GPU Specs Database
Full specifications for datacenter and workstation accelerators — memory, dense compute throughput, interconnect, power, and silicon. Sortable on every column, with dense and sparse figures kept separate so the comparisons stay honest.
Fastest FP16 (dense)
B300
2,250 TFLOPS
Most memory
B300
288 GB HBM3e
Widest bandwidth
B300
8,000 GB/s
Best TFLOPS per watt
B200
2.25 TFLOPS/W
34 of 34 GPUs
| Arch | ||||||||
|---|---|---|---|---|---|---|---|---|
| B300 NVIDIA | Blackwell | 288 GB | 8,000 GB/s | 2,250 | 4,500 | 1400 W | 1.61 | Specs |
| B200 NVIDIA | Blackwell | 192 GB | 8,000 GB/s | 2,250 | 4,500 | 1000 W | 2.25 | Specs |
| MI325X AMD | CDNA 3 | 256 GB | 6,000 GB/s | 1,307.4 | 2,614.9 | 1000 W | 1.31 | Specs |
| MI300X AMD | CDNA 3 | 192 GB | 5,300 GB/s | 1,307.4 | 2,614.9 | 750 W | 1.74 | Specs |
| GH200 Grace Hopper NVIDIA | Hopper | 144 GB | 4,900 GB/s | 989.5 | 1,979 | 1000 W | 0.99 | Specs |
| H200 SXM NVIDIA | Hopper | 141 GB | 4,800 GB/s | 989.5 | 1,979 | 700 W | 1.41 | Specs |
| H100 SXM NVIDIA | Hopper | 80 GB | 3,350 GB/s | 989.5 | 1,979 | 700 W | 1.41 | Specs |
| H100 NVL NVIDIA | Hopper | 94 GB | 3,900 GB/s | 890 | 1,780 | 400 W | 2.23 | Specs |
| H100 PCIe NVIDIA | Hopper | 80 GB | 2,000 GB/s | 756 | 1,513 | 350 W | 2.16 | Specs |
| L40S NVIDIA | Ada Lovelace | 48 GB | 864 GB/s | 362 | 733 | 350 W | 1.03 | Specs |
| A100 SXM 80GB NVIDIA | Ampere | 80 GB | 2,039 GB/s | 312 | n/a | 400 W | 0.78 | Specs |
| A100 PCIe 80GB NVIDIA | Ampere | 80 GB | 1,935 GB/s | 312 | n/a | 300 W | 1.04 | Specs |
| A100 PCIe 40GB NVIDIAlegacy | Ampere | 40 GB | 1,555 GB/s | 312 | n/a | 250 W | 1.25 | Specs |
| RTX 5090 NVIDIA | Blackwell | 32 GB | 1,792 GB/s | 209.5 | 838 | 575 W | 0.36 | Specs |
| RTX 6000 Ada NVIDIA | Ada Lovelace | 48 GB | 960 GB/s | 182 | 728 | 300 W | 0.61 | Specs |
| L40 NVIDIA | Ada Lovelace | 48 GB | 864 GB/s | 181 | 362 | 300 W | 0.60 | Specs |
| RTX 4090 NVIDIA | Ada Lovelace | 24 GB | 1,008 GB/s | 165.2 | 660 | 450 W | 0.37 | Specs |
| A30 NVIDIA | Ampere | 24 GB | 933 GB/s | 165 | n/a | 165 W | 1.00 | Specs |
| A6000 NVIDIA | Ampere | 48 GB | 768 GB/s | 154.8 | n/a | 300 W | 0.52 | Specs |
| A40 NVIDIA | Ampere | 48 GB | 696 GB/s | 149.7 | n/a | 300 W | 0.50 | Specs |
| V100 SXM2 32GB NVIDIAlegacy | Volta | 32 GB | 900 GB/s | 125 | n/a | 300 W | 0.42 | Specs |
| A10G NVIDIA | Ampere | 24 GB | 600 GB/s | 125 | n/a | 150 W | 0.83 | Specs |
| L4 NVIDIA | Ada Lovelace | 24 GB | 300 GB/s | 121 | 242 | 72 W | 1.68 | Specs |
| Tesla V100 PCIe 32GB NVIDIAlegacy | Volta | 32 GB | 900 GB/s | 112 | n/a | 250 W | 0.45 | Specs |
| RTX 3090 NVIDIAlegacy | Ampere | 24 GB | 936 GB/s | 71 | n/a | 350 W | 0.20 | Specs |
| T4 NVIDIAlegacy | Turing | 16 GB | 320 GB/s | 65 | n/a | 70 W | 0.93 | Specs |
| Tesla P100 SXM2 NVIDIAlegacy | Pascal | 16 GB | 732 GB/s | 21.2 | n/a | 300 W | 0.07 | Specs |
| Rubin R100 NVIDIA | Rubin | 288 GB | — | — | n/a | — | — | Specs |
| Tesla K80 NVIDIAlegacy | Kepler | 24 GB | 480 GB/s | — | n/a | 300 W | — | Specs |
| Tesla M40 NVIDIAlegacy | Maxwell | 24 GB | 288 GB/s | — | n/a | 250 W | — | Specs |
| Tesla P40 NVIDIAlegacy | Pascal | 24 GB | 346 GB/s | — | n/a | 250 W | — | Specs |
| Tesla M2090 NVIDIAlegacy | Fermi | 6 GB | 177 GB/s | — | n/a | 225 W | — | Specs |
| Tesla C1060 NVIDIAlegacy | Tesla | 4 GB | 102 GB/s | — | n/a | 187 W | — | Specs |
| Tesla C870 NVIDIAlegacy | Tesla | 1 GB | 77 GB/s | — | n/a | 170 W | — | Specs |
All TFLOPS figures are dense tensor throughput. Vendors often headline sparse numbers, which are 2× higher and only achievable on models with structured sparsity — most workloads will not see them. FP8 shows n/a for architectures predating hardware FP8 (Ampere and earlier).