RTX 4000 Ada vs Tesla P40 for local AI
The Tesla P40 fits more: up to Qwen3 32B, against Gemma 3 27B on the RTX 4000 Ada. They generate at about the same speed, since their memory bandwidth is close (360 and 346 GB/s).
Memory and bandwidth
Memory decides which models fit. Bandwidth decides how fast they generate, because every token reads the model's active weights once.
| RTX 4000 Ada | Tesla P40 | |
|---|---|---|
| Memory | 20 GB VRAM | 24 GB VRAM |
| Bandwidth | 360 GB/s | 346 GB/s |
| Biggest model that fits | Gemma 3 27B | Qwen3 32B |
| Best everyday pick | gpt-oss 20B, ~63 tok/s | Qwen3 30B-A3B, ~61 tok/s |
Every model, side by side
Estimated generation speed at the largest memory size each was sold in (20 GB and 24 GB).
| Model | RTX 4000 Ada | Tesla P40 |
|---|---|---|
| Qwen3 4B About 4 GB | ~79 tok/s Runs great | ~76 tok/s Runs great |
| Qwen3 8B About 7 GB | ~38 tok/s Runs great | ~37 tok/s Runs great |
| Gemma 3 12B About 10 GB | ~24 tok/s Usable | ~23 tok/s Usable |
| Qwen3 14B About 11 GB | ~21 tok/s Usable | ~20 tok/s Usable |
| gpt-oss 20B About 16 GB | ~63 tok/s Runs great | ~61 tok/s Runs great |
| Gemma 3 27B About 19 GB | ~12 tok/s Usable | ~11 tok/s Usable |
| Qwen3 30B-A3B About 21 GB | Needs 21 GB | ~61 tok/s Runs great |
| Qwen3 32B About 22 GB | Needs 22 GB | ~10 tok/s Slow |
| Llama 3.3 70B About 45 GB | Needs 45 GB | Needs 45 GB |
| gpt-oss 120B About 67 GB | Needs 67 GB | Needs 67 GB |
| Qwen3 235B-A22B About 144 GB | Needs 144 GB | Needs 144 GB |
| DeepSeek R1 671B About 406 GB | Needs 406 GB | Needs 406 GB |
It depends. The Tesla P40 fits bigger models; the RTX 4000 Ada runs the models both fit faster. Price matters as much as either number: check used prices on the builds page, where every parts list has a measured or estimated speed.
Both columns use the same formula: memory bandwidth divided by the weights read per token, calibrated on our own rack, with Ollama's default 4-bit sizes. Real results are usually within about a third. Measured either one? Send us the number.