Training and fine-tuning large language models from 7B to 70B+ parameters. The H100 SXM has 80 GB of GPU memory against a 24 GB minimum for this workload.
The GPU's published specification against the workload's stated minimum.
Listing-weighted median across 2 observed H100 SXM listings at Massed Compute. The same GPU is tracked at 19 marketplaces total.
The 5 lowest-priced other marketplaces listing the same GPU, sorted by price ascending.
Other GPUs at Massed Compute that meet LLM Training's minimum GPU memory, sorted by price ascending.
Training and fine-tuning large language models from 7B to 70B+ parameters. LLM Training requires at least 24 GB VRAM and benefits from Datacenter-class compute.
Full LLM Training guide and the GPUs that meet its minimumTrack H100 SXM rental pricing at Massed Compute, updated daily — useful for sustained llm training workloads.