Written by Jakub Rusinowski · Last updated July 21, 2026
Ranked for fully offline and air-gapped use: A model that works with no internet connection at all — on a laptop, in the field, or inside an air-gapped network.
Best overall: AMD Radeon RX 7800 XT
16 GB VRAM at 624 GB/s. It runs 64 of the models that qualify for this workload; the strongest is Qwen 3 14B at an estimated 43.9 tokens/sec.
| GPU | VRAM | MSRP | Models that fit | Best model it runs | Est. speed | |
|---|---|---|---|---|---|---|
| Best overall | AMD Radeon RX 7800 XT | 16 GB | $499 | 64 | Qwen 3 14B | ~43.9 tok/s |
| Best value | Intel Arc B570 | 10 GB | $219 | 56 | Gemma 4 12B (Unified) | ~33.4 tok/s |
| Budget pick | Intel Arc B570 | 10 GB | $219 | 56 | Gemma 4 12B (Unified) | ~33.4 tok/s |
| Most memory | NVIDIA DGX Spark | 128 GB | $4,699 | 92 | Qwen 3.5 122B-A10B (MoE) | ~25.8 tok/s |
| GPU | Score | VRAM | MSRP | Models fit | Est. speed | Tok/s per watt | Cost per model |
|---|---|---|---|---|---|---|---|
| AMD Radeon RX 7800 XT | 63.6 | 16 GB | $499 | 64 | ~43.9 | 0.17 | $8 |
| AMD Radeon RX 9070 | 63.4 | 16 GB | $549 | 64 | ~44.9 | 0.2 | $9 |
| AMD Radeon RX 9070 XT | 62.7 | 16 GB | $599 | 64 | ~49.7 | 0.23 | $9 |
| NVIDIA GeForce RTX 5070 Ti | 60.1 | 16 GB | $749 | 64 | ~60.5 | 0.2 | $12 |
| NVIDIA GeForce RTX 4070 Ti Super | 59.8 | 16 GB | $799 | 64 | ~46.9 | 0.16 | $12 |
| NVIDIA GeForce RTX 4080 Super | 58.3 | 16 GB | $999 | 64 | ~50.9 | 0.16 | $16 |
| NVIDIA GeForce RTX 5080 | 58 | 16 GB | $999 | 64 | ~64.1 | 0.18 | $16 |
| NVIDIA GeForce RTX 4080 | 57.6 | 16 GB | $1,199 | 64 | ~49.7 | 0.16 | $19 |
| NVIDIA GeForce RTX 5060 Ti 16GB | 57 | 16 GB | $429 | 64 | ~32.4 | 0.18 | $7 |
| NVIDIA GeForce RTX 5060 | 56.4 | 8 GB | $299 | 46 | ~49.5 | 0.34 | $7 |
| NVIDIA GeForce RTX 5060 Ti 8GB | 52 | 8 GB | $379 | 46 | ~49.5 | 0.28 | $8 |
| AMD Radeon RX 9060 XT 8GB | 51.7 | 8 GB | $299 | 46 | ~36.5 | 0.24 | $7 |
The AMD Radeon RX 7800 XT — 16 GB of VRAM runs 64 qualifying models, the strongest being Qwen 3.
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.