Get started
Choosing a GPU
Match model memory, precision, GPU count, network, and security.
Hard compatibility first
Estimate memory from model parameters, precision, workload overhead, and GPU count. Filter out machines that do not meet VRAM, GPU-count, CUDA, or explicitly requested CVM requirements.
Then choose the operating fit
Compare the final hourly rate, observed reliability, network throughput, location, tier, and whether your template is likely cached. Unknown values remain Not verified.