B200 vs H200

The two largest-memory NVIDIA cards a provider will rent you. Closer on capacity than the names suggest; the gap is bandwidth, compute, FP4 — and price.

Illustrative render of a NVIDIA B200 SXM6 180GB: a generic body for its form factor, carrying the model name. Not a product photograph.
B200 SXM6 180GBBlackwell
$4.50/ GPU-hour
cheapest at Sesterce · 23 providers list it
Memory180 GB
Bandwidth8,000 GB/s
Every B200 SXM6 180GB offer →
Illustrative render of a NVIDIA H200 SXM5 141GB: a generic body for its form factor, carrying the model name. Not a product photograph.
H200 SXM5 141GBHopper
$2.25/ GPU-hour
cheapest at Sesterce · 31 providers list it
Memory141 GB
Bandwidth4,800 GB/s
Board power700 W
Every H200 SXM5 141GB offer →
Short answer
What separates them
Memory is the difference
180 vs 141 GB
1.28× the capacity per card — the number that decides how many cards a model needs.
Bandwidth moves with it
1.7× more
8,000 GB/s against 4,800 GB/s, as the vendor publishes them.
Only one has FP4
B200 SXM6 180GB only
9,000 dense TFLOPS at FP4 on the B200 SXM6 180GB; the H200 SXM5 141GB publishes no FP4 figure — the format is not there.
The gap depends on where you rent
1.36–2.00×
Across the 13 providers listing both, that is the range of what the B200 SXM6 180GB costs relative to the H200 SXM5 141GB.
Specs side by side
SpecB200 SXM6 180GBH200 SXM5 141GB
Memory
Capacity GB180141
Bandwidth GB/s8,0004,800
Compute
BF16 dense TFLOPS2,250dense989.5dense
FP8 dense TFLOPS4,500dense1,979dense
FP4 dense TFLOPS9,000dense
TF32 dense TFLOPS1,125dense494.5dense
FP32 TFLOPS7567
FP64 TFLOPS34
INT8 dense TOPS4,500dense1,979dense
Platform
NVLink GB/s1,800900
Board power W700
Max MIG instances7
Silicon
ArchitectureBlackwell 2024Hopper 2024

Violet marks the stronger figure. Throughput numbers are the vendors' own published dense peaks, read from B200 SXM6 180GB's product page and H200 SXM5 141GB'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.

Both cards, by providerPrices confirmed 1h ago
ProviderB200 SXM6 180GBH200 SXM5 141GBRatio
Sestercecheapest$4.50$2.252.00×Deploy
Lium$5.60$3.001.87×Deploy
Cerebrium$6.01$4.201.43×Deploy
Modal$6.25$4.541.38×Deploy
Fal$6.25$4.501.39×Deploy
Vast.ai$6.25$4.611.36×Deploy
Verda$6.68$4.501.48×Deploy
Together AI$8.19$5.991.37×Deploy
CoreWeave$8.60$6.311.36×Deploy

On-demand, per GPU, USD, from providers publishing a comparable rate for both cards. 13 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.

What actually decides it

Where you rent moves more money than which card you pick

  • The same H200 SXM5 141GB is published between $2.25 and $10.00 — a 4.4× spread across 23 providers for identical hardware.
  • The gap between the two cards is 2.00×. Choosing the provider is worth more than choosing the card, and it is the part a spec sheet cannot tell you.
Other comparisons

More pairings need both cards' specs fetched first.

Working out what a whole run costs?
The calculator prices a job by hours and cards, from these same rates.
Open the calculator