[guisalberto]

As you were CPU limited on LambdaLabs, have you tried GPU instances from Latitude.sh? https://www.latitude.sh/accelerate/pricing

Each H100 is paired with 32 3.25GHz CPUs