NVIDIA GB300 NVL72 GPU 288GB vs NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip 252GB

Both GPUs launched in 2025: the GB300 NVL72 GPU 288GB on Blackwell Ultra and the GB300 Grace Blackwell Ultra Desktop Superchip 252GB on Blackwell Ultra.

Dense throughput for the GB300 NVL72 GPU 288GB against the GB300 Grace Blackwell Ultra Desktop Superchip 252GB: FP64 1.4 vs 1.3 TFLOPS, FP32 83 vs 80 TFLOPS, FP16 2,500 vs 2,500 TFLOPS, FP8 5,000 vs 5,000 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 288 GB against 252 GB, fed at 8.0 TB/s versus 7.1 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.