GPU Cloud Options In 2020

What are your options, when it comes to running GPU workloads with a cloud provider?

Here are the results of my initial research. I hope it will help you to discover other options apart from the usual big cloud providers. With more choice, you'll be able to decide for the cloud GPU option which best suits your to-be-deployed AI application.

Setting Expectations

Getting the best bang for the buck is not the only criteria.

This is the result of first, cursory research I did. It doesn't take into account legal requirements, or more the availability of more advanced cloud services apart from basic GPU compute, which might be useful for your AI applciation, for your development workflows or similar synergies.

It also doesn't take into account available cloud credit offers, which might influence the choice significantly.

Source: Mind Share

Personally, I would have listed the usual suspects:

• Amazon Web Services (AWS)
• Google Cloud Platform (GCP)
• Microsoft Azure (Azure)

However, that's by for not the complete list!

Source: NVIDIA

NVIDIA lists a nice overview of their "Cloud Partners" on their GPU Cloud Computing page.

The list from above grows by a lot:

• Alibaba Cloud
• Baidu Cloud
• IBM Cloud
• Oracle Cloud
• Tencent Cloud

In addition, you can search for partners using NVIDIA's "Partner Locator", although I didn't find it very informative.

Source: Google Search

Instead, looking for "GPU cloud" via a simple Google search yields even more results, not emphasized by NVIDIA's partner page.

Here's a list of a few promising-looking entries which caught my eye, because I recognized the companies from other cloud-related topics:

• Lambdalabs
• Linode
• OVHcloud

I hadn't heard about these before, and have no experience whatsover with those, but they showed up pretty high in the search:

• PlusServer
• GPUEater
• GPUonCloud

Additions

Since writing this article, here are some more options I have stumbled upon:

• Hetzner doesn't seem to offer new machines with GPUs, but you can search the auction. Those are root servers, and you pay per month instead of cloud-usual pricing units.
• HostAG
• GenesisCloud
• Paperspace
• Scaleway
• Nimbix

In Conclusion

There are a bunch of options, but relatively few big ones.

Sure, you can run your GPU workloads with a smaller cloud provider, but the choice should be made carefully. It's unlikely that Azure will close down overnight, but it's less unlikely that a smaller company won't be around anymore eventually for a number of possible reasons.

I hope this small list has helped you to find more GPU cloud hosting options.

All the best for choosing the right GPU cloud to run your deep learning workloads and deploy your AI products!