I know gpuarray could improve for large sized matrix multiplication, but does it also improve if it has very large symbolic elements?

Dear all,

Just as the question says, I know that gpuarray could improve large sized matrix multiplication. However, what if I have a huge symbolic expression for each elements? Would gpuarray still outperform cpu computation?

In another words, does GPGPU perform better than CPU in multiplication of both large matrix and large elements? Or does GPGPU excel in either one of the departments over CPU?

I'm asking this because if GPGPU only excel in one of the two departments, then it is possible that CPU can outperform GPGPU when the other department is significantly large. If so, then use of gpuarray may possibly worsen the performance.

Answer 1

[Walter Roberson]

There is no gpu support for the symbolic toolbox.

[Hayao]

Oh okay. Thank you.

[Walter Roberson]

I see that there is some theoretical work on using GPU for some kinds of symbolic tasks, such as https://www.researchgate.net/publication/221922914_Parallel_Symbolic_Analysis_of_Large_Analog_Circuits_on_GPU_Platforms