Since the last beta I had a very unexpected exchange with an AMD engineer, which led to a bit of research, and long story short: I now have the OpenCL portions of the AMDGPU-PRO drivers deployed on my nodes with AMD GPUs, and this means that those nodes are now OCL 1.2 compliant. Which, in turn, means that OPNG now runs successfully on those nodes (well, only one of the three got WUs, but they're all using the same GPU series, so this result should apply to all of them). 8 WUs completed, all in very reasonable times:

WUs matching 'OPNG' for World Community Grid in past 24 hours: 8
Total CPU time used:  00h 09min 52s
        Min runtime: 00h 00min 57s
        Max runtime: 00h 01min 25s
        Avg runtime: 00h 01min 14s
WUs by quintile:
        <= 00h 01min 02s         1       (12.5%)
        <= 00h 01min 07s         0       (00.0%)
        <= 00h 01min 12s         1       (12.5%)
        <= 00h 01min 17s         2       (25.0%)
        <= 00h 01min 25s         4       (50.0%)

All validated already. No infinite hangs or other weirdness. Woo!

Thanks for the details

Task in question had 30 jobs inside it. Time spent on the job was 0.03/0.04 0000162_00228. Looks like between 2 and 4 seconds were spent on each job within the task

I noticed I had another task that had 18 jobs inside and granted me with 0.3/820.7 with the runtime of 0.03/0.03 Does seem excessively high considering my other task only had 30 jobs? 0030007_00530_0 Looks like around 7 seconds were spent on each job with this task
I have included the task names above for you my Device name is DESKTOP-FVL1L8F If you would like to look into it.

I am sure I talk for more than myself when I say this I/we appreciate. The time you spend on things making sure they work correctly.

Thanks Speedy,

Those credits look in line with what I'm expecting to see. For a CPU work unit, we estimate they can run X jobs based on what each job has inside of it. This is based off how many atoms are in a given ligand.

( 0.0000000122 * Atoms^2 + 0.0000000751 * Atoms + 0.0000105946 ) * ga_num_evals * ga_run = how long we estimate it'll take for an average cpu.

Each job has a different number of atoms and structure, which changes the equation by evals being different and higher generally with more atoms in a ligand. This is 100% just an estimate but gets us a pretty good average runtime on similar processors.

When a work unit is created, we package multiple jobs together or split them up based on how difficult they are. We try to target say 3 hours per CPU work unit. For the GPU version, we create them with 20 times the difficulty as CPU version. These are split the exact same way, thus they get 20 times more points because they were originally created 20 times harder.

If we ran one of the GPU work units on CPU, it would on average take them 60 hours to complete the same task.

This is the basis for why points are granted the way they are for this application. One thing that is different as I have mentioned just above, is that when a job finds a good answer, it does not need to continue and stops early. This is why you are granted a percentage of the total max points allowed.

Thanks,
-Uplinger

Looking good so far my side as bell. Just the same issue repeating for my since the start of the first beta test, that my 1660S always suddenly stops after a certain amount of time. If I dont baby sit these tasks ans suspend/unsuspend them, they run into runtime exceeded errors, while otherwise applying this strategy, sets back runtime by a few minutes and let's them finish within minutes.

No issues on a 970. I'll try a clean driver intstall before going live trying to fix that issue hopefully.

Thanks Keith. Glad to see all seems well. Just one more question. When do you sleep?