Do you have any plans to support GPU

May 4, 2016 14.5k views

Hi - we are starting to require GPU support for our next release. Any plans from your side to support it in the very near future?

1 comment
11 Answers

I'm also very interested in this question. Taking into account the popularity of ML and Neural Networks, that are currently use GPU-based computation, this topic will be very actual.

My company can chime in here as well. We need to constantly train and evaluate Deep Neural Networks on 20-50 GPUs in parallel.

  • Not sure if you were able to figure out the way out for your requirement? However we have been using something on cloud instances powered by GPU's at reasonable rates here on - check it out, it may help.

I think you'd need dedicated hardware for that ...

  • Nope, I helped set up our office's dev/testing server which runs VM instances much like DO does. However, our server has 4 CUDA compatible GPUs where the VM instances have access to the GPUs inside. If you're interested in how this works, look up IOMMU, and KVM/QEMU or KVM/LibVirt are two common ways of making VMs with nearly native access to GPUs and other PCI bus devices by using IOMMU compatible hardware.

No need for something extreme but just enough above the current one in the droplets.

I would use the hell out of this if it were anywhere near as affordable as AWS

+1 (also for Deep Learning purposes)

Digital ocean being my #1 goto place to spin up an instance when I quickly want to learn something new, the skillsets that seem to be a requirement for future survival in the industry lean towards knowledge of Machine Learning. Digital Ocean being high on the convenience/cost ratio for me, I think the lack of a GPU/TPU droplet is adding to my research on other platforms. I still have high hopes that DO will get GPU support.

Have another answer? Share your knowledge.