GPU not boosting

[silverpulser]

Running f@h on Windows 10 64bit Pro and latest F@H software. GPU is 750 Ti.
For the last two days GPU-Z is showing my GPU clock as 1019 Mhz where as it normally runs at 1162 Mhz; i.e. no boost. The GPU has been running normally 24/7 for about 4 years now. All temps look low so I am wondering what could be causing this.

[foldy]

Some work units don't trigger boost. No boost is more power efficient. If you still want to boost nevertheless then run any 3D app in parallel, e.g. FAH Viewer.

[silverpulser]

Ok, thanks for the reply. I did wonder whether it was the units. Certainly makes for an easier life for the GPU and fans. Temps are down to around 40C.

[MeeLee]

usually Core 22 would do the opposite to high end GPUs.
I guess the core isn't optimized for older GPUs?
I'm thinking, if the GPU runs at PCIE 1.0 8x or PCIE 2.0 x4 or slower, it may also be PCIE bandwidth starved.

[bruce]

FAHCore_22 is not optimized for some GPUs at the expense of other GPUs. The drivers make whatever decisions are appropriate for specific GPU models. Blame the chip manufacturer and it's driver.

[foldy]

Would be nice to have that option in FahCore if user wants a work unit to to max GPU usage with boost for best performance or lower GPU usage without boost for better UI responsiveness and power efficency. Maybe OpenMM framework could trigger both behaviors depending on OpenCL allocation? Would be a feature

[bruce]

Write an enhancement suggestion in https://github.com/FoldingAtHome/fah-issues. Be sure to explain the issue clearly.

One of these days (soon) FAH's code is all going to be open-source and there will be enough programmers looking for something they can improve.

[MeeLee]

Sounds great!
A program where you can assign the amount of cores on your GPU.
The lower the amount of simultaneous threads, the lower the power consumption (but also the lower the ppd).

[bruce]

I'm not sure that makes sense.

Every application that uses the GPU generates blocks of work called kernels and submits them to the GPU. The application knows the characteristics of the GPU and generates kernels that will run within those constraints. Whether the kernel comes from FAHCore_2x or from the OS (to update the desktop) or from youtube (to process a video) it's up to the GPU to schedule them internally.

UI responsiveness depends on whether there are free resources at the time the OS submits a UI update kernel as well as whether it has to wait for a free moment on the PICe bus.

[MeeLee]

I see...
Another alternative, is to issue a small delay between data feeding from the CPU to GPU, causing the GPU to use a little less power.
It's kind of counter intuitive to slow down a process in hopes to lower TDP.
I wished FAH could have more direct access to the power settings from the driver.

[PantherX]

...is to issue a small delay between data feeding from the CPU to GPU, causing the GPU to use a little less power....

That would cause thermal stress on the GPU since you would be cycling from high GPU load to low GPU load. Thermal stress isn't ideal for the longevity of the system

...I wished FAH could have more direct access to the power settings from the driver.

Once the GPU Vendors build a driver with a scheduling system and the OS supports it, it would then be up to F@H to use those features. Currently, there's no GPU scheduler so it's a bit of a hit-and-miss.