A few weeks ago I shared an H100/A100 availability post on HN (
https://news.ycombinator.com/item?id=36333321).
But the capacity and prices change regularly!
At first I was manually checking them and updating them on the page linked in the HN post above.
But I didn't want to keep checking manually, and I wanted more data points, historical views, and more real time info. Siddharth who I work with built a small tool to do that.
It's the best way to check live pricing and availability for Nvidia A100 and H100 GPUs. Pretty niche, but still kinda cool.
Why these 3 clouds? I believe they are, for most people, the best places to rent on demand instances of H100s or A100s. (I've looked at quite a few GPU clouds! See: https://gpus.llm-utils.org/alternative-gpu-clouds/)
If you happen to fall into the small niche that's interested in this, please let us know what you like about it!
Would like to know:
1) What's most helpful?
2) What could be improved?
3) Which GPUs, and which clouds, should we add next?
If you need a huge number of A100s/H100s - talk to Oracle, FluidStack, Lambda Labs. Capacity is very low though for large quantities, especially of H100s, based on a couple of cloud founders/execs I've talked with.
If you need a couple A100s: FluidStack or Runpod.
If you need 1x H100: FluidStack or Lambda Labs.
If you need cheap 3090s: Tensordock.
If you need Stable Diffusion inference only: Salad.
If you want to play around with templates / general hobbyist: Runpod.
Assuming you're not tied to / required to use a specific large cloud by your enterprise.