Why not?

Betreger
Betreger
Joined: 25 Feb 05
Posts: 991
Credit: 1,550,265,188
RAC: 714,761
Topic 197687

This project wants more production as all do so the question I ask is why don't we optimise the apps to be more platform specific? The only thing I have seen is the option to run multiple work units on GPUs.
An example of what I am talking about is we don't use CUDA42, or CUDA50 only CUDA32, another is as far as I can tell we don't differentiate between AMD and Intel CPUs much less the various flavors of them. I know from my years at SETI that can make a huge diffence in production. The latest release from Lunatics has increased my out put over 20%. That is huge!

paul milton
paul milton
Joined: 16 Sep 05
Posts: 329
Credit: 35,825,044
RAC: 0

Why not?

they do, the application list can be found at http://einstein.phys.uwm.edu/apps.php i may not be explaining it right but if i recall the app detects instruction set and uses which ever is best, i "think" they tried x86 vs x64 before and there wasnt much of an improvement. though i do see some x64 (EM64T) applications.

seeing without seeing is something the blind learn to do, and seeing beyond vision can be a gift.

Holmis
Joined: 4 Jan 05
Posts: 1,118
Credit: 1,055,935,564
RAC: 0

If I remember correctly all

If I remember correctly all CPU apps has up to SSE2 optimization and above that tests have shown that it doesn't really speed things up so no point.

As for different CUDA version, all versions above 3.2 have caused errors or invalid tasks, there were talk about 5.5 or 6.0 a while ago but I don't think it led to much improvement in speed so again no point.

One also needs to take into account the speed improvement vs the burden to build and maintain all the apps, if there were apps for all SSE levels times 2 or 3 for different platforms and all CUDA levels then the number of apps would be much higher. I'd imagine it would be close to a nightmare for Bernd to keep control of it.

Betreger
Betreger
Joined: 25 Feb 05
Posts: 991
Credit: 1,550,265,188
RAC: 714,761

RE: I'd imagine it would

Quote:
I'd imagine it would be close to a nightmare for Bernd to keep control of it.


Well we need more manpower after all this seems to be a well funded project unlike S@H for example.

Alex
Alex
Joined: 1 Mar 05
Posts: 451
Credit: 503,281,640
RAC: 10,229
Betreger
Betreger
Joined: 25 Feb 05
Posts: 991
Credit: 1,550,265,188
RAC: 714,761

Well that reinforces my

Well that reinforces my orriginal statement.

Alex
Alex
Joined: 1 Mar 05
Posts: 451
Credit: 503,281,640
RAC: 10,229

RE: Well that reinforces my

Quote:
Well that reinforces my orriginal statement.

If you follow the post before these two ones you will find that the dev's found a bug in the cuda software package which was confirmed but not solved for a long time. I assume it was a more idealistic than realistic message. Now we have cuda 6 available. Maybe it makes more sense to try this package than to run an unstable 5 or 5.5 version.
I'm shure that the dev's did not forget that. But an information update would be nice.

Betreger
Betreger
Joined: 25 Feb 05
Posts: 991
Credit: 1,550,265,188
RAC: 714,761

Both 42 abd 50 qre being used

Both 42 abd 50 qre being used bug free over at Seti so either the bug was fixed or it does not matter there.

Alex
Alex
Joined: 1 Mar 05
Posts: 451
Credit: 503,281,640
RAC: 10,229

It looks like it matters

It looks like it matters here.
http://einsteinathome.org/node/197131&nowrap=true#126563
Maybe they found another bug then. But as long as we have no further info ...

Betreger
Betreger
Joined: 25 Feb 05
Posts: 991
Credit: 1,550,265,188
RAC: 714,761

And nearly a year later

And nearly a year later nothing for the unwashed masses.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.