Best CPUs for AI work 2026
Only CPUs with a built-in NPU are listed — a separate block for neural networks. It handles background blur in video calls, live captions, Copilot features and other Windows AI tasks without loading the cores or draining the battery. Microsoft requires an NPU of 40+ TOPS for Copilot+ PC features.
| # | CPU | Type | Task score | NPU (AI accelerator) | Score |
|---|---|---|---|---|---|
| 1 | Ryzen AI Max+ 395 | Laptop | 50 TOPS | 70.63 | |
| 2 | Ryzen AI 9 HX 370 | Laptop | 50 TOPS | 48.22 | |
| 3 | Ryzen AI 7 350 | Laptop | 50 TOPS | 37.84 | |
| 4 | Ryzen AI 5 340 | Laptop | 50 TOPS | 31.88 | |
| 5 | Core Ultra 7 258V | Laptop | 47 TOPS | 32.53 | |
| 6 | Core Ultra 5 226V | Laptop | 40 TOPS | 30.58 | |
| 7 | Ryzen 7 8700G | Desktop | 16 TOPS | 48.69 | |
| 8 | Core Ultra 9 275HX | Laptop | 13 TOPS | 68.09 | |
| 9 | Ryzen 9 8945HS | Laptop | 16 TOPS | 42.68 | |
| 10 | Ryzen 7 8845HS | Laptop | 16 TOPS | 41.53 | |
| 11 | Ryzen 5 8600G | Desktop | 16 TOPS | 40.70 | |
| 12 | Core Ultra 9 285H | Laptop | 13 TOPS | 53.36 | |
| 13 | Core Ultra 7 255H | Laptop | 13 TOPS | 47.48 | |
| 14 | Core Ultra 9 185H | Laptop | 11 TOPS | 43.03 | |
| 15 | Core Ultra 7 155H | Laptop | 11 TOPS | 37.91 | |
| 16 | Ryzen 9 7940HS | Laptop | 10 TOPS | 42.68 | |
| 17 | Ryzen 7 7840HS | Laptop | 10 TOPS | 41.53 | |
| 18 | Core Ultra 5 125H | Laptop | 11 TOPS | 31.46 |
Compare the top two: Ryzen AI Max+ 395 vs Ryzen AI 9 HX 370 →
Graphics cards for neural networks
To run language models and image generation on your own PC, video memory matters most: the model has to fit into it completely. So memory comes first here, then speed. Roughly: 8 GB fits small quantized models, 16 GB medium ones, 24–32 GB large ones. Most AI software works best with NVIDIA cards (CUDA).
| # | Graphics card | VRAM | Score | AI TOPS (as stated by the maker) |
|---|---|---|---|---|
| 1 | GeForce RTX 5090 | 32 GB GDDR7 | 3352 | |
| 2 | GeForce RTX 4090 | 24 GB GDDR6X | 1321 | |
| 3 | Radeon RX 7900 XTX | 24 GB GDDR6 | — | |
| 4 | GeForce RTX 3090 Ti | 24 GB GDDR6X | — | |
| 5 | GeForce RTX 3090 | 24 GB GDDR6X | — | |
| 6 | Radeon RX 7900 XT | 20 GB GDDR6 | — | |
| 7 | GeForce RTX 5080 | 16 GB GDDR7 | 1801 | |
| 8 | GeForce RTX 4080 SUPER | 16 GB GDDR6X | 836 | |
| 9 | GeForce RTX 5070 Ti | 16 GB GDDR7 | 1406 | |
| 10 | GeForce RTX 4080 | 16 GB GDDR6X | 780 | |
| 11 | Radeon RX 9070 XT | 16 GB GDDR6 | 779 | |
| 12 | GeForce RTX 4070 Ti SUPER | 16 GB GDDR6X | 706 | |
| 13 | Radeon RX 7900 GRE | 16 GB GDDR6 | — | |
| 14 | Radeon RX 9070 | 16 GB GDDR6 | 578 | |
| 15 | Radeon RX 6950 XT | 16 GB GDDR6 | — |
How we rank
Order: NPU power first (TOPS as stated by the manufacturer), then overall speed. For local language models and Stable Diffusion the graphics card and memory size matter more — that section will come with graphics cards.
The score is our own estimate from the specs (cores, threads, clocks, architecture generation), not a benchmark run. 100 = the fastest CPU in our database.