Hello.
I've got a computer with two OS'es installed - ASPLinux11 and WindowsXP SP2. I've tested recently a super_pi mod 1.4(pi caltulation) on both operating systems (I've used same Win32 binary file) . It looks like it works faster under Linux, while it crashes under windowsXP at certain conditions. What's going on? My friend tried same trick on his computer(taht is much more powerful). He got slightly different results. IN his case program works fastr under windows, when calculations aren't very complicated (pi with 16K digits), but it works faster under linux when calculating "pi" with 1M digits. What's going on? I just want to know if this is a proof of WindowsXP being slower, or does that mean that there is some timer-related bug? ASPLinux claims their wine version is 9.5 while there is mention only about 9.24 version on this site.
Here are timing results and links to original program: link to the program: http://www.techpowerup.com/downloads/126/mirrors.php checksum of result can be checked (if provided) at: http://www.xtremesystems.com/pi/
My configuration: 1. AMD Athlon (Thunderbird) 900 Mhz 2. 3x128 Мб DIMM SDRam 3. DFI AK75-EC /OCtect Rhino KT133A-ASP motherboard
WindowsXP timinig: 16К - 1.199 s. 32K - 2.727 s 64K - 6.433 s 128K and more - crashes with an accessviolation.
ASPLinux11 (via Wine) timinig: 16K - 01.199s 32K - 02.339s 64K - 04.676s 128K - 12.409s 256K - 31.005s 512K - 1m 17.630s 1M - 2m 53.044s
results with checksum: ASPLinux11: 64K time 4.703s checksum 06874B42 WindowsXP: 64K time 06.461s checksum 72955C87
My friend's configuration and results (unfortunately he didn't provide any checksum): configuration: GigabitP4Titan 845PE/Cel2400/1Gb/WXPSp2
WindowsXP: 16K - 0.922s 32K - 1.828s 64K - 3.437s 128K - 9.266s 256K - 26.453s 512K - 1m 07.391s 1M - 2m 43.078s
ASPLinux11: 16K - 1.511s 32K - 2.318s 64K - 4.372s 128K - 9.156s 256K - 23.454s 512K - 1m 3.082s 1M - 2m 07.937s
Victor "ErV" Eremin.
"ErV" ErV2005@yandex.ru wrote:
I've got a computer with two OS'es installed - ASPLinux11 and WindowsXP SP2. I've tested recently a super_pi mod 1.4(pi caltulation) on both operating systems (I've used same Win32 binary file) . It looks like it works faster under Linux, while it crashes under windowsXP at certain conditions. What's going on? My friend tried same trick on his computer(taht is much more powerful). He got slightly different results. IN his case program works fastr under windows, when calculations aren't very complicated (pi with 16K digits), but it works faster under linux when calculating "pi" with 1M digits. What's going on? I just want to know if this is a proof of WindowsXP being slower, or does that mean that there is some timer-related bug?
Calculating PI is not a representative OS performance test at all, since it doesn't involve actual OS facilities and mostly uses CPU math power only.
Dmitrydmitry@codeweavers.com wrote:
Calculating PI is not a representative OS performance test at all, since it doesn't involve actual OS facilities and mostly uses CPU math power only.
Well, I'm not talking about a performance. Since calculating PI is a pure CPU-related task, results must be same or linearly dependand. As I understand, WINE does not modify program, just loads it and resolve references to dlls to correct equivalents. But calculation times surely aren't linearly dependant nor equal. Why?
Victor "ErV" Eremin.
On 11/5/06, ErV ErV2005@yandex.ru wrote:
Dmitrydmitry@codeweavers.com wrote:
Calculating PI is not a representative OS performance test at all, since it doesn't involve actual OS facilities and mostly uses CPU math power only.
Well, I'm not talking about a performance. Since calculating PI is a pure CPU-related task, results must be same or linearly dependand. As I understand, WINE does not modify program, just loads it and resolve references to dlls to correct equivalents. But calculation times surely aren't linearly dependant nor equal. Why?
Victor "ErV" Eremin.
You forget, there are different processes between different OSs that are running, which affects cpu load...
Am Sonntag 05 November 2006 19:46 schrieb Aaron Slunt:
On 11/5/06, ErV ErV2005@yandex.ru wrote:
Dmitrydmitry@codeweavers.com wrote:
Calculating PI is not a representative OS performance test at all, since it doesn't involve actual OS facilities and mostly uses CPU math power only.
Well, I'm not talking about a performance. Since calculating PI is a pure CPU-related task, results must be same or linearly dependand. As I understand, WINE does not modify program, just loads it and resolve references to dlls to correct equivalents. But calculation times surely aren't linearly dependant nor equal. Why?
Victor "ErV" Eremin.
You forget, there are different processes between different OSs that are running, which affects cpu load...
What also affects the performance is the performance of the operating systems process managament and memory management. Linux does a pretty good job with memory management, and this can be seen when windows apps are run with wine.
Benchmarks in the past have shown that cpu-dependent operations are equally fast, memory intensive ones are faster in wine, while graphics(2D gdi) are much slower in wine.
In this case, I think, background processes make the difference. Was the windows box free of malware? Any firewall, virus scanner, quick launch tray applets running?
Sunday, November 5, 2006, 10:40:33 PM, Stefan wrote:
Am Sonntag 05 November 2006 19:46 schrieb Aaron Slunt:
On 11/5/06, ErV ErV2005@yandex.ru wrote:
Dmitrydmitry@codeweavers.com wrote:
Calculating PI is not a representative OS performance test at all, since it doesn't involve actual OS facilities and mostly uses CPU math power only.
Well, I'm not talking about a performance. Since calculating PI is a pure CPU-related task, results must be same or linearly dependand. As I understand, WINE does not modify program, just loads it and resolve references to dlls to correct equivalents. But calculation times surely aren't linearly dependant nor equal. Why?
Victor "ErV" Eremin.
You forget, there are different processes between different OSs that are running, which affects cpu load...
What also affects the performance is the performance of the operating systems process managament and memory management. Linux does a pretty good job with memory management, and this can be seen when windows apps are run with wine.
Benchmarks in the past have shown that cpu-dependent operations are equally fast, memory intensive ones are faster in wine, while graphics(2D gdi) are much slower in wine.
In this case, I think, background processes make the difference. Was the windows box free of malware? Any firewall, virus scanner, quick launch tray applets running?
Well, it isn't possible to make OS states identical, since they are different. Situation was as "identical" as possible. No additional background jobs\applets were running. I had opened browser and established connection in both cases. But connection/browser were inactive/doing nothing... It's difficult to make sure what were doing something and what were not...
I guess the reason of speedup is memory managment. I suppose it run slower on my friends' computer because he uses KDE, and that desktop uses more background jobs|memory...
Thanks for the answer.
-
Aaron Slunt -
Dmitry Timoshkov -
ErV -
Stefan Dösinger