Improve Speed of Matrix Multiplication by Taking Advantage of GPUArray

4 次查看（过去 30 天）
Chimalume Atuchukwu
评论： Edric Ellis 2015-6-19
I actually want to improve the speed of the multiplication of matrix A (m-by-n) and vector x (n-by-1) using GPUArray. Using:
gpuA = gpuArray(A);
gpux = gpuArray(x);
gpuC = gpuA * gpux
C = gather(gpuC)
didn't give me the needed improvement in speed. Any ideas on how I can achieve this?
  5 个评论
显示 3更早的评论
Chimalume Atuchukwu
Chimalume Atuchukwu 2015-6-19
Hi Azzi. Thanks for your contribution. I have edited the question to make it clearer. Any assistance you can offer me will be highly appreciated.
Edric Ellis
Edric Ellis 2015-6-19
How large is A? What GPU device do you have? Also, note that ideally you'd compare timings for only the matrix multiplication portion - you shouldn't time the data transfers too as you shouldn't need to keep transferring data. I would compare:
timeit(@() A * x);
against
gputimeit(@() gpuA * gpux);
On my K20 GPU I see break-even when A is about 1000x1000.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 GPU Computing 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by