1

In our institution we have two RTX 3080 GPUs and we want to find out, if it is better to have every GPU in separate PC or if it'd be better to have both GPUs in one PC. Would it be possible to e.g. run one analysis on 1st GPU and use 2nd GPU for training DL model (both GPUs connected to one MB) or maybe parallelise DL model training to both GPUs to speed it up? (although what I have red it is quite challenging to program DL model to make it work).

I just want to know, if it'd be more cost-efficient to place both GPUs to one PC (so we'd buy one PC case, one MB and one power supply) or if it'd be better to separate both GPUs to two PCs.

Thank you :)

Improve this question
1
  • It's almost certainly going to cost less to buy two GPUs and build something around that, than an individual PC and all the associated parts for each. Commented Sep 30, 2022 at 1:58

1 Answer 1

Reset to default
0

having mentioned RTX, as in Nvidia, assuming it's CUDA code then yes there is the programming capability to choose which physical GPU among many that are present in a single system. But if you already have that code, and you are dependent on whomever wrote the code, then said deep learning code may not have been written with multiple physical GPGPU's in mind.

Regarding cost and multiple GPGPU's or GPU's (graphics cards) the cost of however many of those is the baseline.... to fit many into one server or workstation I believe will be more expensive than buying a $100 motherboard off newegg having one PCIe 4.0 x16 slot to accomodate just one graphics card. To fit two RTX3080 size graphics card into one PC type workstation you are generally looking at eATX size motherboards, a larger case, power supply and all associated stuff, such a mobo will run $500+ I believe.

And then there's servers: such as Supermicro 4U gpu models, and Dell r940xa, having up to 10 pcie slots that could accomodate 4+ double width graphics cards like the RTX3080... you're looking at BMW 5 or 7 series car price range.

Improve this answer
1
  • Thanks for answer, but we have already decided to separate both GPUs to separate PCs, so we'll buy everything twice. We have some institutional restrictions and simply we are not able to plug both GPUs to one PC. Commented Oct 5, 2022 at 18:16

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.