once the tasks don't utilize the GPU to full extent any more (which happens either on purpose or inadvertently), each tasks takes markedly longer to get finished, thus reducing the network traffic with all the jams we had before.

I'm experiencing some strange behaviour after modifying the app_config file.

I forced BOINC to run up to 8 GPU workunits in parallel:



This works absolutely fine. I run both GPU and CPU workunits and my GPU and CPU are able to process that many in parallel. This obviously has a dramatic effect on throughput.

However the BOINC client is not able to fetch GPU workunits anymore. It tries to fetch both CPU and GPU workunits but only receives CPU workunits. Anybody who experienced the same?

Just for reference. I am running 7 WUs concurrently on my 1660S and 6 WUs on a 970 card. Works like a charm, with the downside of my 8 core CPU actually being fully pegged. The CPU is running at 90-95% load at all times and just keeps on feeding the GPUs. Boincstats recorded ~1.4m points just for these cards alone for the last 24hrs.

And about GPU util. While throughput could eventually be tweaked higher by reducing the number of parellel WUs leading to shorter runtimes and lower memory bandwidth restriction, I reckon that a higher and more continious GPU CUDA load effectively yield an increase in power efficiency. Both cards run at ~100W each. Not too shabby IMO.

Curiously enough, ~1.4m credits is pretty much equivalent to what the two cards can achieve per day over at F@H.

Both, and that's also the intention: I want to process both GPU and CPU work.

In the mean time I was able to catch some GPU workunits but not nearly enough to keep my GPU busy full time.

As a WCG newbie, I'd just like to say thank you very much to KoopaTroopa and others whose comments have enabled me to get my machine crunching 6 GPU WUs at a time instead of the 1 which it was doing yesterday. Cheers!