Description
Architecture: Blackwell Ultra
Process: TSMC 4NP
Transistor: 208 billion
CUDA core: 20480 units
Tensor Core: 640 fifth generation Tensor Cores
SM unit: 160 units
Video Memory: 288GB HBM3e
Video memory bit width: 8192 bits
Video memory bandwidth: 8TB/s
High speed cache: 40MB TMEM
Interconnection: Fifth generation NVLink, single link bandwidth of 1.8TB/s; The interconnection bandwidth between NV-HBI Die is 10TB/s
Interface: Supports PCIe 6.0
FP4 Sparse Computing Power: 20 PFLOPS
FP8 Dense Computing Power: 5 PFLOPS
Maximum power consumption (TGP): 1400W
Process: TSMC 4NP
Transistor: 208 billion
CUDA core: 20480 units
Tensor Core: 640 fifth generation Tensor Cores
SM unit: 160 units
Video Memory: 288GB HBM3e
Video memory bit width: 8192 bits
Video memory bandwidth: 8TB/s
High speed cache: 40MB TMEM
Interconnection: Fifth generation NVLink, single link bandwidth of 1.8TB/s; The interconnection bandwidth between NV-HBI Die is 10TB/s
Interface: Supports PCIe 6.0
FP4 Sparse Computing Power: 20 PFLOPS
FP8 Dense Computing Power: 5 PFLOPS
Maximum power consumption (TGP): 1400W
