The NVIDIA B200 is a flagship Blackwell data-center GPU with 192 GB of HBM3e memory and 8 TB/s of bandwidth, delivering up to 20 petaFLOPS of FP4 AI compute. Available from $3.75/GPU-hour.
A dual-die design with 208 billion transistors, engineered for trillion-parameter AI and the largest training runs.
FP4 and FP8 compute. 2.25 PFLOPS FP4 for the largest transformer models at maximum throughput.
192 GB handles 405B at FP16 natively. 8 TB/s bandwidth, the fastest memory access of any GPU.
1.8 TB/s NVLink for tight multi-GPU coupling. Scale to 1,024+ GPUs with full bandwidth coherence.
Native FP4 mixed precision with micro-scaling. 2× the throughput of H100 on transformer inference.
Pre-train and fine-tune 70B+ and trillion-parameter models with 2nd-gen Transformer Engine and NVLink scale.
Serve the largest LLMs in production with FP4 precision and 192 GB of memory per card.
Adapt frontier open models on bursty, hourly capacity without reserving a card.
| Configuration | On-Demand / hr | Monthly | 3 Months | 6 Months | Annually |
|---|---|---|---|---|---|
| Dynamic & Dedicated | Dedicated only | ||||
| Most Popular1× NVIDIA B200 | Dynamic $3.75/hr Dedicated $5.90/hr | Dynamic $2,728/mo $3.74/hr eff. Dedicated $4,307/mo $5.90/hr eff. | Dynamic $8,184/3mo Dedicated $12,921/3mo | Dynamic $16,368/6mo Dedicated $25,842/6mo | Dynamic $32,736/yr Dedicated $51,684/yr |
| 2× NVIDIA B200Dedicated only | $11.80/hr | $8,614/mo $11.80/hr eff. | $25,842/3mo | $51,684/6mo | $103,368/yr |
| 4× NVIDIA B200Dedicated only | $23.60/hr | $17,228/mo $23.60/hr eff. | $51,684/3mo | $103,368/6mo | $206,736/yr |
| 8× NVIDIA B200Dedicated only | $47.20/hr | $34,456/mo $47.20/hr eff. | $103,368/3mo | $206,736/6mo | $413,472/yr |
| Deploy hourly → | Subscribe to Monthly → | ||||
The B200 is NVIDIA's flagship Blackwell data-center GPU with 192 GB of HBM3e memory at 8 TB/s, delivering up to 20 petaFLOPS of FP4 AI compute.
B200 starts at $3.75/GPU-hour Dynamic and $5.90/GPU-hour Dedicated.
The B200 is roughly 2.5× faster for LLM workloads than the H100, with 192 GB vs 80 GB memory.
Each B200 has 192 GB of HBM3e at 8 TB/s.
Large-scale training, fine-tuning frontier models, high-throughput LLM inference with FP4, and HPC.
On Dynamic, SSH-ready in under 5 minutes.
Launch a B200 on-demand from $3.75/GPU-hour, $5.90/hr dedicated, or $4,307/mo.
On-demand · hourly billing · US & EU regions