H100 SXM vs H200 SXM

H100 SXM vs H200 SXM

Same compute, 76% more memory (141 GB) and 43% more bandwidth. For inference of large or long-context models the H200 is strictly better; for training, whichever is cheaper per hour on the panel.

H200 SXM vs H100 SXM

Identical compute; 76% more memory and 43% more bandwidth. Training jobs that fit in 80 GB see little difference - pick on price. Inference and memory-bound work favor the H200 outright.

Workload fit

H100 SXM: llm pre-training, 70b+ fine-tuning, high-throughput inference. H200 SXM: large-model inference, memory-bound training, moe serving.

Which is cheaper to rent, H100 SXM or H200 SXM?

By today's published on-demand rates: H100 SXM starts at $0.890/GPU/hr (median $5.49, 77 providers) and H200 SXM starts at $0.774/GPU/hr (median $4.59, 58 providers). The cheapest live listing is the H200 SXM at OVHcloud. Figures recompute from the panel on every refresh.

H100 SXM or H200 SXM - which should I rent?

H100 SXM: The H100 SXM is the chip the current generation of frontier models was trained on, and the benchmark every other datacenter GPU is priced against. H200 SXM: The H200 is an H100 with the memory subsystem rebuilt: 141 GB of HBM3e at 4.8 TB/s versus 80 GB at 3.35. The side-by-side spec table and live price rows above give the full picture.

Where can I rent H100 SXM and H200 SXM?

77 providers publish H100 SXM rates and 58 publish H200 SXM rates on Compute Cafe. Every row links to the provider's own published price - committed-use and reserved rates are excluded.