NVIDIA Tesla P40 vs NVIDIA GeForce RTX 4090
For running models locally, priced in Canada. At 8k context the first runs 28 models fully in memory, the second 28.
Specifications
| Spec | NVIDIA Tesla P40 | NVIDIA GeForce RTX 4090 |
|---|---|---|
| GPU | Tesla P40 | GeForce RTX 4090 |
| Memory | 24 GB GDDR5 | 24 GB GDDR6X |
| Memory bus | 384-bit | 384-bit |
| Memory bandwidth | 346 GB/s | 1,008 GB/s |
| FP16 compute | 11.8 TFLOPS | 165 TFLOPS |
| Software | CUDA | CUDA |
| Board power | 250 W | 450 W |
| PCIe generation | Gen 3 | Gen 4 |
| PCIe lanes | x16 | x16 |
What each runs
Estimated decode speed with the most precise quant that fits each one est.
| Model | NVIDIA Tesla P40 | NVIDIA GeForce RTX 4090 |
|---|---|---|
| Qwen 3.8 27B | 10tok/sUD-Q6_K | 28tok/sUD-Q6_K |
| Qwen 3.6 27B | 9tok/sQ6_K | 27tok/sQ6_K |
| Gemma 4 31B | 9tok/sQ5_K_M | 27tok/sQ5_K_M |
| Qwen 3.6 35B-A3B | 62tok/sUD-Q4_K_M | 181tok/sUD-Q4_K_M |
| Qwen 3.5 27B | 9tok/sQ6_K | 27tok/sQ6_K |
| Gemma 4 26B-A4B | 32tok/sUD-Q6_K | 95tok/sUD-Q6_K |
| Qwen 3.5 35B-A3B | 62tok/sQ4_K_M | 181tok/sQ4_K_M |
| GLM 4.7 Flash | 32tok/sQ5_K_M | 94tok/sQ5_K_M |
| Devstral Small 2 24B | 10tok/sQ6_K | 30tok/sQ6_K |
| gpt-oss 20B | 49tok/sMXFP4 | 142tok/sMXFP4 |
| Qwen3 Coder 30B-A3B | 44tok/sQ5_K_M | 128tok/sQ5_K_M |
| Qwen3 32B | 10tok/sQ4_K_M | 29tok/sQ4_K_M |
| Gemma 4 12B | 16tok/sQ8_0 | 48tok/sQ8_0 |
| Qwen 3.5 9B | 22tok/sQ8_0 | 64tok/sQ8_0 |
| Mistral Small 3.2 24B | 10tok/sQ6_K | 30tok/sQ6_K |
| DeepSeek R1 Distill Qwen 32B | 10tok/sQ4_K_M | 28tok/sQ4_K_M |
| Gemma 3 27B | 9tok/sQ6_K | 27tok/sQ6_K |
| Ministral 3 14B | 14tok/sQ8_0 | 40tok/sQ8_0 |
| Qwen3 14B | 13tok/sQ8_0 | 37tok/sQ8_0 |
| DeepSeek R1 0528 Qwen3 8B | 22tok/sQ8_0 | 63tok/sQ8_0 |
| Qwen3 8B | 22tok/sQ8_0 | 63tok/sQ8_0 |
| Phi-4 14B | 12tok/sQ8_0 | 36tok/sQ8_0 |
| Qwen 3.5 4B | 45tok/sQ8_0 | 130tok/sQ8_0 |
| Ministral 3 8B | 21tok/sQ8_0 | 62tok/sQ8_0 |
| Granite 4.2 8B | 20tok/sQ8_0 | 59tok/sQ8_0 |
| Gemma 4 E4B | 25tok/sQ8_0 | 74tok/sQ8_0 |
| LFM2.5 8B-A1B | 88tok/sQ8_0 | 256tok/sQ8_0 |
| Llama 3.1 8B | 28tok/sQ6_K | 82tok/sQ6_K |
| Qwen 3.5 122B-A10B | offload | offload |
| Qwen3 Coder Next | offload | offload |
| Mistral Small 4 119B | offload | offload |
| GLM 4.5 Air | offload | offload |
| Qwen3 Next 80B-A3B | offload | offload |
| Nemotron 3.5 Lightning 30B-A3B | offload | offload |
| Kimi Linear 48B-A3B | offload | offload |
| Llama 3.3 70B | offload | offload |
| DeepSeek V4 Flash | won't run | won't run |
| GLM 5.3 Flash | won't run | won't run |
| Qwen 3.8 Flash Next | won't run | won't run |
| MiniMax M2.7 | won't run | won't run |
| Nemotron 3 Super 120B-A12B | won't run | won't run |
| gpt-oss 120B | won't run | won't run |