NVIDIA B100 SXM 192GB vs NVIDIA B200 SXM 180GB

The B200 is the newer part: Blackwell, launched in 2025, against the B100's Blackwell from 2024. Newer architectures typically add lower-precision formats and better throughput per watt, so check the precision rows your workload actually uses.

Dense throughput for the B100 against the B200: FP64 30 vs 37 TFLOPS, FP32 60 vs 75 TFLOPS, FP16 1,750 vs 2,250 TFLOPS, FP8 3,500 vs 4,500 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 192 GB against 180 GB, fed at 8.0 TB/s versus 7.7 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.

Power budgets are 700 W for the B100 and 1.0 kW for the B200. At FP32 that works out to 0.09 against 0.07 TFLOPS per watt.

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.