I eventually got seven of these tasks, all resends with _2 or _3 suffixes. Most of the run time was at 3X, though at the beginning and end I was running 1X and 2X for a little while.
I ran them together and turned off the Beta permission to run test applications, ceased requesting these new GW tasks, and resumed running GRP for the moment. All seven of my tasks validated immediately on return.
A couple of observations, which well may not differ at all from the immediate previous GW GPU tasks:
1. There was a very long pause at 99.0% reported completion, amounting to over a quarter of the run time. During this pause there was heavy CPU use and little GPU use. I think I observed pause times at this step of somewhat over 4 minutes elapsed time on my system.
2. Running even a single GRP GPU task while one of these was running in the GPU-rich phase before the 99% pause greatly slowed the rate of advance of the GW task.
With that effect added to the always-present violent changes in DCF and hence pre-fetching, I think most participants won't like to mix these tasks with Einsteing GRP GPU tasks on the same machine at the same time.
I eventually got seven of
)
I eventually got seven of these tasks, all resends with _2 or _3 suffixes. Most of the run time was at 3X, though at the beginning and end I was running 1X and 2X for a little while.
I ran them together and turned off the Beta permission to run test applications, ceased requesting these new GW tasks, and resumed running GRP for the moment. All seven of my tasks validated immediately on return.
A couple of observations, which well may not differ at all from the immediate previous GW GPU tasks:
1. There was a very long pause at 99.0% reported completion, amounting to over a quarter of the run time. During this pause there was heavy CPU use and little GPU use. I think I observed pause times at this step of somewhat over 4 minutes elapsed time on my system.
2. Running even a single GRP GPU task while one of these was running in the GPU-rich phase before the 99% pause greatly slowed the rate of advance of the GW task.
With that effect added to the always-present violent changes in DCF and hence pre-fetching, I think most participants won't like to mix these tasks with Einsteing GRP GPU tasks on the same machine at the same time.