NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip 252GB vs NVIDIA B300 SXM 288GB

Both GPUs launched in 2025: the GB300 on Blackwell Ultra and the B300 on Blackwell Ultra.

Dense throughput for the GB300 against the B300: FP64 1.3 vs 1.3 TFLOPS, FP32 80 vs 75 TFLOPS, FP16 2,500 vs 2,250 TFLOPS, FP8 5,000 vs 4,500 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 252 GB against 288 GB, fed at 7.1 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.