Can anybody explain benefits of hardware acceleration? [closed]

asked 2017-10-02 07:01:23 -0600

I try to sort out what really graphical processors do and what acceleration ratio they provide. I am interested primarily in matrix algebra such as multiplying a vector by matrix (the result being another vector). Both are large (10000 and 10000x10000 respectively).

edit retag flag offensive reopen merge delete

Closed for the following reason question is off-topic or not relevant by LBerger
close date 2017-10-02 08:07:18.658042

Comments

1

What does this have to do with OpenCV?

Anyway, it depends on the hardware you are using. With serious GPU you can have an important acceleration factor (you can even add several cards to your system for even more performance). Even simple configurations (let's say the dedicated graphics card of a laptop) can be 10x faster than CPU.

The matrix multiplication is typically an operation where you can have important acceleration when using GPUs.

kbarni gravatar imagekbarni ( 2017-10-02 08:05:42 -0600 )edit
1

@ya_ocv_user

My suggestion is to use directly a BLAS library that already do matrix vector multiplication efficiently (and basic linear algebra):

Eduardo gravatar imageEduardo ( 2017-10-02 10:09:42 -0600 )edit

To Eduardo. This is a valuable comment. I use AMD processor which has internal GPU. My question was about what specifically it does. So far as I understand, it has no full scale parallel computing. Can load just 4x4 matrix or so. That is acceleration will be 16 times at best. Right?

ya_ocv_user gravatar imageya_ocv_user ( 2017-10-04 05:47:31 -0600 )edit
1

In my opinion, you will have to benchmark yourself to know which speed-up you could get between BLAS on AMD integrated GPU vs BLAS on AMD CPU. It really depends on the GPU hardware but as it is an integrated GPU, probably the speed-up would be minimal but I can't tell.

Eduardo gravatar imageEduardo ( 2017-10-05 05:06:20 -0600 )edit

There is also:

Eduardo gravatar imageEduardo ( 2017-10-10 12:29:19 -0600 )edit