Are all steps equal within a WU?

[arisu]

On a system without a varying load, would the number of steps completed per second remain relatively constant within a single work unit? Or are the number of operations needed to complete a step more-or-less the same irrespective of the state the simulation is in (for realistic states)?

[muziqaz]

Each WU from any project has 100 frames aka 100%.
Every project might have different number if steps. Step number usually decided before release, depending how quickly that WU runs on test systems.

[arisu]

I know I mean within a given WU. So let's say a WU has 500,000 steps. Is the progress from 100,000 to 100,100 going to be significantly faster or slower than the progress from 400,000 to 400,100 for example?

[muziqaz]

I know I mean within a given WU. So let's say a WU has 500,000 steps. Is the progress from 100,000 to 100,100 going to be significantly faster or slower than the progress from 400,000 to 400,100 for example?

There is no variation in length and size within the same WU

[calxalot]

Seems to me the speed could vary if atoms migrate into a larger partition (with more atoms)

[muziqaz]

Seems to me the speed could vary if atoms migrate into a larger partition (with more atoms)

But that shouldn't happen within same WU. Maybe different runs, but not the same WU

[calxalot]

My understanding is that the space is partitioned three dimensionally and atoms move around during the simulation. What’s to stop them from wiggling into another partition? Simulation is only as fast as the slowest thread.

[Joe_H]

Depends on the project. Some I see almost the same exact TPF for every frame. Other projects the time will vary between frames more than a few seconds. There was one project where I recall where every other frame took about 10-15 seconds longer based on the timestamps I saw in the log file.

But you shouldn't see the time for a 1% frame vary by very large amounts such as a couple minutes for one and 5 minutes for another except if something else is actively using the system. Folding uses low priority so other processes will run first, folding will use the remaining CPU cycles.

[arisu]

My understanding is that the space is partitioned three dimensionally and atoms move around during the simulation. What’s to stop them from wiggling into another partition? Simulation is only as fast as the slowest thread.

I'm sure that it is possible to put the simulation into a pathological state but on average I would expect atoms to be relatively evenly distributed.

[Joe_H]

The atoms may be distributed evenly, but they are not all the same. The simulation is of a protein system in a solvent, the bonds between elements of the protein will be different from those of the solvent. They will also differ between the solvent and the adjacent protein.