GPU comparison
B200 vs H100: Specs, monthly price and which to rent
B200 (Blackwell) has 2.25× the memory of an H100 (180 vs 80 GB), 2.4× the bandwidth (8 vs 3.35 TB/s), about 2.3× the dense FP8 (4.5 vs 1.98 PF) and FP4 Tensor Cores that Hopper does not have, for about 2.2× the monthly price. Per FP8 PFLOPS the two cost about the same: the B200 wins on memory, FP4 and the number of GPUs a model needs, the H100 on entry price and Hopper-tuned code.
- Updated Nov 4, 2026
- Specs: vendor datasheets, dense Tensor figures
- Prices: live CoreGPU catalog
| Spec | NVIDIA HGX B200 | NVIDIA H100 SXM |
|---|---|---|
| Architecture | Blackwell | Hopper |
| GPU memory | 180 GB HBM3e | 80 GB HBM3 |
| Memory bandwidth | 8 TB/s | 3.35 TB/s |
| FP8 dense | 4.5 PF | 1.98 PF |
| FP4 dense | 9 PF | n/a |
| Interconnect | NVLink 5 · 1.8 TB/s | NVLink 4 · 900 GB/s |
| Power (TDP) | 1,000 W | 700 W |
| Price per GPU / month | $4,449 | $1,999 |
| Price per GB of memory | $24.72 | $24.99 |
| Price per FP8 PFLOPS | $989 | $1,010 |
| Availability | Back Nov 20 | Only 3 left |
Green marks the better value on each line. Prices are per GPU per month at CoreGPU, at least 25% below the market median.
- Training and fine-tuning that fills 180 GB per GPU
- FP4 inference with Blackwell kernels
- Fewer GPUs per model thanks to 2.25× the memory
- Models that fit comfortably in 80 GB
- Code and recipes tuned for Hopper
- The lowest monthly price for an 8-GPU NVLink node
B200 vs H100: monthly price by server size
| Server | B200 | H100 | Difference |
|---|---|---|---|
| 1 GPU | $4,449/mo | $1,999/mo | $2,450 less on the H100 |
| 2 GPUs | $8,898/mo | $3,998/mo | $4,900 less on the H100 |
| 4 GPUs | $17,796/mo | $7,996/mo | $9,800 less on the H100 |
| 8 GPUs | $35,592/mo | $15,992/mo | $19,600 less on the H100 |
Flat monthly price per GPU, the same at every server size. USD, taxes excluded.
Largest model that fits in GPU memory
| Weights | 1× B200 | 1× H100 | 8× B200 | 8× H100 |
|---|---|---|---|---|
| BF16 | 75B | 33B | 600B | 266B |
| FP8 | 150B | 66B | 1.2T | 533B |
| FP4 | 300B | 133B | 2.4T | 1.1T |
Parameters whose weights fit with 20% headroom for activations and runtime. Long contexts and large batches need more room for the KV cache: see how much GPU memory you need.
More GPU comparisons
B200 vs H100: common questions
Already a customer? Open a ticket from the console: engineers reply within one business day.
How support worksIs the B200 faster than the H100?
On dense FP8 Tensor throughput the B200 is faster: 4.5 PF against 1.98 PF per GPU. Memory bandwidth is 8 TB/s on the B200 and 3.35 TB/s on the H100, which matters most for LLM decoding.
How much does the B200 cost per month compared with the H100?
At CoreGPU the B200 is $4,449 and the H100 is $1,999 per GPU per month, so an 8-GPU server costs $35,592 or $15,992 a month. The H100 is 55% cheaper per GPU. Both are at least 25% below the market median, billed monthly from a prepaid balance.
Which is better for LLM inference, the B200 or the H100?
For large models and long contexts the B200 has the edge with 180 GB of memory per GPU, so a model needs fewer GPUs. Choose the B200 for training and fine-tuning that fills 180 GB per GPU; the H100 for models that fit comfortably in 80 GB.
Can I rent the B200 and the H100 by the month?
Yes. Both are dedicated GPUs you rent by the month, from 1 to 8 per server, with root SSH access. Right now the B200 is sold out, back Nov 20 and the H100 is only 3 left. You pay in BTC, ETH, XMR, USDT (TRC-20) or SOL.