NVIDIA B200 Blackwell Ultra — next-generation AI supercomputing accelerator. 192GB HBM3e at 8 TB/s, up to 4.5 PFLOPS FP4 with Transformer Engine, 1000W TDP. Purpose-built for trillion-parameter LLM training and inference at unprecedented scale.
B200 Blackwell Ultra
Blackwell GB200
192 GB HBM3e
8 TB/s
4.5 PFLOPS
2.25 PFLOPS
1.125 PFLOPS
562 TFLOPS
SXM6 / NVLink 5
1000W
1.8 TB/s bidirectional
5th Gen FP4/FP6
The NVIDIA B200 Blackwell Ultra represents the cutting edge of AI acceleration. With 192GB of HBM3e memory at 8 TB/s bandwidth and 5th-generation Tensor Cores delivering 4.5 PFLOPS of FP4 compute, the B200 is engineered for training trillion-parameter foundation models, real-time generative AI inference at massive scale, and scientific computing workloads that were previously infeasible. NVLink 5 enables 1.8 TB/s GPU-to-GPU communication in DGX B200 configurations, making 8-way and beyond multi-GPU scaling practical for the largest models.
Need NVIDIA B200 Blackwell Ultra?
Contact QS Compute for DGX B200 system configurations, pricing, and availability.
Request Quote