Using Tesla K80 as two Tesla K40

On this page

https://blogs.nvidia.com/blog/2014/11/18/tesla-k80-perf/

it is mentioned that Tesla K80 can be used as two separate Tesla K40.

When I execute “nvidia-smi”, it shows only the K80 GPU. Is there a way to make it show two K40 instead? or how can use this GPU as two K40 in CUDA?

I looked at that page. There is only one mention of K40, and it is in this quote:

“Xcelerit has tested the new K80 and found that for many of our financial customers, it can achieve a straight 2x speed-up over its predecessor, the K40. "

I don’t see anywhere that it is mentioned that “Tesla K80 can be used as two separate Tesla K40”.

There is no way to make nvidia-smi show a K80 as two K40 instead.

The K80 can be used as two separate GPUs, exactly the same way you would use two K40 GPUs.

Codewise, there is no difference.

cudaSetDevice(0);
// launch work for one of the K80 devices (or one K40)
cudaSetDevice(1);
// launch work for the other K80 device   (or the other K40)

If you look at the output of either nvidia-smi or the deviceQuery sample app, you will see two entries for each K80 GPU in your system, because there are two physical devices on a K80 GPU.

Here is the statement:
“… the K80 gives the user flexibility to either use both GPUs on the board to maximize throughput for a single simulation, or use them independently to maximize aggregate simulation…”

cudaSetDevide(0); // this is for the first K80 
cudaSetDevide(1); // this is for the second K80

Yes, the K80 has two GPU on board (I said: “The K80 can be used as two separate GPUs,”), and you can use them together or independently.

They are not K40’s, however, and there is no connection with a K40.

If you have a single K80 GPU in your system, and you look at the output of either nvidia-smi or deviceQuery, you should see 2 devices reported.

Thank you I understand now.

If nvidia-smi gives me a list of four K80, it means I have two of them and I can use them as four.

Yes.