Skip to content

Instantly share code, notes, and snippets.

@m4r1k
Last active April 15, 2026 21:53
Show Gist options
  • Select an option

  • Save m4r1k/2c5fde62a8e2a8cf386a26be9aadc8cc to your computer and use it in GitHub Desktop.

Select an option

Save m4r1k/2c5fde62a8e2a8cf386a26be9aadc8cc to your computer and use it in GitHub Desktop.
gpu-comparison
GPU Arch Year FP32 FP16/BF16 FP8 FP4 Bandwidth Memory Ops:byte
P100 SXM2 Pascal 2016 10.6 21.2 (FP16) -- -- 720 GB/s 16 GB HBM2 29
V100 SXM Volta 2017 15.7 125 (FP16) -- -- 900 GB/s 32 GB HBM2 139
A100 SXM Ampere 2020 19.5 312 (BF16) -- -- 2039 GB/s 80 GB HBM2e 153
H100 SXM Hopper 2022 67 989 (BF16) 1979 -- 3350 GB/s 80 GB HBM3 295
B200 HGX Blackwell 2024 75 2250 (BF16) 4500 9000 7700 GB/s 180 GB HBM3e 292
B300 NVL72 Blackwell Ultra 2025 ~90 2500 (BF16) 5000 15000 8000 GB/s 288 GB HBM3e 313
DGX Spark Blackwell 2025 31 ~125 (BF16) ~250 ~500 273 GB/s 128 GB LPDDR5x ~458
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment