B200 vs H100
Blackwell against Hopper: more than twice the memory and bandwidth, and an FP4 path the H100 does not have. What the published figures say, and what providers are charging for each today.
- The B200 SXM6 180GB holds 180 GB to the H100 SXM5 80GB's 80 GB — 2.25× the capacity, which is the difference between a model fitting on one card and needing two.
- The B200 SXM6 180GB publishes 2.4× the memory bandwidth but only 2.3× the dense BF16 throughput — the bigger gain is in moving weights, not in arithmetic.
- It rents for 3.46× the price cheapest-to-cheapest, across the 12 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 B200 SXM6 180GB's product page and H100 SXM5 80GB's product page — not speeds we measured. NVIDIA publishes no per-card page for the B200 SXM6 180GB; its figures are the per-system totals on the DGX page and the HGX B200 table, divided by the 8 GPUs in the system, and only where that divides exactly — memory and bandwidth do, board power does not and is left blank. 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. 16 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.
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 3.46×. 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.