ChasR
09-23-05, 02:21 PM
Today, I discovered an error in a bit of data I've been spouting about at times, QMD checkpoints. A couple of months ago, I noticed frame time varying on QMDs by 46 seconds every 4 frames or so. The only difference I saw at the time was that every fourth frame had a checkpoint in it. I assumed that since the rig I was looking at was folding almost exactly 15:00 per frame that sometimes a checkpoint was triggered and sometimes not. What I later learned was that there aren't the same number of steps per frame in a QMD. They vary from 20 to 22 steps, after convergence, depending on the total number of steps in the QMD. If a frame with 20 steps completes in 15:00 a frame with 21 steps will take 45 seconds longer. On the rig I was looking at, the extra step took the frame past the checkpoint time. The difference which I reported at 46 seconds was actually the extra step plus the checkpoint time.
I recalculated the checkpoint time today and find that one checkpoint per frame adds about 2 seconds per frame on sig rig #2 or 3:20 per WU. In comparison to the 600 pointers, which take about 8-9 seconds, QMD checkpoints are actually pretty quick. The slower the HD, the longer the checkpoint will take so your mileage may vary.
I recalculated the checkpoint time today and find that one checkpoint per frame adds about 2 seconds per frame on sig rig #2 or 3:20 per WU. In comparison to the 600 pointers, which take about 8-9 seconds, QMD checkpoints are actually pretty quick. The slower the HD, the longer the checkpoint will take so your mileage may vary.