SMP 6.34 sometime using fahcore_a4.exe

[Stene]

When i finish a WU from fahcore_a3.exe which is multithreaded and then i get a WU from fahcore_a4.exe. What is the problem? I have added the smp flag.

[bruce]

I do not believe there is a problem. FahCore_a4 is mutithreaded too.

[sswilson]

Can I get some clarification here?

I'm running the V7 front end on two pcs. One is set for SMP -6, GPU, and a uniprocessor client. and one is setup to run a uni and a gpu. All three of the cpu clients are currently running 0xa4 cores including the SMP client. Shouldn't the SMP client be running a different core?

Sorry if there is a quick answer to this somewhere in the forums, I tried several different searches and it kept telling me that words like smp, a4, and core were too common to be used in a search.

Thanks in advance

[7im]

a4 work units are compatible with both -smp 1, and -smp > 1, i.e. single core or multi core.

[sswilson]

So is it normal that I'm getting nothing but a4 cores on my SMP client?

These are both fresh windows installs (busy weekend... ) and I want to ensure that everything is working fine..... and while we're on the topic of uniproc clients... do they now give a bonus? The V7 frontend suggests so.

[7im]

Yes, it is normal for the project to send you the work units that move the project forward in the most orderly manner. If that is a diet of a4s for a day or a month or 4 months, then that is what will happen. Like the weather, it changes often...

[bruce]

The code developed by gromac.org is used in almost all of the active FahCores. The core_78 goes back many years and is only for the uniprocessor.. This branched out into several variations of SMP and GPU and PS3 clients that were adapted to different hardware. Meanwhile, other people contributed other types of advancements to the code and it has gone through several new Gromacs versions (not to be confused with the version of the FahCore that includes it. I'm reasonably sure that core_a4 includes all of the capabilities of _78 and _a2 and _a3 plus it has support for things like improved force-field models. Thus if the scientist wants to use those models -- or simply wants to avoid some bugs that were probably fixed in later versions of Gromacs, it makes sense to start the project with Core_a4 rather than with a core that contains one of the older versions of Gromacs.

To prevent the possible "contamination" of results that were started with Gromacs Version X, those projects will be finished with X, but new projects will likey be started with version Y (and Core_a4 does contain the latest version of Gromacs).

[CBT]

http://folding.typepad.com/news/2012/07 ... oject.html

[AtwaterFS]

I have started getting A4's since Aug 24th, and they run single threaded instead of smp -8... Needless to say, points have taken a beating:

http://folding.extremeoverclocking.com/ ... =&u=450216

Is there something I should be doing to avoid this? I dont feel like upgrading the client - I'm running 6.30 so I think I should be 6.34 per previous link, correct?
You guys should update ur WU assignment logic - thanks

[AtwaterFS]

yea updating to 6.34 core enabled multithreaded performance, so just take previous post as an FYI

[Jesse_V]

http://folding.typepad.com/news/2012/07 ... oject.html

[7im]

http://folding.typepad.com/news/2011/11/not-getting-smp-wus-heres-a-fix.html

[bruce]

What you're seeing here is a gradual migration toward a newer version of Gromacs (Some important features are in FahCore_a4; not in FahCore_78) and a gradual migration to the application of the Quick Return Bonus more broadly. I still fold on a couple of old/slow computers, but they can't really compete with newer/faster hardware as the Pande Group continues to emphasize speed over simple WU-count due to the serial nature of most projects.

I have started getting A4's since Aug 24th, and they run single threaded instead of smp -8... Needless to say, points have taken a beating

What hardware are you seeing this on and which flags do you use with which client?

[AtwaterFS]

Dual Xeon E5420 (Penryn Quad @ 2.5) - smp -8.
Like I said I was at core 6.30 - updating to 6.34 fixed the single thread issue - in fact u can see points going back up already: http://folding.extremeoverclocking.com/ ... =&u=450216

I think I was using 6.30 cause I got used to using drop-in replacement for some reason. Don't want to go to v7 or v8 cause I'm just not interested - 6.X console works fine
Thanks

[bruce]

For a dual-quad with no supported GPUs, 6.34 is a fine place to be. If you decide to add a GPU or if the PG adds some V7-only projects, then you might seriously consider upgrading to V7. The current V7.1.52 is pretty clean but offers no major improvements for experienced folks like you running V6.34 SMP on Windows (except for contemporary GPUs). Development is actively working on some of the remaining oddities/bugs but even that may not matter to you (yet).