Page 1 of 1

Increasing GPU memory allocation and utilization

Posted: Thu Apr 23, 2020 1:00 pm
by ragho
Hi,

We have nodes with 10x RTX8000s that each have 48GB of graphics memory. When running the client with multiple <slot> options, the maximum memory used by FAHClient does not exceed 250MB (i.e. < 1% of total available).

Is there a global config.xml reference and if not, are there attributes that can be passed to <slot>s in such a way to maximize the GPU memory utilization as well as GPU cores utilization?

Thanks,
Ragu

Re: Increasing GPU memory allocation and utilization

Posted: Thu Apr 23, 2020 2:21 pm
by JimboPalmer
Welcome to Folding@Home!

Once you have enough memory to model the entire protein, that is all you need.

The amount of memory used and the GPU utilization achieved are both based on the number of atomic bonds in the Protein. Big proteins are going to be more complex than little proteins.

Over time, F@H will be taking on bigger and bigger proteins. Turing based cards like the RTX will get the biggest while older cards like my Pascal based models get less complex proteins.

Re: Increasing GPU memory allocation and utilization

Posted: Thu Apr 23, 2020 5:34 pm
by Joe_H
The folding FAHCore_2n are mainly using the shader cores to process data, little use of the video RAM is involved. However if you check utilization of the shader cores through monitoring software, that will be in the 90-100% range almost all of the time.