Greetings,

I recently cobbled together a computer to run Einstein@Home. I originally was running integrated graphics as I didn't have a graphics card for it. I recently obtained one but do not have a VGA to HDMI adapter to run the monitor off the card, so I enabled the integrated graphics in BIOS to run the monitor.

BOINC is currently switching between integrated graphics and the NVIDIA graphics card, something I have never seen before. That in and of itself is fine for me, however, when an E@H task is running that uses the graphics card, at most the card is only running at 50%-60%.

has anyone run into this particular situation before? While using integrated graphics for the monitor, is there a way to get BOINC to run the graphics card at or near 100% for those specific tasks? Or maybe it doesn't utilize the full card when an integrated graphics process is also running?
Thanks for any input!



The specs (nothing OC):

Intel I5 6600K @3.5Ghz

MSI Duke GeForce 1070ti

Gigabyte mATX motherboard (can't remember which one specifically)

32 gigs of RAM

---

END OF LINE

Einstein hardly ever does a modern GPU at full load. The WUs are more made for 1030 GPUs than any more modern ones. Even with 3 to 5 WUs per GPU, I still only get it up to about 75%.

If you have a laptop with an Intel IGP and shared with an Nvidia GPU, you might run into issues.
However, if each GPU runs their own WUs, you can increase GPU usage on Einstein by running 2 or 3 WUs at the same time.

Einstein hardly ever does a modern GPU at full load.

Not true. If you use a modern app you can use 90-95% of the gpu with the latest stock app.
I use 98-100% of the gpu with a custom special sauce app. Best production is 2X or 3X concurrency.

Einstein hardly ever does a modern GPU at full load. The WUs are more made for 1030 GPUs than any more modern ones. Even with 3 to 5 WUs per GPU, I still only get it up to about 75%.

firstly, the above bolded text is not true at all.

second, whenever talking about Einstein GPU tasks, you need to be aware of the difference between different types of tasks and which app version you are running. not all tasks are the same and you can't just say "Einstein" without clarifying the subproject. Einstein has two major sub-projects: Gravitational Wave (O3AS) and Gamma Ray (FGRPB1G).

Gravitational Wave GPU tasks *are* notorious for lower GPU utilization. but it's not because they are "made for 1030 GPUs" or other slow GPUs, it's because they are CPU-bound. if you drop in a faster CPU, you see increased GPU utilization. the GW app does some calculations on the CPU as the project devs weren't able to find a method to do them on the GPU (which might be precision related rather than coding methods or hardware capabilities). if you use a fast GPU with a slow CPU on GW, you will see low GPU utilization, which can be somewhat mitigated with running multiple tasks at a time.

now, onto Gamma Ray tasks. these tasks have ALWAYS had rather high GPU utilization as they are not at all CPU-bound. but even here, you do need to factor different sub-types of tasks for this project.
- LAT4000-series tasks will run 95-100% utilization on even the fastest GPUs with only one task at a time.
- the LAT3000-series tasks use a little less (around 85-90% utilization on fast GPUs), and you can bump that up to 100% by running 2 or 3 at time and get better overall production.


many people running Einstein aren't aware of this differentiation in work. and just attach to the project with default preferences, which will try to send you all app types. because of the older server software and the way BOINC calculates DCF, people who do this will predominately be sent GW work unfortunately.

---