Hi, We are conducting tests on our laptop, and I have observed that when running the program locally, the bottleneck for matrix multiplication shifts from the CPU to the memory. Saying when the size of the matrices exceeds 50000, dynamic memory allocation results in a bad_alloc error. But all of these can be easily done in 10 minutes. In this case, how should we implement our experiment? Thanks for your reply.

This might happen depending on how much memory you have on your laptop. This is perfectly fine -- you'll simply have to explain this observation in your report. You can still talk about how your implementation scales below these values, and then explain that you are running out of memory when exceeding N=50,000.

My current idea is to reduce the time limit from ten minutes to one minute. This way, I can make full use of the experimental data I have obtained. Is that okay?

If you have some one minute experiments from Setonix, you can use other one minute experiments for comparison. But you should try to do longer experiments (i.e. 10 minutes) where you can.