Average runtime listing

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0
Topic 198435

I am interested in how well my Haswell does with the AVX work units on O1AS20-100T (among other things). Several projects list average, minimum and maximum runtimes according to the type of work units and hardware (Intel v. AMD, or Nvidia v. AMD, for example), as well as OS (Windows v. Linux, etc.). Can this be done for Einstein?

noderaser
noderaser
Joined: 9 Feb 05
Posts: 50
Credit: 91057840
RAC: 637057

Average runtime listing

Have you checked http://wuprop.boinc-af.org/ for the kind of info you're looking for?

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

I have participated in

I have participated in WUProp@Home in the past, but seldom found its data to very complete. It seems to vary all over the place, probably based on how the computer is used (whether 24/7 or intermittent) and the load on the CPU cores, etc. The websites that give actual runtimes for their projects appear to be much better.

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513211304
RAC: 0

RE: I am interested in how

Quote:
I am interested in how well my Haswell does with the AVX work units on O1AS20-100T (among other things).

This is a really common desire "how does my host compare?".

Perhaps a simple light weight work around might be to put some filters on the top hosts to find matching hosts. Filter on CPU / GPU / Operating system strings and explore tasks from there.

There is a boinc RPC suggestion here which if existed could be used to extract the tasks but no join to the processor or other host details.

The downside of all these is the load it puts on project servers to produce the numbers.

Edit: there is the cpu list

archae86
archae86
Joined: 6 Dec 05
Posts: 3157
Credit: 7230418153
RAC: 1161094

Perhaps you might best try to

Perhaps you might best try to interest one of the BOINC-wide statistics sites to prepare the data the way you would like. An upside would be that if you succeeded, the way you like would be available for lots of projects, not just this one.

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

Perhaps it would be overkill.

Perhaps it would be overkill. I will be getting a comparison shortly, at least for AVX, and I ran one of the SSE2 O1AS20-100T work units a few days ago (10 hours 17 minutes). So I can compare them directly on my own equipment, which is probably the best-controlled environment and the most relevant comparison anyway.

However, this is on Win 7 64-bit, and I would still be interested in how Linux does. Perhaps one of the moderators could start a thread, or at least a few of the Linux users give their results? My completed SSE2 work unit took 10 hours 17 minutes running 4 cores of an i7-4771 on Einstein, but I am only 17% complete on the AVX version and don't think the estimate is very reliable yet, but shows as 9 hours 36 minutes. I think it will speed up, looking at a second one that come in later.

Christian Beer
Christian Beer
Joined: 9 Feb 05
Posts: 595
Credit: 188580413
RAC: 163754

I think I know what list you

I think I know what list you mean but can you please give me an example from another project?

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

About the closest to what I

About the closest to what I was hoping for is from GPUGrid:
http://www.gpugrid.net/server_status.php
They list minimum, maximum and average run times per "application", though that term covers a lot of ground. And they don't distinguish between Windows and Linux. However, they use only Nvidia GPUs, and the difference between Windows and Linux is a relatively constant amount, about 20% as I recall.

Universe gives a finer-grained breakdown per application, though there is a relatively large spread between Linux and Windows there (about 80% the last time I checked) which they don't distinguish in the statistics http://universeathome.pl/universe/server_status.php.

Also, POEM gives minimums, maximums and averages, though they don't distinguish between the type of GPU used or the OS http://boinc.fzk.de/poem/server_status.php. However, POEM has only three applications, and they all have relatively similar runtimes on a given type of GPU, so they don't really need to show the applications separately.

So I suppose there is no single ideal breakdown, but that should give some idea.

By the way, I have completed three AVX work units by now and get 9:09, 9:19 and 9:22, versus the 10:17 on the SSE2 work unit, so the AVX helps, but I expect Linux v. Windows might make a bigger difference.

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

I don't want to cause you

I don't want to cause you unnecessarily work just for my sake. I can get pretty much what I want (for the present case) just by looking at the validation results, and the machines I am being validated against. Fortunately, they include Linux machines (I know of one project where Windows have to be validated only against other Windows machines).

I am the 11368189 machine here, and it looks good.
https://einsteinathome.org/workunit/239442405

The idea may have validity for various other purposes, but I think other people should express there interest in it before you put yourself out too much.

noderaser
noderaser
Joined: 9 Feb 05
Posts: 50
Credit: 91057840
RAC: 637057

RE: I have participated in

Quote:
I have participated in WUProp@Home in the past, but seldom found its data to very complete.


Only as complete as the data volunteered to it...

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

I ran it for several months

I ran it for several months on 5 of my machines and didn't see the results in the statistics. But then I suspected that it contributed to a crash or two, and got rid of it.

Comment viewing options

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