Digital Ocean recently launched a NVIDIA H100 GPU cluster which provides on-demand (pay-by-the-hour) H100 cloud GPUs backed by a 3.2TB GPU fabric. The cluster is optimal for bursty training/fine-tuning jobs, no commit required.

Server specs (8-way box) 8 NVIDIA H100 HGX GPUs connected by NVLink/NVSwitch 1.76 TB RAM 128 vCPU

DigitalOcean now has H100 GPU clusters available with 3.2TB networking

https://twitter.com/digitalocean/status/1747974629572911140

That's interesting because I block all publicly facing ips from DigitalOcean from every server I own. the amount of spam and hackers probing from there ips is beyond a joke. Anyway thanks for your input.

I'd be curious to know how quickly they fill up capacity for customers on whatever size DO has deployed? (how hungry is the market for this offering?)