Hi all,
I have been folding for spcr for a few weeks with my humble 1.3 Athlon. When I started, it gave me a 854 gromac, which takes ages on my machine. Ok, more accurately it takes 7-8 days. These gave 155 points in about a week. Which I gather is really low...
Last week, when I had done about 2400/2500 of the same gromac, I had a power outage. When the electricity was back, it started not from ~2400 but from 0! I had the checkpointing frequency at 3 min! I don't understand why this happened? After that the same thing happened when I was at ~1000.
So, I got really angry and changed over to no deadline units. I thought this would give me less points but maybe smaller WUs. Guess what, yesterday and the day before I got two tinkers that were 49 and 42 points and finished in less than a day. Now that makes 45*7=315 PPW.
With gromacs: 155 PPW
With tinkers: ~300 PPW
I just don't understand this? Aren't the deadlineless units supposed to be less points? Any advice?
Gromacs or Tinkers?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
-
- Patron of SPCR
- Posts: 2674
- Joined: Sun Jul 04, 2004 6:07 am
- Location: Houten, The Netherlands, Europe
Yes, you get about twice as much points on Gromacs units if your processor can use SSE in stead of no optimisations. The difference between using 3Dnow and using SSE is less dramatic but still very noticable.
What happened to your Gromacs units is a known problem. When F@H writes a checkpoint it overwrites the previous one. When your system shuts down while it is writing a checkpoint, the checkpoint gets corrupted and the work has to start from the beginning. On the Folding-community forums it has been argued that this could be solved by using alternating checkpoints. But the F@H programmers have other things higher on their todo list.
If you set your checkpoint interval to 3 minutes the chance that the system is writing a checkpoint at the moment it shuts down is higher than with a checkpoint interval of 30 minutes.
What happened to your Gromacs units is a known problem. When F@H writes a checkpoint it overwrites the previous one. When your system shuts down while it is writing a checkpoint, the checkpoint gets corrupted and the work has to start from the beginning. On the Folding-community forums it has been argued that this could be solved by using alternating checkpoints. But the F@H programmers have other things higher on their todo list.
If you set your checkpoint interval to 3 minutes the chance that the system is writing a checkpoint at the moment it shuts down is higher than with a checkpoint interval of 30 minutes.