BTC
ETH
USDT
XMR
LTC
GPU comparison · Hopper vs Blackwell
NVIDIA H200 vs B200.
A generation apart. The B200 (Blackwell) has 180 GB of HBM3e at 8 TB/s, about 2.3× the dense FP8 and BF16 throughput of the H200, FP4 tensor cores and NVLink 5. The H200 (Hopper) costs $1,160 less per GPU per month and keeps stronger FP64 tensor math.
- 141 vs 180 GBGPU memory
- 4.8 vs 8 TB/smemory bandwidth
- $2,279 vs $3,439per GPU, per month
H200 vs B200: which one to rent.
Choose the H200 if
- Your models fit in 141 GB per GPU and the H200 is fast enough for your traffic.
- You rely on FP64 tensor math for simulation (67 TFLOPS on Hopper).
- You want the lower monthly price: $2,279 per GPU against $3,439.
Choose the B200 if
- You need throughput: 4.5 PFLOPS of dense FP8 per GPU against 1.98, and 1.7× the memory bandwidth.
- You want FP4 inference, which Hopper does not support.
- You train across 8 GPUs: NVLink 5 moves 1.8 TB/s per GPU, twice NVLink 4.
H200 vs B200 specs, side by side.
Figures from NVIDIA’s published specifications, per GPU. Tensor figures are peak dense throughput.
| Per GPU | NVIDIA H200 | NVIDIA B200 | B200 / H200 |
|---|---|---|---|
| Architecture | Hopper | Blackwell, TSMC 4NP | — |
| Form factor | SXM module | SXM module (HGX B200) | — |
| GPU memory | 141 GB HBM3e | 180 GB HBM3e | 1.28× |
| Memory bandwidth | 4.8 TB/s | Up to 8 TB/s | 1.67× |
| GPU-to-GPU link | NVLink 4, 900 GB/s per GPU | NVLink 5, 1.8 TB/s per GPU | 2× |
| FP4 tensor (dense) | Not supported | 9.0 PFLOPS | — |
| FP8 tensor (dense) | 1,979 TFLOPS | 4,500 TFLOPS | 2.3× |
| FP16 / BF16 tensor (dense) | 989 TFLOPS | 2,250 TFLOPS | 2.3× |
| FP64 | 34 / 67 TFLOPS | 37 TFLOPS | 1.09× |
| Max power | Up to 700 W, configurable | Up to 1,000 W, configurable | — |
| GPUs per server | 1, 2, 4, 8 | 1, 2, 4, 8 | — |
| Per GPU in our servers | 24 vCPU · 256 GB RAM · 3.84 TB | 28 vCPU · 288 GB RAM · 3.84 TB | — |
Sources: NVIDIA H200 · NVIDIA HGX · NVIDIA DGX B200. Full sheets: H200 · B200
H200 vs B200 rental price.
Our monthly prices, set 30% below the market median of public on-demand prices and paid in crypto. Hourly figures are the monthly price divided by 730 hours.
| CryptGPU | NVIDIA H200 | NVIDIA B200 | B200 / H200 |
|---|---|---|---|
| Price per GPU, per month | $2,279 | $3,439 | 1.51× |
| Equivalent per GPU-hour | $3.12 | $4.71 | — |
| Per GB of GPU memory, per month | $16.16 | $19.11 | 1.18× |
| Per PFLOPS of dense FP8, per month | $1,152 | $764 | 0.66× |
| Largest server | 8× · $18,232/mo | 8× · $27,512/mo | — |
| Market median, per GPU | $3,259 | $4,920 | — |
| Below the median | −30% | −30% | — |
Same price per GPU at every server size, no hourly metering. How we compare prices · Configure an H200 server · Configure a B200 server
Which models fit on each.
Smallest server that holds each model, by precision. Rule of thumb with headroom for the KV cache, 92% of GPU memory usable.
| Model (total parameters) | H200 · FP8 | B200 · FP8 | H200 · 4-bit | B200 · 4-bit |
|---|---|---|---|---|
| Qwen3.5-9B | 1× | 1× | 1× | 1× |
| Gemma 4 31B | 1× | 1× | 1× | 1× |
| Llama 3.3 70B | 1× | 1× | 1× | 1× |
| gpt-oss-120b | 2× | 1× | 1× | 1× |
| DeepSeek-V4-Flash | 4× | 4× | 2× | 2× |
| Qwen3.5-397B-A17B | 4× | 4× | 2× | 2× |
| DeepSeek-R1 (671B) | 8× | 8× | 4× | 4× |
| Kimi K2.6 (1T) | — | 8× | 8× | 4× |
FP8 ≈ 1.2 bytes and 4-bit ≈ 0.65 bytes per parameter. — = larger than the biggest server of that GPU. Size another model · How much VRAM does an LLM need?
H200 vs B200 FAQ.
More on each GPU: NVIDIA H200 · NVIDIA B200.
Is the B200 worth the extra cost over the H200?
Per unit of compute it is cheaper: dense FP8 costs about $764 per PFLOPS per month on the B200 against $1,152 on the H200. If your workload does not use the extra throughput, the H200 costs less.
How much memory does each have?
141 GB of HBM3e at 4.8 TB/s on the H200, 180 GB of HBM3e at 8 TB/s on the B200. An 8-GPU server holds 1,128 GB or 1,440 GB.
Does my CUDA code run on the B200?
Yes, with a CUDA version that supports Blackwell (CUDA 12.8 or later) and current frameworks. Kernels tuned for Hopper run but may need Blackwell builds to reach full speed.
Other GPU comparisons.
Rent the H200 or the B200.
Dedicated servers with 1 to 8 GPUs, one monthly price, paid in crypto. Online in under 10 minutes, no KYC.