Comparator
Compare GPUs for local AI
Up to three cards or machines side by side, with the expected speed on the model of your choice. The best of each row is highlighted.
RTX 5090 vs RTX 4090NVIDIA DGX Spark vs RTX 5090RTX 3090 vs RTX 4090Mac Studio M4 Max vs NVIDIA DGX SparkMini-PC Ryzen AI Max+ 395 vs NVIDIA DGX SparkH100 vs H200RTX PRO 6000 Blackwell vs RTX 6000 AdaRadeon RX 7900 XTX vs RTX 4090
| RTX 4090 24 GB | RTX 5090 32 GB | NVIDIA DGX Spark 128 GB | |
|---|---|---|---|
| Usable memory | 24 GB | 32 GB | 109 GB |
| Bandwidth | 1,008 GB/s | 1,792 GB/s | 273 GB/s |
| FP16 compute | ~165 TF | ~210 TF | ~100 TF |
| Power | 450 W | 575 W | 240 W |
| Indicative price | €2,000 | €2,600 | €4,000 |
| Complete machine | €2,900 | €3,500 | €4,000 |
| € per usable GB | €121 | €109 | €37 |
| Qwen3 8B (Q4_K_M) | ~109 tok/s | ~194 tok/s | ~30 tok/s |
| € per tok/s on Qwen3 8B | €27 | €18 | €135 |
| Biggest model alone (Q4) | DeepSeek R1 Distill 32B | Mixtral 8x7B | Mistral Large 123B |
For Qwen3 8B: the fastest is RTX 5090 32 GB (~194 tok/s), the best speed for the price is RTX 5090 32 GB.
Can I run it?
Pick a model and your hardware: instant verdict, and the cheapest alternatives if it doesn’t fit.
Yes, but it’s tight
47.4 GB needed48 GB available
Speed~18 tok/s
Occupancy99 %
Machine€2,400
Other options
| Mini-PC Ryzen AI Max+ 395 128 GB | ~3.3 tok/s | €2,300 |
| NVIDIA DGX Spark 128 GB | ~3.5 tok/s | €4,000 |
| Mac Studio M4 Max 128 GB | ~7.1 tok/s | €4,400 |
| 4 × RTX 4060 Ti 16 GB | ~11 tok/s | €4,600 |
| 4 × Radeon RX 7800 XT 16 GB | ~24 tok/s | €4,800 |