Used Model
GPU Architecture
NVIDIA Hopper
NVIDIA Blackwell
NVIDIA Blackwell
GPU Memory
141GB
96GB
32GB
Cuda Cores
16,896
24,064
21,760
Tensor Cores
528
752
680
RT Cores
×
188
170
Bandwidth
4,890 GB/s
1,792 GB/s
1,790 GB/s
Base Clock
1.365 GHz
1.590 GHz
2.017 GHz
Boost Clock
1.980 GHz
2.617 GHz
2.407 GHz
Memory Clock
1.539 GHz
1.750 GHz
1.75 GHz
Memory Bus
6,144-bit
512-bit
512-bit
NVLink
O
×
×
Double-Precision Performance (FP64)
FP64 : 30 TFLOPS
FP64 Tensor Core : 60 TFLOPS
FP64 : 1.968 TFLOPS
FP64 : 1.637 TFLOPS
Single-Precision Performance (FP32)
FP32 : 60 TFLOPS
FP32 Tensor Core : 835 TFLOPS
FP32 (Float) [TF32] : 126 TFLOPS
FP32 Float [TF32] : 104.8 TFLOPS
Half-Precision Performance (FP16)
FP32 Tensor Core : 1,671 TFLOPS
126 TFLOPS
104.8 TFLOPS
Infer Performance
INT8 : 3,341 TOPS
×
3,352 TOPS
