Loading...
Loading...
192GB HBM3e | 8TB/s Bandwidth | Blackwell | 1000W TDP | NVLink 5.0
The NVIDIA B200 SXM is the most powerful single-chip AI accelerator ever built, featuring 192GB of HBM3e with 8TB/s bandwidth on the Blackwell architecture. Its 1000W TDP powers fifth-gen Tensor Cores and a second-gen Transformer Engine for unprecedented AI training throughput.
Architecture
Blackwell (GB200)
Memory
192 GB HBM3e
Memory Bandwidth
8 TB/s
Full hardware specifications for the NVIDIA B200 SXM
Architecture
Blackwell (GB200)
Memory
192 GB HBM3e
Memory Bandwidth
8 TB/s
TDP
1000W
Interface
NVLink 5.0 / SXM6
FP4 Performance
18,000 TFLOPS
141GB HBM3e | 4.8TB/s Bandwidth | Hopper | 700W TDP | PCIe 5.0
Hopper (GH200)
80GB HBM3 | 3.35TB/s Bandwidth | Hopper | 700W TDP | NVLink 4.0
Hopper (GH100)
48GB GDDR6 | 864GB/s Bandwidth | Ada Lovelace | 350W TDP | PCIe 4.0 x16
Ada Lovelace (AD102)
80GB HBM2e | 2TB/s Bandwidth | Ampere | 400W TDP | NVLink 3.0
Ampere (GA100)