2.4V updated apps.

[Crunch3r]

Great news! 
BTW, could you please point out test WU that best suitable for speed comparision?
Sure whole bunch gives most precise picture but maybe some faster approach is valuable for comparision of different systems?

Better ask Joe or Mark about this one  this weekend i did compile about 50 versions of the apps so i only provided the apps and msattler and Joe did the speed testing. I only did a quick check if the apps are working 

[Raistmer]

Ok
Hope they post some info about testbench used here. Wonna get comparable results on my systems too
Some database with test times on different app builds/OSes/ CPUs will be useful IMHO

[SkOrPn]

Can someone shed some light on the diferences of the three x64 versions please? Or do these three versions correspond with what type of cpu you might have? i.e sse2, sse3 and ssse3 capable cpu's.

I have a Core 2 Duo E6700 @ 3.2ghz (on a P5B Deluxe) which has all three of these extensions, but Im not sure which extension version would run the best. lol

Any help would be appreciated.

And thanks crunch3r for the hard work,
 
SkOrPn

[Raistmer]

According my own tests SSSE3 64bit under 64-bit OS is the best one for such CPU
So right now probably KWSN_2.4V_SSSE3_MB.exe is the leader (from 2.4V_Windows_x64_SSSE3 archive)

[Crunch3r]

According my own tests SSSE3 64bit under 64-bit OS is the best one for such CPU
So right now probably KWSN_2.4V_SSSE3_MB.exe is the leader (from 2.4V_Windows_x64_SSSE3 archive)

That's what i'm telling people all day long  However... i do see a possibility to gain another 10 to max 15% in performance... but ONLY for the 64 bit app.

Anyhow, we need to get a common base (2.4V changes) for ALL apps. That' Linux,Windows,UNIX before we can start figuring out how to get some more performance...

[Raistmer]

But for AMD 64 leader is SSE2 32-bit version under 64-bit OS it seems.
Is it because worse Intel compiler optimization for Athlons or something wrong with SSE2 64-bit AMD realization? Just curious...

[msattler]

Ok
Hope they post some info about testbench used here. Wonna get comparable results on my systems too
Some database with test times on different app builds/OSes/ CPUs will be useful IMHO

The testbench I used was the 1.43 knabench available in the downloads 'test and benchmark tools' section.
The test WUs used are attached below.  For a short test, just run the 6 FMXXXX wus..  For a longer, more accurate test, use the MBXXXX wus, but they will take much time to run, closer to a full lenght Seti wu.
And I substiuted the 2.2b app (use the one that matches your cpu) for the reference app rather than the old 5.15 that is in the test package.
You had also asked earlier about the best wu to use for comparing apps and platforms.  That is the problem, there is no such wu.  Different AR wus crunch differently with different apps, some are faster on one than another.  So the only way you can test is by running a range of wus and see which app does better overall.
And as far as the result database, Simon has mentioned working on such a thing on the home page of this site.  Hopefully he has the time soon to continue work on that project.

[attachment deleted by admin]

[attachment deleted by admin]

[Raistmer]

I use 1.43 knabench  too, will run with your WUs now
Thank you!

P.S. this test loads only 1 core. As I see from your test file you run it on quad. Did you manage all cores loading with app binding to core (some CPU affinity) ?
imagecfg.exe from test bench doesnt work for me (is it only for NT4 ?)...

[msattler]

I use 1.43 knabench  too, will run with your WUs now
Thank you!

P.S. this test loads only 1 core. As I see from your test file you run it on quad. Did you manage all cores loading with app binding to core (some CPU affinity) ?
imagecfg.exe from test bench doesnt work for me (is it only for NT4 ?)...

No, the testbench only runs a single instance of the test app at a time on one core..  My quad is x64 and is my fastest rig, that's why I run the tests there, because I can test 32 bit and 64 bit apps.

[SkOrPn]

According my own tests SSSE3 64bit under 64-bit OS is the best one for such CPU
So right now probably KWSN_2.4V_SSSE3_MB.exe is the leader (from 2.4V_Windows_x64_SSSE3 archive)

Thanks for the reply, thankfully that is the version I am running. I used the version with graphics "2.4V_Windows_x64_SSSE3_GFX" but when the screen saver starts I get a blank screen "no graphics"... Wonder if there was something else I was sappose to do to get the nice seti 3D screen saver?

Thanks again for your reply.

[Raistmer]

2 msattler
Yes, this testbench uses only 1 core. My question is : is any test-bench exist to correctly test full-cores-load run times?
AFAIK conroe has common L2 cache for cores so real situation for app when it runs in parrallel on all available cores quite different from single core run.
It's interesting to compare full-load run-times on Core2 Duo/Quadro with Athlon 64 X2 version IMHO.

2 SkOrPn I don't use any graphic or screensaver so cant give any help/comments on your situation, sorry.

[msattler]

Yes, this testbench uses only 1 core. My question is : is any test-bench exist to correctly test full-cores-load run times?
AFAIK conroe has common L2 cache for cores so real situation for app when it runs in parrallel on all available cores quite different from single core run.
It's interesting to compare full-load run-times on Core2 Duo/Quadro with Athlon 64 X2 version IMHO.

Interesting point, but I don't know if it's ever been done.  You would have to get each app to run on all four cores at the same time.  I have no idea if some versions would react differently to sharing the L2 cache than others, but I think the app that tests the best on a single core would still be the best to use running on all four.

[Idefix]

I know, i fixed that one  2 hours ago.

Many thanks 

Regards,
Carsten

[Josef W. Segur]

...
is any test-bench exist to correctly test full-cores-load run times?

When running more than one application instance, there is contention for resources. To test all the possible scenarios of different WU angle ranges, offsets of start time, etc. simply would have too many variable effects to permit repeatable tests. That's why the test program shuts down the BOINC service if it's running and does everything on one core.

I don't say a multiple test is impossible, but the amount of time spent testing many scenarios would be more than we can expect volunteer testers to contribute.
                                                            Joe

[msattler]

...
is any test-bench exist to correctly test full-cores-load run times?

When running more than one application instance, there is contention for resources. To test all the possible scenarios of different WU angle ranges, offsets of start time, etc. simply would have too many variable effects to permit repeatable tests. That's why the test program shuts down the BOINC service if it's running and does everything on one core.

I don't say a multiple test is impossible, but the amount of time spent testing many scenarios would be more than we can expect volunteer testers to contribute.
                                                            Joe

If there were a way to test the same app on 2 cores or 4 cores simultaneously, I wouldn't mind knowing if it can be done and trying it..............would it be a hard thing to modify the knabench script to do it,  or really just not worth the bother?