Many companies now use GPU instances to train models for their machine learning-powered app or service. Some companies also leverage these cloud instances to generate 3D models and other GPU-intensive tasks. If you don’t want to buy a bunch of expensive GPUs, you can leverage GPUs on demand with your favorite cloud-hosting company. Once you’re done, you can shut down your cloud instance.
Scaleway’s RENDER-S instance is powered by an Nvidia Tesla P100 with 16GB of HBM2 memory. It comes with 45GB of RAM, 400GB of storage (local NVMe SSD, so it should be super-fast for video processing) and 10 cores of an Intel Xeon Gold 6148 with the AVX-512 instruction set. If you plan to keep your instance running for a while, the RENDER-S instance costs €1 per hour or €500 ($567) per month, whichever is lower.
On Google Cloud, you can get an on-demand instance with an Nvidia P100 for $1.60 per hour in Europe and Asia, or $1.46 per hour in the U.S. Microsoft also sells cloud instances with a P100 GPU for $2.07 per hour. It seems like Scaleway is competing with those offers in particular.
Amazon also has GPU instances on Amazon Web Services. You can find instances with an Nvidia Tesla V100 GPU, a more powerful graphics processing unit. Those instances are also more expensive, at more than $3 per hour — prices vary slightly depending on the data center. You also can find AWS instances with older GPUs, but they don’t perform that well.
OVH also offers instances with Tesla V100 GPUs for €2.30 per hour ($2.61). I couldn’t find GPU instances on DigitalOcean or Linode.
Most people probably don’t need GPU instances. But it can be an important factor for companies looking for their next cloud provider. If you want to centralize everything under one bill, you need to pick a company with a large offering.