NVIDIA GB200 Grace Blackwell Superchip 372GB vs NVIDIA GB200 NVL72 GPU 186GB

Both GPUs launched in 2024: the GB200 Grace Blackwell Superchip 372GB on Blackwell and the GB200 NVL72 GPU 186GB on Blackwell.

Dense throughput for the GB200 Grace Blackwell Superchip 372GB against the GB200 NVL72 GPU 186GB: FP64 80 vs 40 TFLOPS, FP32 160 vs 80 TFLOPS, FP16 5,000 vs 2,500 TFLOPS, FP8 10,000 vs 5,000 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 372 GB against 186 GB, fed at 16.0 TB/s versus 8.0 TB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.

For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.