Value Index
Dollars per gigabyte of VRAM and token throughput per dollar, ranked. Prices are the cheapest tracked (usually used-market) at the latest weekly snapshot (Oct 17, 2026). Tokens/s is the best community-verified single-stream generation run for that card — it shows what the hardware can do, not what every model will hit. Methodology.
| GPU | VRAM | Price | $/GB | Best tok/s | tok/s per $1000 | runs |
|---|---|---|---|---|---|---|
| Intel Arc Pro B70 | 32 GB | $1,300 new | $41 | 283.1 | 218 | 28 |
| RTX 3090 | 24 GB | $1,720 used | $72 | 709.5 | 412 | 25 |
| NVIDIA H200 SXM x4 | 564 GB | — | — | 495.9 | — | 1 |
| RTX PRO 6000 Blackwell x4 | 384 GB | — | — | 231.6 | — | 9 |
| RTX PRO 6000 Blackwell x2 | 192 GB | — | — | 204.5 | — | 6 |
| DGX Spark | 128 GB | — | — | 86.8 | — | 48 |
| DGX Spark x2 | 128 GB | — | — | 474.0 | — | 22 |
| DGX Spark x3 | 128 GB | — | — | 152.9 | — | 2 |
| DGX Spark x4 | 128 GB | — | — | 190.0 | — | 11 |
| RTX PRO 6000 Blackwell | 96 GB | — | — | 1271.3 | — | 32 |
| H100 NVL x2 | 94 GB | — | — | 293.3 | — | 1 |
| Multi-GPU x3 | 72 GB | — | — | 63.3 | — | 4 |
| CMP 170HX | 64 GB | — | — | 1392.4 | — | 23 |
| CMP 170HX x10 | 64 GB | — | — | 29.7 | — | 1 |
| CMP 170HX x2 | 64 GB | — | — | 186.9 | — | 5 |
| CMP 170HX x4 | 64 GB | — | — | 264.6 | — | 2 |
| Mac Studio (M4 Max) | 64 GB | — | — | 226.9 | — | 3 |
| Mac Studio (Mac16,9) with Apple M4 Max | 64 GB | — | — | 110.3 | — | 1 |
| RTX A6000 48GB | 48 GB | — | — | — | — | 0 |
| T4 x3 | 48 GB | — | — | 46.6 | — | 1 |
| A100 | 40 GB | — | — | 33.7 | — | 1 |
| Alveo V80 | 32 GB | — | — | — | — | 0 |
| Instinct MI100 x4 | 32 GB | — | — | 132.4 | — | 1 |
| Intel Arc Pro B70 x2 | 32 GB | — | — | 136.4 | — | 6 |
| Intel Arc Pro B70 x4 | 32 GB | — | — | 161.9 | — | 4 |
| Radeon AI Pro R9700 | 32 GB | — | — | 969.4 | — | 20 |
| Radeon AI Pro R9700 x2 | 32 GB | — | — | 131.4 | — | 3 |
| Radeon AI Pro R9700 x3 | 32 GB | — | — | 250.6 | — | 10 |
| Radeon AI Pro R9700 x4 | 32 GB | — | — | 53.1 | — | 1 |
| RTX 5090 | 32 GB | — | — | 4546.8 | — | 55 |
| RTX 5090 x2 | 32 GB | — | — | 195.6 | — | 1 |
| Tenstorrent P150 | 32 GB | — | — | 38.6 | — | 2 |
| Tesla P100 x2 | 32 GB | — | — | 51.3 | — | 1 |
| Tesla V100 | 32 GB | — | — | 4.3 | — | 1 |
| Tesla V100 x2 | 32 GB | — | — | 96.1 | — | 6 |
| Intel Arc Pro B60 | 24 GB | — | — | 77.2 | — | 6 |
| Multi-GPU x2 | 24 GB | — | — | 160.9 | — | 2 |
| RTX 3090 Ti | 24 GB | — | — | 97.0 | — | 1 |
| RTX 3090 x2 | 24 GB | — | — | 240.8 | — | 10 |
| RTX 3090 x4 | 24 GB | — | — | 145.8 | — | 1 |
| RTX 3090 x8 | 24 GB | — | — | 91.2 | — | 4 |
| RTX 4090 | 24 GB | — | — | 141.2 | — | 2 |
| RTX 4090 D | 24 GB | — | — | 70.9 | — | 2 |
| RX 7900 XTX | 24 GB | — | — | 550.2 | — | 6 |
| RX 7900 XTX x2 | 24 GB | — | — | 198.8 | — | 2 |
| RX 7900 XT | 20 GB | — | — | 60.6 | — | 1 |
| RX 7900 XT x2 | 20 GB | — | — | 116.0 | — | 1 |
| AMD RX 9060 XT 16GB | 16 GB | — | — | 39.8 | — | 1 |
| RTX 4060 Ti x2 | 16 GB | — | — | 44.0 | — | 2 |
| RTX 5000 | 16 GB | — | — | 76.6 | — | 1 |
| RTX 5060 Ti 16GB | 16 GB | — | — | — | — | 0 |
| RTX 5060 Ti x2 | 16 GB | — | — | 63.2 | — | 1 |
| RTX 5080 | 16 GB | — | — | 148.3 | — | 1 |
| RX 6900 XT | 16 GB | — | — | 346.1 | — | 2 |
| RX 7800 XT | 16 GB | — | — | 42.0 | — | 2 |
| RX 9070 XT | 16 GB | — | — | 545.3 | — | 3 |
| T4 | 16 GB | — | — | 53.3 | — | 1 |
| Tesla V100 PCIe 16GB | 16 GB | — | — | 40.7 | — | 1 |
| Tesla V100 x4 | 16 GB | — | — | 391.0 | — | 1 |
| VII | 16 GB | — | — | 88.2 | — | 1 |
| RTX 3060 | 12 GB | — | — | 310.4 | — | 19 |
| RTX 3060 x2 | 12 GB | — | — | 198.1 | — | 31 |
| RTX 4070 | 12 GB | — | — | 64.4 | — | 13 |
| RTX 5070 | 12 GB | — | — | 185.8 | — | 6 |
| GTX 1080 Ti | 11 GB | — | — | 94.9 | — | 1 |
| RTX 3080 | 10 GB | — | — | 277.8 | — | 10 |
| BC-250 | 8 GB | — | — | 42.1 | — | 1 |
| RTX 2060 Super | 8 GB | — | — | 97.1 | — | 3 |
| RTX 2080 | 8 GB | — | — | 22.9 | — | 2 |
| RTX 3060 Ti | 8 GB | — | — | 79.9 | — | 3 |
| RX 5700 XT | 8 GB | — | — | 273.6 | — | 3 |
| GTX 1060 x2 | 6 GB | — | — | 18.3 | — | 1 |
| RX 570 4GB x2 | 4 GB | — | — | 158.1 | — | 1 |