Making conclusions about performance comparison without taking into account the cycle can certainly be very misleading.
Quite!
Thank you guys for the look-see. I'm going to "sit on this" for another week or so and see how it trends.
Had the times shown no improvement on just one AMD box, I would not have even posed the question yet. Showing no improvement on two boxes with almost identical "insides" seemed unlikely to be caused by WU freq cycle on both boxes.
But, being a low probability event does not preclude its occurance.
Only the first post and the last Y posts (of the X posts in this thread) are displayed.
and a hotlink like:
Click here to also display the remaining posts.
Cheers, Mike.
Yes, I see it now! Thank you, Mike.
I was looking for a specific post number using the "find in page" function.
Then I looked for something like what you describe at the end of the page, suspecting that some planned exclusion was at work. But I missed that it was tucked in between two posts in place of the omitted posts. (I'm also set for "display latest first" here.)
I guess it is getting near time to get that cataract surgury done! :(
The version of BOINC your using 5.10.13 is pretty dated. I'd try going up to version 6.2.19 or 5.10.45 at the very least and see if you don't get better results with that
Bear in mind that Version 6.2.19 installs as a service by default, and the project files are installed elsewhere than c:/boinc/projects
Waiting for Godot & salvation :-)
Why do doctors have to practice?
You'd think they'd have got it right by now
I know thats not on topic, but is there a date for Optimized Linux Apps release? Both 32bit and 64bit?
My Core 2 Quad Q6600 @2,8Ghz (Win32) need about 24,5k seconds and the
Phenom X4 9500 @2,2Ghz (Linux x64) need about 45k seconds for the same WU.
The stock Linux app (which features an SSE2 enabled version used automatically whenever possible) is still a bit more efficient than the 6.05 Win power user app discussed here, so at least from a POV of "cross-platform fairness", there is little need for new Linux apps. Of course if improvements are found, they will result in further optimizations, but now that the same source code can be used for all three major OSes, it's easier to bring any new optimizations to all platforms at the same time.
The stock Linux app (which features an SSE2 enabled version used automatically whenever possible) is still a bit more efficient than the 6.05 Win power user app discussed here, so at least from a POV of "cross-platform fairness", there is little need for new Linux apps. Of course if improvements are found, they will result in further optimizations, but now that the same source code can be used for all three major OSes, it's easier to bring any new optimizations to all platforms at the same time.
CU
Bikeman
Are there any improvements in using SSE3 or SSE4? SSE4a or SSE4.1 would be exclusive for now, but SSE3 is quite widespread now, with all Intel Macs having processors that support this and pretty well all processors from the last two years. I haven't seen any improvement from GCC using SSE3 over SSE2, but it may be faster using hand-tweaked code.
Are there any improvements in using SSE3 or SSE4? SSE4a or SSE4.1 would be exclusive for now, but SSE3 is quite widespread now, with all Intel Macs having processors that support this and pretty well all processors from the last two years. I haven't seen any improvement from GCC using SSE3 over SSE2, but it may be faster using hand-tweaked code.
I would not expect miracles, but some % should be possible. If I remember correctly, SSE3 introduced some nice instructions for complex vector multiplications and integer truncation.
CU
BRM
Wow... Just switched my main BOINC OS over from XP 32-bit to Linux x86_64 for 64-bit BOINC projects. But even just using the 32-bit applications, I see quite a significant speedup. XP averaged 32,000 seconds, or about 8 hours and 50 minutes. Using Linux averages about 26,000 seconds, or 7 hours and 15 minutes. To shave 100 minutes off, just switching from Windows to Linux, there shouldn't be that big a difference? XP was running the 6.05 SSE2 app.
EDIT: Dang it, I still have 6.04 tasks in there... Dunno why, but XP was averaging about 25,500 with SSE2, which makes a better comparison. Also means SSE2 shaves almost 2 hours off the computation time on my machine, a Core 2 Duo 2.53GHz, 6MB L2 cache (T9400 Penryn if anyone's looking).
RE: Making conclusions
)
Quite!
Thank you guys for the look-see. I'm going to "sit on this" for another week or so and see how it trends.
Had the times shown no improvement on just one AMD box, I would not have even posed the question yet. Showing no improvement on two boxes with almost identical "insides" seemed unlikely to be caused by WU freq cycle on both boxes.
But, being a low probability event does not preclude its occurance.
Stan
RE: Only the first post and
)
Yes, I see it now! Thank you, Mike.
I was looking for a specific post number using the "find in page" function.
Then I looked for something like what you describe at the end of the page, suspecting that some planned exclusion was at work. But I missed that it was tucked in between two posts in place of the omitted posts. (I'm also set for "display latest first" here.)
I guess it is getting near time to get that cataract surgury done! :(
Stan
RE: The version of BOINC
)
Bear in mind that Version 6.2.19 installs as a service by default, and the project files are installed elsewhere than c:/boinc/projects
Waiting for Godot & salvation :-)
Why do doctors have to practice?
You'd think they'd have got it right by now
I installed the app on an
)
I installed the app on an Intel Celeron 550 2GHz Laptop and on my AMD Phenom 9100e Quad.
Times are:
Appr. 37000 seconds on the laptop
Appr. 64000 seconds on my AMD Quad
Unfortunately, I cant compare them to results wih the previous standard app...
I'm surprised that the AMD needs so much longer! Is this normal behaviour?
Cheers!
My NEW BOINC-Site
Why people joined BOINC Synergy...
I know thats not on topic,
)
I know thats not on topic, but is there a date for Optimized Linux Apps release? Both 32bit and 64bit?
My Core 2 Quad Q6600 @2,8Ghz (Win32) need about 24,5k seconds and the
Phenom X4 9500 @2,2Ghz (Linux x64) need about 45k seconds for the same WU.
RE: I know thats not on
)
The stock Linux app (which features an SSE2 enabled version used automatically whenever possible) is still a bit more efficient than the 6.05 Win power user app discussed here, so at least from a POV of "cross-platform fairness", there is little need for new Linux apps. Of course if improvements are found, they will result in further optimizations, but now that the same source code can be used for all three major OSes, it's easier to bring any new optimizations to all platforms at the same time.
CU
Bikeman
RE: The stock Linux app
)
Are there any improvements in using SSE3 or SSE4? SSE4a or SSE4.1 would be exclusive for now, but SSE3 is quite widespread now, with all Intel Macs having processors that support this and pretty well all processors from the last two years. I haven't seen any improvement from GCC using SSE3 over SSE2, but it may be faster using hand-tweaked code.
RE: Are there any
)
I would not expect miracles, but some % should be possible. If I remember correctly, SSE3 introduced some nice instructions for complex vector multiplications and integer truncation.
CU
BRM
For what it's worth, about
)
For what it's worth, about 14% improvement on an Opteron 285 using Vista Ultimate 64 bit.
I feel bad for my wingman, had a little app_info screw up and lost 20 results, so whoever you are please hang in there. Sorry.
Wow... Just switched my main
)
Wow... Just switched my main BOINC OS over from XP 32-bit to Linux x86_64 for 64-bit BOINC projects. But even just using the 32-bit applications, I see quite a significant speedup. XP averaged 32,000 seconds, or about 8 hours and 50 minutes. Using Linux averages about 26,000 seconds, or 7 hours and 15 minutes. To shave 100 minutes off, just switching from Windows to Linux, there shouldn't be that big a difference? XP was running the 6.05 SSE2 app.
EDIT: Dang it, I still have 6.04 tasks in there... Dunno why, but XP was averaging about 25,500 with SSE2, which makes a better comparison. Also means SSE2 shaves almost 2 hours off the computation time on my machine, a Core 2 Duo 2.53GHz, 6MB L2 cache (T9400 Penryn if anyone's looking).