you are viewing a single comment's thread
view the rest of the comments

The cheapest way to run larger models locally is the Vega architecture Radeon Instinct MI60. 32gb of HBM for 3-500 usd used, add 30-40 per unit for active cooling.

  • source
  • parent