A CPU for our GPUs?

Reference the Runpod instance tippi is running.

What matters for our backend at the moment:

• Speed benchmarks:
  • 2 images * 20 steps DPM++ 2M Kerras returned in under 8 seconds
  • 2 images * 50 steps Eular a returned in under 18 seconds
  • 2 images * 150 steps Eular a returned in under 50 seconds
• One at a time, and/or many in parallel
• A single stable URL with 55% uptime for 5 days straight
• Documentation for the endpoint functions
• At least 10 days of access for $200 or less