Matlab GPU processing: Nvidia Quadro k2200 vs Geforce GTX 970

Hello. I'm going to buy a desktop for numerical calculation of huge matrix and iterate them.

I wonder which one is better for the purposed of numerical calculation: Nvidia Quadro k2200 VS Geforce 980. I use MATLAB 2016a and Parallel computing toolbox, Windows 7 professional 64bit.

Could you please let me know which one is better or more stable? Many thanks in advance.

Answer 1

[Joss Knight]

I cannot speak from experience. The Quadro series are intended for high-end visualisation, and GeForce for high speed graphics (gaming, mainly). They differ in their power requirements and memory architecture as well as their raw FLOPS. The 980 is a very popular card for training deep neural networks, since these do not require high precision and are good value for money. The general idea is that the Quadro cards will be better at double precision computation but that isn't necessarily the case. If your use case is more to do with accurate numerical analysis then you should be looking to buy a card from the Tesla series.

You might find this Wikipedia article listing the specs of all NVIDIA's GPUs helpful.

The only thing I can say from experience is that we have had a lot of reports of people using the GeForce 980 and finding it works for them, it has blistering single precision performance, and double precision still often beats the CPU even though it's 32 times slower. I've never heard of anyone using the Quadro K2200 so I cannot comment.