Written by Jakub Rusinowski · Last updated July 21, 2026
Ranked for writing and editing: Drafting, rewriting and editing prose where voice and readability matter as much as correctness.
Best overall: NVIDIA GeForce RTX 5070 Ti
16 GB VRAM at 896 GB/s. It runs 64 of the models that qualify for this workload; the strongest is Mistral Small 3.1 24B at an estimated 40.9 tokens/sec.
| GPU | VRAM | MSRP | Models that fit | Best model it runs | Est. speed | |
|---|---|---|---|---|---|---|
| Best overall | NVIDIA GeForce RTX 5070 Ti | 16 GB | $749 | 64 | Mistral Small 3.1 24B | ~40.9 tok/s |
| Best value | Intel Arc B570 | 10 GB | $219 | 56 | Gemma 3 12B Instruct | ~30.2 tok/s |
| Budget pick | Intel Arc B570 | 10 GB | $219 | 56 | Gemma 3 12B Instruct | ~30.2 tok/s |
| Most memory | NVIDIA DGX Spark | 128 GB | $4,699 | 92 | Nemotron 70B Instruct | ~4.6 tok/s |
| GPU | Score | VRAM | MSRP | Models fit | Est. speed | Tok/s per watt | Cost per model |
|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5070 Ti | 63.8 | 16 GB | $749 | 64 | ~40.9 | 0.14 | $12 |
| NVIDIA GeForce RTX 5080 | 61.7 | 16 GB | $999 | 64 | ~43.5 | 0.12 | $16 |
| AMD Radeon RX 9070 XT | 60.1 | 16 GB | $599 | 64 | ~33.3 | 0.15 | $9 |
| NVIDIA GeForce RTX 5060 | 57.8 | 8 GB | $299 | 46 | ~49.5 | 0.34 | $7 |
| AMD Radeon RX 9070 | 57.7 | 16 GB | $549 | 64 | ~30 | 0.14 | $9 |
| AMD Radeon RX 7800 XT | 57.5 | 16 GB | $499 | 64 | ~29.3 | 0.11 | $8 |
| NVIDIA GeForce RTX 4080 Super | 56.7 | 16 GB | $999 | 64 | ~34.2 | 0.11 | $16 |
| NVIDIA GeForce RTX 4070 Ti Super | 55.7 | 16 GB | $799 | 64 | ~31.4 | 0.11 | $12 |
| NVIDIA RTX 6000 Ada Generation | 55.2 | 48 GB | $6,799 | 84 | ~15.8 | 0.05 | $81 |
| NVIDIA GeForce RTX 4080 | 55.1 | 16 GB | $1,199 | 64 | ~33.3 | 0.1 | $19 |
| AMD Radeon RX 9060 XT 8GB | 54.2 | 8 GB | $299 | 46 | ~36.5 | 0.24 | $7 |
| NVIDIA L40S | 53.6 | 48 GB | $7,499 | 84 | ~14.3 | 0.04 | $89 |
The NVIDIA GeForce RTX 5070 Ti — 16 GB of VRAM runs 64 qualifying models, the strongest being Mistral Small 3.1.
The Intel Arc B570 at $219, which runs 56 qualifying models.
8 GB is the entry point at which a model for this workload will run at all. More memory buys a stronger model, not just a faster one.