression? At least for HL2 & CSS it never did that for me. (And at least a while ago CSS stress test actually was faster with FBOs)
No it's not a regression. Not in Wine at least. I went back to older versions and it still painfully slow. Of course it could be some change in CSS. I also checked with older video drivers and results the same.
The major slowdown comes from ... SterchRect. And even with that disabled FBOs are still slower then default.
Vitaliy.
Am Montag, 22. Oktober 2007 08:24:18 schrieb Vitaliy Margolen:
H. Verbeet wrote:
On 21/10/2007, Vitaliy Margolen wine-devel@kievinfo.com wrote:
GLSL maybe, but not FBO. It causes major slowdown in most source games. And I'm talking about drop to 0.1 FPS. If that can be fixed then I don't see major reasons why FBO can't be the default. If anything it does fix a lot of lighting and geometry related problems in number of other games.
Is that a regression? At least for HL2 & CSS it never did that for me. (And at least a while ago CSS stress test actually was faster with FBOs)
No it's not a regression. Not in Wine at least. I went back to older versions and it still painfully slow. Of course it could be some change in CSS. I also checked with older video drivers and results the same.
The major slowdown comes from ... SterchRect. And even with that disabled FBOs are still slower then default.
Apparently the source engine now uses a depth buffer with stencil, something it didn't use to do in the past. Perhaps some fallout of the Episode 2 extension of the engine.