Which GPU should I choose for parallel computing(RTX a4500, 4000ada, 4090,4080, 3090Ti,3080Ti...)?

Hi,

I'll be buying a laptop for parallel computing with matlab. Specifically, for solving partial differential equation with space discretization and time evolution.

According to my budget, there are several NVIDIA graphics cards can be chosen, like

NVIDIA RTX: a5000, 5000ada, a4500, 4000ada...

Geforce RTX: 4090, 4080, 3090Ti, 3080Ti...

The Geforce cards are designed mainly for games. I'm not sure how is their performance on numerical computing. So can you rank the above GPUs about the performance on numerical computing?

Or any suggestions are grateful.

Thank you!

Answer 1

[Christopher McCausland]

Hi Wang,

There are a couple of things to note here, firstly the preformance from a laptop card will never be as good as a desktop card. There are just to many limiting factors such as heat dissapation and power requirements.

Secondly, if you are planning on using parallel compute, (and from the sounds ot it using this for research/ University work) I would enquire if the institution has large cluster compute resources that you can send batch jobs too as this can offload a lot of local compute time and keep you more productive.

There are several similar questions on the forurm, most will point to the useful GPUBench here. Have a look though some of the reviews and see if you can find the cards you are looking at.

In my personal experiacnce within both MATLAB and python, looking at mathematical transforms and deep learning, I typically find that GeForce cards preform better as they have a higher clock, however are more power hungry and less reliable (driver updates etc.). That being said, this may not hold true for your workloads. However, more vRAM and higher clock speeds are usually a good thing, especially looking to the future.

Kind regards,

Christopher