NVIDIAMaxwell

Best AI Models for NVIDIA Tesla M40 24GB (24.0GB)

VRAM:24.0 GB GDDR5·Bandwidth:288.0 GB/s·CUDA Cores:3,072·TDP:250W

Maxwell (2015): no Tensor Cores, slow FP16, GDDR5 — the cheapest 24 GB card but also the slowest here. Passive server card; needs added cooling.

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 M40 24GB Run?

123 models · 16 excellent · 41 good

Showing compatibility for NVIDIA Tesla M40 24GB

LLM models compatible with NVIDIA Tesla M40 24GB — ranked by performance
ModelVRAMGrade
Qwen3.6 27B27.8B
Q4_K_M·10.7 t/s tok/s·262K ctx·DECENT
17.4 GBB49
Gemma 3 27B IT27.4B
Q4_K_M·10.3 t/s tok/s·131K ctx·DECENT
18.1 GBB49
Q4_K_M·11.3 t/s tok/s·262K ctx·DECENT
16.6 GBB50
Q4_K_M·10.0 t/s tok/s·262K ctx·DECENT
18.7 GBB48
Q4_K_M·10.0 t/s tok/s·262K ctx·DECENT
18.7 GBB48
Q4_K_M·10.4 t/s tok/s·8K ctx·DECENT
18.0 GBB49
Q4_K_M·11.6 t/s tok/s·262K ctx·DECENT
16.1 GBB50
Hy MT2 30B A3B30.1B
Q4_K_M·10.2 t/s tok/s·262K ctx·DECENT
18.4 GBB48
Q4_K_M·10.0 t/s tok/s·262K ctx·DECENT
18.7 GBB48
North Mini Code 1.030.5B
Q4_K_M·10.0 t/s tok/s·500K ctx·DECENT
18.7 GBB48
Q4_K_M·12.4 t/s tok/s·131K ctx·DECENT
15.1 GBB51
Tmax 27B26.9B
Q4_K_M·11.1 t/s tok/s·262K ctx·DECENT
16.9 GBB49
BF16·12.2 t/s tok/s·4K ctx·DECENT
15.4 GBB51
BF16·12.2 t/s tok/s·8K ctx·DECENT
15.3 GBB51
Q4_K_M·12.6 t/s tok/s·33K ctx·DECENT
14.9 GBB51
GPT OSS 20B21.5B
Q4_K_M·14.1 t/s tok/s·131K ctx·DECENT
13.3 GBB53

NVIDIA Tesla M40 24GB Specifications

Brand
NVIDIA
Architecture
Maxwell
Compute Capability
5.2 (CUDA SM version)
VRAM
24.0 GB GDDR5
Memory Bandwidth
288.0 GB/s
CUDA Cores
3,072
Tensor Cores
0
TDP
250W
Release Date
2015-11-10

Get Started

Ollama (Recommended)

$curl -fsSL https://ollama.com/install.sh | sh
$ollama run llama3:8b

LM Studio

LM Studio

Download LM Studio, search for a model, and run it with one click.

Prompt Processing

Estimated for Mellum2 12B A2.5B Thinking SFT, the compute-bound phase that reads your prompt before the first reply token appears.

614.7tok/s prefill

Short chat

833 ms

512 tok prompt

Long chat

6.7 s

4,096 tok prompt

Document / codebase

53.3 s

32,768 tok prompt

Prefill is compute-bound and a different number from the decode tok/s shown elsewhere on this page — how prompt processing works →

Efficiency & Value

Based on Mellum2 12B A2.5B Thinking SFT at ~24.4 tok/s decode.

Tokens per watt

0.10tok/s per W

Higher is better.

How efficiency & value are calculated →

Performance figures are estimates calibrated as of 2026-07-30 see calibration basis →

GPUs to Consider Over NVIDIA Tesla M40 24GB

Similar GPUs and upgrades with more VRAM or higher bandwidth for AI

Frequently Asked Questions

Can NVIDIA Tesla M40 24GB run Qwen3.6 27B?

Yes, the NVIDIA Tesla M40 24GB with 24 GB can run Qwen3.6 27B, Gemma 3 27B IT, Gemma 4 26B A4B IT, and 1520 other models. 447 models run at excellent quality, and 528 at good quality. Check the compatibility table above for the full list with VRAM usage and estimated speed.

Is NVIDIA Tesla M40 24GB good for AI?

The NVIDIA Tesla M40 24GB has 24 GB of GDDR5, making it excellent for running local AI models. It supports 975 models at good quality or better. With 288.0 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 M40 24GB handle?

With 24 GB, the NVIDIA Tesla M40 24GB 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 M40 24GB?

For the best balance of quality and speed on the NVIDIA Tesla M40 24GB, 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 M40 24GB for AI inference?

With 288.0 GB/s memory bandwidth, the NVIDIA Tesla M40 24GB achieves approximately 42 tokens/sec on a 7B model at Q4_K_M — that's very fast, well above conversational speed. A 14B model runs at ~21 tok/s. Token generation speed scales inversely with model size — smaller models are significantly faster.

tok/s = (288 GB/s ÷ model GB) × efficiency

Smaller models = faster inference. Memory bandwidth is the main bottleneck for token generation speed.

Estimated speed on NVIDIA Tesla M40 24GB

Real-world results typically within ±20%. Speed depends on quantization kernel, batch size, and software stack.

Learn more about tok/s estimation →

What's the best model for NVIDIA Tesla M40 24GB?

The top-rated models for the NVIDIA Tesla M40 24GB are Qwen3.6 27B, Gemma 3 27B IT, 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 M40 24GB need?

The NVIDIA Tesla M40 24GB 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 M40 24GB?

Maxwell (2015): no Tensor Cores, slow FP16, GDDR5 — the cheapest 24 GB card but also the slowest here. Passive server card; needs added cooling.