H200 vs H100
Same Hopper silicon, 141 GB of memory instead of 80. The published compute figures are identical; what you are paying for is capacity, bandwidth, and fewer cards per model.
- The H200 SXM5 141GB holds 141 GB to the H100 SXM5 80GB's 80 GB — 1.76× the capacity, which is the difference between a model fitting on one card and needing two.
- Identical dense BF16 throughput — 989.5 TFLOPS on both. The die is the same; what changed is the memory around it, at 1.43× the bandwidth.
- It rents for 1.73× the price cheapest-to-cheapest, across the 18 providers listing both.
- We have not measured either card on any workload. Everything above is the vendors' own peak figures beside today's published rates.
Violet marks the stronger figure. Throughput numbers are the vendors' own published dense peaks, read from H200 SXM5 141GB's product page and H100 SXM5 80GB's product page — not speeds we measured. Process node and transistor count are absent because they are not on the pages we read; architecture comes from our hand-compiled hardware registry, which has no fetch behind it.
On-demand, per GPU, USD, from providers publishing a comparable rate for both cards. 11 more list one of the two but not both. Spot and committed rates are excluded — they are not comparable to on-demand, and mixing them would turn a different product into what looks like a discount.
Same die, more memory around it
- Dense BF16 is identical at 989.5 TFLOPS, so arithmetic-bound work has nothing to gain. What the H200 SXM5 141GB adds is 141 GB against 80 GB, and 1.43× the bandwidth to feed it.
- That pays where the H100 SXM5 80GB is memory-bound: a model that needed two of them fitting on one, longer contexts, more concurrent sequences per card. Where it is not, the two run the same job at different prices.
Where you rent moves more money than which card you pick
- The same H100 SXM5 80GB is published between $1.30 and $12.84 — a 9.9× spread across 24 providers for identical hardware.
- The gap between the two cards is 1.73×. Choosing the provider is worth more than choosing the card, and it is the part a spec sheet cannot tell you.
More pairings need both cards' specs fetched first.