For all I remember, not interested in any extended debate, HCMD was launched when the UD agent still had the fastest share of WCG CPU time. Considering the anticipated limited CPU years needed, the decision not to port it to the BOINC platform was made then. History.

P3/Athlon not supported... refer to post by I think uplinger early in this thread. If this is changed, I don't know. Any written thought on later change is speculation. It's a matter for WCG to decide to revert to the very lowest common denominator. If it impacts performance, I would not. WCG techs will know from the error reports how many devices are causing error in that class. A similar decision to ending for W98 was made, else WCG/We'd be forever stuck with legacy. So, let's see what WCG decides.

You've got to be kidding. As things stand, HCMD2 WUs are being given to P3 machines, which may error immediately, fetch another, and repeat. If specifically set to HCMD2, they'll hit their daily quota, then sit idle for the rest of the day, then do the same the next day (with the daily quota halved). Result: Lots of idle machines, huge numbers of error WUs, massive amounts of useless work for for the scheduler, lots of pointless downloading (some of us pay for downloads, you know). Plus, we are very limited in number of boinc profiles. I'm sure I'm not the only one running the same profile for some pre-P4 and some P4+ machines. Having to divide up all profiles by CPU type, so as to avoid P3 machines fetching P4+ WUs becomes really messy.

Getting some tiny performance increase on some machines while creating this huge mess can't possibly be worth it. At the very least, the scheduler needs to recognise pre-P4 machines and not send them P4+ WUs. If it were to do that, then the obvious next step is not to ignore pre-P4 machines, but to have the best of both worlds: Simply have 2 copies of the WU executable, one with the P4 optimisations and one without, and send them to the appropriate machines.

This is pretty obviously just a screw-up, as P3/Athlon machines are apparently working fine under Windows. What's amazing is that there's still not even a mention of this in the Known Issues forum.

Thanks Uplinger... very much appreciate your time to respond and clarify for me. I have now used another profile for these machines and set them to not run HCMD2

What is so different about HCMD2 that it requires using a different compiler?

Also, making it work across all x86 processors might be unnecessary. There probably aren't many P2 and older machines left. How is performance if you do P3 optimisations, but not P4?

Hello again! Let me try to respond to all the posts since my last.

Matt, no problem on explaining the issue.

TXR13, The reason it is limited to HCMD2 linux for older processors is that we use a different compiler for this application. Most of our science applications are programmed in C/C++ but some are in Fortran. Which caused the different compiler. Since without optimizations the work units were taking forever to run we had to add some optimizations causing this issue. We are going to try our best to make it work on all processors in linux. This is our goal.

Dadinck, Each science application uses the BOINC api to have it communicate with the client/manager. But each work unit is specifically run on the science application which is downloaded through BOINC onto your machine. The problem people are seeing is that on older machines (I believe it is processors with out SSE2) that are running linux they error out immediately due to an invalid operation being passed through the processor.
The best example I can give on this is basically someone gave you a book to read English, and in the middle of the book a quote was written in Spanish. Since I am not fluent in Spanish, I would stop reading and close the book as I can not continue the story properly. But other members who know Spanish would be able to continue the story properly to then end.

Hope this helps,
-Uplinger