Which workstation or system better for matlab parallel procesing(GPU)?

Respected sir/madam,

I wish to buy a workstation which having Quadro gpu, so will mitalb parallel processing like highend image/vidio processing will work out properly?

Kindly guide which workstation is best for latest matlab.

[Walter Roberson]

Why a Quadro ? What GPU features are required for your purposes? Are you going to be working in single precision or double precision? Are you looking to get the new Quadro RTX 6000 ?

[Pranav Shingane]

I am new in image/video procesing on matlab. so, which gpu will best for me. Quadro or tesla?

Please guide which precision will best for parallel procesing(image/video)?

[Walter Roberson]

what kind of calculations will you be doing?

[Pranav Shingane]

Calculation part is small but for image procesing i need fast processing.I will perform image and video procesing using cuda or alexnet. Related to image processing coding, algorithms and mathematical part for it. . Also there will be training part; if i have proper gpu it will perform accurate and fast.

[Walter Roberson]

https://blogs.mathworks.com/deep-learning/2017/10/06/deep-learning-with-matlab-r2017b/ hints that using GPU Coder product could be important during classification .

Answer 1

[Matt J]

You need to select a workstation whose power supply is enough to meet the power requirements of whichever Quadro card you are considering. You also need to meet the Matlab system requirements described here.

[Pranav Shingane]

Thank you ,

matlab requirments are meeting properly but for image/video processeng in matlab which gpu will best for me? Quadro or tesla? and which model of gpu will be best?

[Matt J]

The stronger the better, right?

[Walter Roberson]

if double precision work needs to be done then you have to choose more carefully . If there is aa need to access graphics by remote desktop software then you need to choose carefully . if you expect compute heavy custom GPU work you should get aa device with tcc support

[Walter Roberson]

Of these factors, the required double precision performance is the trickiest. There are nominally faster GPUs that have much worse double precision performance, that some of the other devices: some of the GPU have specialized double precision hardware. The other things like TCC can be easily researched on a Present / Absent basis, but double precision performance is trickier.

My understanding is that Alexnet does not require double precision, and that it is common for Deep Learning techniques to be computed in single precision or even "half precision". But if you were to create your own algorithms, or use one of the less common Deep Learning techniques, then double precision could be pretty important.