NEXT-GEN AI
Blackwell Data Center B200 SXM Blackwell
NVIDIA B200 SXM6
The world's most powerful AI accelerator. The B200 SXM6 rewrites the record books with 192 GB of HBM3e, 8 TB/s of memory bandwidth, and 9,000 TFLOPS of FP8 AI compute — purpose-built for next-generation foundation model training and inference at planetary scale.
20,480 CUDA Cores
192 GB HBM3e
8000 GB/s Bandwidth
1000W TDP
MSRP $40,000
Architecture Blackwell
GPU Chip GB100
Fabrication TSMC 4NP
CUDA Cores 20,480
SM Count 160
Tensor Cores 640
RT Cores 0
Boost Clock 1,884 MHz
Memory Size 192 GB HBM3e
Memory Bus 8192-bit
Memory Bandwidth 8000 GB/s
L2 Cache 192 MB
TDP 1000 W
Launch Date November 19, 2024
- 5th-Gen Tensor Cores (FP4 / FP8)
- Second-Gen Transformer Engine
- NVLink 5.0 (1,800 GB/s)
- NVSwitch 4.0 for GB200 NVL72
- RAS (Reliability / Availability)
- Confidential Computing
- SXM6 Form Factor
Foundation Model Training
Frontier AI Research
Agentic AI Systems
HPC Simulation
Genomics