GPU comparison
B300 vs H200: Specs, monthly price and which to rent
B300 (Blackwell Ultra) has about twice the memory of an H200 (288 vs 141 GB), 1.7× the bandwidth (8 vs 4.8 TB/s), 2.3× the dense FP8 (4.5 vs 1.98 PF) and 13.5 PF of FP4 that Hopper does not have, for about 1.8× the monthly price. It costs less per unit of FP8 compute and fits much larger models; the H200 stays the best value when a model fits in 141 GB and your code targets Hopper.
- Updated Nov 4, 2026
- Specs: vendor datasheets, dense Tensor figures
- Prices: live CoreGPU catalog
| Spec | NVIDIA HGX B300 | NVIDIA H200 SXM |
|---|---|---|
| Architecture | Blackwell Ultra | Hopper |
| GPU memory | 288 GB HBM3e | 141 GB HBM3e |
| Memory bandwidth | 8 TB/s | 4.8 TB/s |
| FP8 dense | 4.5 PF | 1.98 PF |
| FP4 dense | 13.5 PF | n/a |
| Interconnect | NVLink 5 · 1.8 TB/s | NVLink 4 · 900 GB/s |
| Power (TDP) | 1,100 W | 700 W |
| Price per GPU / month | $4,899 | $2,699 |
| Price per GB of memory | $17.01 | $19.14 |
| Price per FP8 PFLOPS | $1,089 | $1,363 |
| Availability | Only 9 left | 34 available |
Green marks the better value on each line. Prices are per GPU per month at CoreGPU, at least 25% below the market median.
- FP4 and FP8 inference of reasoning models
- Models and KV caches beyond 141 GB per GPU
- Fewer GPUs per model
- 70B models at FP8 on a single GPU
- CUDA code and recipes tuned for Hopper
- A lower monthly price when 141 GB is enough
B300 vs H200: monthly price by server size
| Server | B300 | H200 | Difference |
|---|---|---|---|
| 1 GPU | $4,899/mo | $2,699/mo | $2,200 less on the H200 |
| 2 GPUs | $9,798/mo | $5,398/mo | $4,400 less on the H200 |
| 4 GPUs | $19,596/mo | $10,796/mo | $8,800 less on the H200 |
| 8 GPUs | $39,192/mo | $21,592/mo | $17,600 less on the H200 |
Flat monthly price per GPU, the same at every server size. USD, taxes excluded.
Largest model that fits in GPU memory
| Weights | 1× B300 | 1× H200 | 8× B300 | 8× H200 |
|---|---|---|---|---|
| BF16 | 120B | 58B | 960B | 470B |
| FP8 | 240B | 117B | 1.9T | 940B |
| FP4 | 480B | 235B | 3.8T | 1.9T |
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
B300 vs H200: common questions
Already a customer? Open a ticket from the console: engineers reply within one business day.
How support worksIs the B300 faster than the H200?
On dense FP8 Tensor throughput the B300 is faster: 4.5 PF against 1.98 PF per GPU. Memory bandwidth is 8 TB/s on the B300 and 4.8 TB/s on the H200, which matters most for LLM decoding.
How much does the B300 cost per month compared with the H200?
At CoreGPU the B300 is $4,899 and the H200 is $2,699 per GPU per month, so an 8-GPU server costs $39,192 or $21,592 a month. The H200 is 45% 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 B300 or the H200?
For large models and long contexts the B300 has the edge with 288 GB of memory per GPU, so a model needs fewer GPUs. Choose the B300 for fP4 and FP8 inference of reasoning models; the H200 for 70B models at FP8 on a single GPU.
Can I rent the B300 and the H200 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 B300 is only 9 left and the H200 is 34 available. You pay in BTC, ETH, XMR, USDT (TRC-20) or SOL.