Talking about frames

[PureChaos]

I have read a few threads concerning frames, the time it takes to finish them, and the command line additions some have added to make their machines run faster or more efficiently. Below is a couple of lines from my log file.

03:44:04] Finished a frame (37)
[03:51:28] Finished a frame (38)
[03:58:52] Finished a frame (39)
[04:06:14] Finished a frame (40)


I am currently running a p4 2.53 with 512 ram, is there any reason I would need the sse flag? It seems the two main flags people are talking about are also being associated with AMD systems.

 

[The[H]uman]

[03:28:48] Completed 1450000 out of 2500000 steps (58)
[03:42:48] Writing local files
[03:42:50] Completed 1475000 out of 2500000 steps (59)
[03:56:49] Writing local files
[03:56:51] Completed 1500000 out of 2500000 steps (60)
[04:10:49] Writing local files
[04:10:51] Completed 1525000 out of 2500000 steps (61)
[04:24:49] Writing local files
[04:24:51] Completed 1550000 out of 2500000 steps (62)

Amd xp 2000 - 512mb ram - with both flags enabled

 

[LenBit]

I am currently running a p4 2.53 with 512 ram, is there any reason I would need the sse flag?

No, P4's automatically use sse
The advmethods flag will get you newer "gromacs" proteins which may be worth more points

Len

 

[Chugiak]

If your system works just fine without flags, there is no need to change them. If you are interested in improving the throughput of your system, then you will most likely process more work units in a day or week by using the -advmethods flag.

Without the -advmethods flag your client will probably process mostly proteins that use the Tinker core. This core is the program that does the real work in the folding project, the client program is simply the helper program that takes care of the details.

The Tinker core uses code that was written a few years ago, and provides very consistent and predictable performance from one system to another. More recent code includes the Gromacs core, which generally can process more complex proteins than the Tinker code in a given amount of time. In addition the Gromacs core has optimized code that makes use of the SIMD instruction sets available on newer processors, namely 3DNow!+ and SSE (no handy intel reference, sorry). If you are processing a Gromacs protein, the core will default to running SSE on a P4 or 3DNow!+ on a capable AMD processor. No intervention is required to have the Gromacs core run those optimizations.

However, if the program detects a crash (assumes it was due to system instability) it will revert back to no optimizations the next time it is run. the -forceasm and -forcesse flags override this behavior. Note that the Tinker core has no optimized code, so using these flags will have no effect on the frame time on Tinker proteins. Same holds for the Genome core.

The best way to increase your likelihood of getting Gromacs core proteins is to use the -advmethods flag and also tell the client you have an 'fah' preference in the advanced configuration. if you have any problems with your system crashing, locking up, or otherwise rebooting when not shut down, then you can add the -forcesse flag to ensure that the optimized code is enabled on every Gromacs protein regardless of the prior termination status. however on some AMD systems this leads to more frequent lockups, so the use of -forceasm is recommended to have it use 3DNow!+ optimizations instead of SSE.

Hope this helps!

 

[BillR]

Chugiak is right, Intel cpu should do SSE without flags, and he is right that if your Intel/AMD machine poochs a WU it sets a flag in your folding folder to stop the optimizations. I run both flags for all processors to prevent this. No harm that I can see.

BillR