[BLACKWELL]
NVIDIA B200
The Blackwell flagship — 2.5× H100 FP8 throughput.
The NVIDIA B200 (Blackwell) is the highest-throughput AI accelerator in commercial deployment, delivering up to 2.5× H100 FP8 performance with 192 GB HBM3e and 8 TB/s memory bandwidth. Used for frontier LLM training (1T+ params) and high-density inference.
VRAM
192 GB HBM3e
FP8 TFLOPS
4,500
Mem BW
8 TB/s
TDP
1000 W
Best for
- Frontier LLM training (1T+ parameters)
- High-throughput FP8 inference
- Multi-trillion token pretraining runs
Benchmarks
| WORKLOAD | METRIC | VALUE |
|---|---|---|
| GPT-3 175B pretraining | tokens/sec/GPU | ~3,100 |
| Llama-3 70B inference (FP8) | tokens/sec | ~5,400 |
| FP8 dense matmul | TFLOPS | 4,500 |
NVIDIA B200 availability from $38.00/hr
NVIDIA Blackwell
8x NVIDIA B200
Memory1536GB HBM
RegionUS-East-1
Availability62%
ProviderCoreWeave
$38.00/hr
Deploy →Definition
What is Servers.Computer?
Servers.Computer is an AI compute routing and procurement layer that benchmarks, compares, and deploys GPU clusters (NVIDIA H100, H200, B200 and AMD MI300) across global cloud providers in real time.