Best AI Models for NVIDIA Tesla P40 (24.0GB)
Pascal: no Tensor Cores and 1:64-rate FP16 — fine for GGUF/INT8 inference, poor for FP16/exllama. Passive server card; needs added cooling in a desktop.
24 GB is the enthusiast tier for running AI models locally. It comfortably handles 7B–13B models at high quality and opens the door to larger 30B models at moderate quantization.
This is one of the most popular memory tiers for local AI, found in GPUs like the RTX 4090 and RTX 3090. You can run Llama 3 8B, Mistral 7B, and Qwen 2.5 7B at Q5_K_M or Q6_K quality with fast token generation and generous context windows. Larger 14B models like DeepSeek R1 Distill fit comfortably at Q4_K_M. For even bigger models, 30B class runs at Q2–Q3, but 70B models are generally too heavy for single-GPU inference at this tier.
Runs Well
- 7B models (Llama 3 8B, Mistral 7B) at Q5–Q8 quality
- 13B–14B models at Q4–Q5 quality
- Small models (3B–4B) at FP16 precision
- Multimodal models like LLaVA 7B
Challenging
- 30B models only at Q2–Q3 quantization
- 70B models do not fit in VRAM
- Large context windows with 14B+ models
What LLMs Can NVIDIA Tesla P40 Run?
105 models · 7 excellent · 15 good
Showing compatibility for NVIDIA Tesla P40
| Model | Quant | VRAM | Speed | Context | Status | Grade |
|---|---|---|---|---|---|---|
Q4_K_M·45.7 t/s tok/s·33K ctx·EASY RUN | Q4_K_M | 4.9 GB | 45.7 t/s | 33K | EASY RUN | C36 |
Q4_K_M·77.6 t/s tok/s·41K ctx·EASY RUN | Q4_K_M | 2.9 GB | 77.6 t/s | 41K | EASY RUN | C31 |
Q4_K_M·26.2 t/s tok/s·FAIR FIT | Q4_K_M | 8.6 GB | 26.2 t/s | — | FAIR FIT | B51 |
Q4_K_M·10.8 t/s tok/s·16K ctx·FAIR FIT | Q4_K_M | 20.8 GB | 10.8 t/s | 16K | FAIR FIT | B63 |
Q4_K_M·23.6 t/s tok/s·33K ctx·FAIR FIT | Q4_K_M | 9.5 GB | 23.6 t/s | 33K | FAIR FIT | B55 |
Q4_K_M·65.6 t/s tok/s·131K ctx·EASY RUN | Q4_K_M | 3.4 GB | 65.6 t/s | 131K | EASY RUN | C32 |
Q4_K_M·77.6 t/s tok/s·262K ctx·EASY RUN | Q4_K_M | 2.9 GB | 77.6 t/s | 262K | EASY RUN | C31 |
Q4_K_M·56.8 t/s tok/s·33K ctx·EASY RUN | Q4_K_M | 4.0 GB | 56.8 t/s | 33K | EASY RUN | C34 |
Q4_K_M·26.2 t/s tok/s·FAIR FIT | Q4_K_M | 8.6 GB | 26.2 t/s | — | FAIR FIT | B51 |
Q4_K_M·45.1 t/s tok/s·131K ctx·EASY RUN | Q4_K_M | 5.0 GB | 45.1 t/s | 131K | EASY RUN | C36 |
Q4_K_M·79.2 t/s tok/s·EASY RUN | Q4_K_M | 2.8 GB | 79.2 t/s | — | EASY RUN | C31 |
Q4_K_M·62.6 t/s tok/s·EASY RUN | Q4_K_M | 3.6 GB | 62.6 t/s | — | EASY RUN | C33 |
Q4_K_M·36.9 t/s tok/s·8K ctx·EASY RUN | Q4_K_M | 6.1 GB | 36.9 t/s | 8K | EASY RUN | C40 |
Q4_K_M·40.7 t/s tok/s·131K ctx·EASY RUN | Q4_K_M | 5.5 GB | 40.7 t/s | 131K | EASY RUN | C38 |
Q4_K_M·26.2 t/s tok/s·FAIR FIT | Q4_K_M | 8.6 GB | 26.2 t/s | — | FAIR FIT | B51 |
Q4_K_M·41.7 t/s tok/s·131K ctx·EASY RUN | Q4_K_M | 5.4 GB | 41.7 t/s | 131K | EASY RUN | C37 |
NVIDIA Tesla P40 Specifications
- Brand
- NVIDIA
- Architecture
- Pascal
- Compute Capability
- 6.1 (CUDA SM version)
- VRAM
- 24.0 GB GDDR5
- Memory Bandwidth
- 346.0 GB/s
- CUDA Cores
- 3,840
- Stream Processors
- 3,840
- Tensor Cores
- 0
- FP16 Performance
- 0.18 TFLOPS
- TDP
- 250W
- Release Date
- 2016-09-13
Get Started
GPUs to Consider Over NVIDIA Tesla P40
Similar GPUs and upgrades with more VRAM or higher bandwidth for AI
NVIDIA GeForce RTX 5090
NVIDIA · Blackwell
NVIDIA GeForce RTX 3090 Ti
NVIDIA · Ampere
NVIDIA GeForce RTX 4090
NVIDIA · Ada Lovelace
AMD Radeon RX 7900 XTX
AMD · RDNA 3
NVIDIA GeForce RTX 3090
NVIDIA · Ampere
NVIDIA GeForce RTX 5090 Laptop GPU
NVIDIA · Blackwell
Frequently Asked Questions
- Can NVIDIA Tesla P40 run Gemma 3 27B IT?
Yes, the NVIDIA Tesla P40 with 24 GB can run Gemma 3 27B IT, Qwen3.6 27B, Gemma 4 26B A4B IT, and 1264 other models. 69 models run at excellent quality, and 153 at good quality. Check the compatibility table above for the full list with VRAM usage and estimated speed.
- Is NVIDIA Tesla P40 good for AI?
The NVIDIA Tesla P40 has 24 GB of GDDR5, making it excellent for running local AI models. It supports 222 models at good quality or better. With 346.0 GB/s memory bandwidth, it delivers reasonable token generation speeds. This is an enthusiast-grade GPU that handles most popular open-source LLMs.
- How many parameters can NVIDIA Tesla P40 handle?
With 24 GB, the NVIDIA Tesla P40 supports models from 3B to 30B parameters depending on quantization level. At Q4_K_M (the recommended sweet spot), you can fit roughly 40B parameters. This means 7B models at high quality (Q6/Q8) or 30B+ models at Q4.
- What quantization should I use on NVIDIA Tesla P40?
For the best balance of quality and speed on the NVIDIA Tesla P40, start with Q4_K_M — it preserves ~85% of the original model quality while keeping VRAM usage reasonable. With 24+ GB, you have the headroom to run 7B models at Q5_K_M or even Q6_K for noticeably better output quality. For larger 30B models, Q4_K_M remains the sweet spot.
- How fast is NVIDIA Tesla P40 for AI inference?
With 346.0 GB/s memory bandwidth, the NVIDIA Tesla P40 achieves approximately 50 tokens/sec on a 7B model at Q4_K_M — that's very fast, well above conversational speed. A 14B model runs at ~25 tok/s. Token generation speed scales inversely with model size — smaller models are significantly faster.
tok/s = (346 GB/s ÷ model GB) × efficiency
Smaller models = faster inference. Memory bandwidth is the main bottleneck for token generation speed.
Estimated speed on NVIDIA Tesla P40
~12 tok/s~13 tok/s~14 tok/s~13 tok/sReal-world results typically within ±20%. Speed depends on quantization kernel, batch size, and software stack.
- What's the best model for NVIDIA Tesla P40?
The top-rated models for the NVIDIA Tesla P40 are Gemma 3 27B IT, Qwen3.6 27B, Gemma 4 26B A4B IT. The best choice depends on your use case: coding assistants benefit from code-tuned models, while general chat works well with instruction-tuned models like Llama or Qwen.
- What power supply and cooling does NVIDIA Tesla P40 need?
The NVIDIA Tesla P40 has a TDP of 250 W. A good rule of thumb is to provide at least double the GPU's TDP to cover the rest of the system — that means a 550 W PSU or larger. A mid-tower case with one intake and one rear exhaust is usually sufficient. Keep dust filters clean, as sustained inference generates continuous heat rather than the brief spikes typical of gaming.
- Anything to watch out for with NVIDIA Tesla P40?
Pascal: no Tensor Cores and 1:64-rate FP16 — fine for GGUF/INT8 inference, poor for FP16/exllama. Passive server card; needs added cooling in a desktop.