Page 1 of 1

do we have a new core GROCVS 3.3.99

Posted: Wed Mar 26, 2008 4:12 am
by dschief
Just down-loaded this, it's produceing a log file that is totally different from anything I've previouslly run.
It seems to be running a couple of minutes slower per frame. than the cores on my other machines
It came with a 2619 Wu. Also a first, Any body else getting this combination yet ?

Re: do we have a new core GROCVS 3.3.99

Posted: Wed Mar 26, 2008 4:57 am
by kasson
Yep, this is the A2 core. It is both based on a new version of the Gromacs code and uses a newer version of the FAH interface code (there are a couple cores out there using this latter part). Ultimately, it will give us a lot more flexibility and enable a number of scientific enhancements.

Re: do we have a new core GROCVS 3.3.99

Posted: Wed Mar 26, 2008 12:17 pm
by Ren02
I also got the same p2619 and FahCore_a2 overnight.
I'm running FaH in Linux virtual machines. Got one X2 3800(stock) running single VMWare Player and one Q6600(3.0GHz OC) running two VMWare Players. The VMs had been assigned memory of 512MB. Two out of three crashed with "Not enough memory" error message. I've upped the limit to 640MB, let's see what happens.
Up until now all three had been crunching p2605 and p2653 WUs. The average time on X2 was 30min per frame and on Q6600 12.5min per frame. The X2 is currently crunching p2619 at 27.5 minutes per frame, so basically no change in PPD as p2619 is 1620 points. The situation gets interesting with Q6600. One of the VMs crashed, but the other kept crunching at 11.5min per frame which would also mean no change in PPD. VMWare Player is limited to using 2 CPU cores max, so the other 2 were idle during this time. But when I fired up the second VM , the performance dropped to 17min per frame in both which is pretty drastic. Furthermore, while usually FaH keeps my Q6600 at 50C it is now 4 degrees cooler at 46C. That can only mean that either the FPU or SSE unit is partially idle.
The Project Summary page says p2619 is 308K atoms vs just 77K in p2605 and p2653. I'm guessing that this particular project is memory bound. That's the only logical explanation as to why the frame times increased that much. Seems that there's no point in running 2 VMs on Q6600 for this particular project. Does the VMWare Server support 4 cores? If not then I'll have to try out WinSMP...

Re: do we have a new core GROCVS 3.3.99

Posted: Wed Mar 26, 2008 12:33 pm
by toTOW
For project 2619, you need at least 4 * 200 MB of memory aviable (add what your OS consumes)?

Re: do we have a new core GROCVS 3.3.99

Posted: Wed Mar 26, 2008 5:29 pm
by Lamancha
I seem to have received three of these WU's so far,

User - Crieffvechter

Project: 2619 (Run 2, Clone 361, Gen 0) - [02:38:14] + Results successfully sent
Project: 2619 (Run 0, Clone 436, Gen 0) - [11:49:36] + Results successfully sent
Project: 2619 (Run 4, Clone 369, Gen 1) - [17:23:06] Completed 53130 out of 125000 steps (43)

No sign of points yet though.

Re: do we have a new core GROCVS 3.3.99

Posted: Wed Mar 26, 2008 5:31 pm
by 7im
Nor results listed in the stats DB yet. The new code in the work server probably isn't connected to the stats server just yet, as in the p2605s no credit thread.

Thanks for the report, I'll PM someone.

Re: do we have a new core GROCVS 3.3.99

Posted: Wed Mar 26, 2008 5:48 pm
by kasson
Yes--we upgraded the code on that server, and it took a day or so before the config changes went through to the stats system. Everything after ~9:10 AM PDT today should get credit as usual (tends to post within 2-3 hours of upload completion); I've given a list of the work units returned between the server upgrade and 9:10 this morning to our stats team. Those will take a little longer to credit, but they will post.