Hello. We have purchased a workstation with the above configuration which will be used for running code written in C++/CUDA for CNN deep-learning. DO WE NEED to bridge the two GTXs with the ‘GTX SLI HB Bridge’ which we can’t anyway purchase it from anywhere as suppliers say that they can’t deliver, or CUDA will recognize them both on the motherboard and will then utilize them both for taking full advantage of the 7168 cores? Thank you very much. John Piliounis, Athens, Greece
You don’t need the SLI Bridge and it is not helpful for CUDA anyway.
Having said that, there is no way (with or without the SLI bridge) two make 2 GPUs appear as if they were one, to CUDA, or to anything that depends on CUDA.
Thank you very much txbob. So how could one utilize the use of both the GTX GPUs in order to use as many cores as they are available from both GPUs for running his/her code? I don’t want to make them to appear as one and I only need to know and understand where does the utilization of the GPUs need to take place so that both are available for and exposed to my CUDA code?
Again, thank you very much.
You need to learn a lot more about deep learning, if you want to use multiple GPUs for deep learning. For example, using tensorflow, you can use multiple GPUs. It’s also possible with other deep learning frameworks.
When choosing a motherboard, the best choice is to have two dedicated x16 slots (for 2 GPUs), one for each GPU, both connected to the same CPU. That last part is probably irrelevant in your case, since there is only one CPU on that particular ASUS motherboard.