Run Prime95 with GPU

[tait1k27]

I have a Intel(R) Iris(R) Xe Graphics, a NVIDIA GeForce MX330 card and processor 11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz, 2803 Mhz, 4 Core(s), 8 Logical Processor(s).
I have 2 questions.
1) Is run Prime95 with GPU faster than with CPU?
2) How to run Prime95 with GPU?

[Zhangrc]

You can't run Prime95 with GPU, but you can run GPUowl.
See https://www.mersenneforum.org/showth...t=25624&page=3 for latest builds.

[kriesel]

Originally Posted by tait1k27

I have a Intel(R) Iris(R) Xe Graphics, a NVIDIA GeForce MX330 card and processor 11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz, 2803 Mhz, 4 Core(s), 8 Logical Processor(s).
I have 2 questions.
1) Is run Prime95 with GPU faster than with CPU?
2) How to run Prime95 with GPU?

GPUs are not supported for computation in prime95. (Please read the documentation or reference info.)

While some IGPs can be usefully employed for GIMPS if the user is sufficiently patient, the Iris Xe IGP on i7-1165G7 is probably not worth pursuing for GIMPS GPU computing. There were compatibility issues and performance loss issues when I tried it. IGP use in general would also reduce CPU performance in prime95 because the IGP and the CPU cores share a package power limit, and also memory access.
Discrete GPUs are more promising, because

1. separate power budget
2. separate memory and dedicated memory access paths
3. typically a greater total power budget than for an IGP

There's lots of reference info for GIMPS use of GPUs with various other mostly computation-type-specific applications.

There are no benchmarks at https://www.mersenne.ca/mfaktc.php for the MX330 GPU, but it may be worth pursuing. The MX 150 is listed at 143 (TF) GHD/day there.
It's probably more effective to use the MX330 for TF than anything else. That's the case with most NVIDIA GPUs (except some pro-grade Tesla & Quadro such as can be found on cloud computing services like Google Colab).

An MX 330 has 2GB GPU ram (small but usable for effective wavefront P-1), about 5% of the memory bandwidth and only ~1.1% the 64bit FP performance of a Radeon VII. I don't expect much throughput in P-1, PRP or LL from the MX330. (Would you be ok with a wavefront PRP taking weeks or months each?)
There are no benchmarks at https://www.mersenne.ca/cudalucas.php for the MX 330. The MX150 is listed at ~10 (LL) GHD/day there. The more relevant MX150 specs (ram bandwidth generally, FP32 for TF, FP64 for other computation types) are close to that of the MX330 on the TechPowerUp pages.

Give the MX 330 a try in mfaktc TF, and in gpuowl PRP on a relatively quicker DC assignment, and please report some benchmarks for each at the relevant mersenne.ca pages linked above.
Expect a completely new learning curve for each GPU application, and no GUI, command line only.