NVIDIAAda Lovelace

Best AI Models for NVIDIA L40 (48.0GB)

VRAM:48.0 GB GDDR6·Bandwidth:864.0 GB/s·CUDA Cores:18,176·TDP:300W

With 48 GB of memory, this is a high-end configuration for local AI. You can comfortably run most open-source LLMs including large 70B parameter models at good quantization levels, making it one of the best setups for serious local AI work.

At this memory tier, nearly every popular open-source model is within reach. You can run Llama 3 70B at Q4_K_M or even Q5_K_M quantization with room to spare, handle coding assistants like DeepSeek Coder 33B at high quality, and easily run any 7B–30B model at full or near-full precision. Context windows remain generous even with larger models, so multi-turn conversations and long-document processing work smoothly.

Runs Well

  • 70B models (Llama 3 70B, Qwen 72B) at Q4–Q5
  • 30B models at Q6–Q8 quality
  • 7B–14B models at full FP16 precision
  • Vision models (LLaVA, CogVLM) without compromise

Challenging

  • Mixture-of-experts models like Mixtral 8x22B at higher quants
  • 120B+ models still require lower quantizations

What LLMs Can NVIDIA L40 Run?

Showing compatibility for NVIDIA L40

ModelVRAMGrade
28.6 GBA76
Qwen3 32B
19.8 GBB56
20.5 GBB58
20.5 GBB58
Gemma 3 27B IT
18.1 GBB53
GPT OSS 20B
13.3 GBC43
QwQ 32B
20.0 GBB57
21.4 GBB60

NVIDIA L40 Specifications

Brand
NVIDIA
Architecture
Ada Lovelace
VRAM
48.0 GB GDDR6
Memory Bandwidth
864.0 GB/s
CUDA Cores
18,176
Tensor Cores
568
FP16 Performance
362.10 TFLOPS
TDP
300W
Release Date
2022-10-13

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.

Similar GPUs for Running AI Models

Frequently Asked Questions

Can NVIDIA L40 run Llama 3 8B?

Yes, the NVIDIA L40 with 48 GB can run Llama 3 8B at Q4_K_M quantization with good performance. At this VRAM level, you can expect smooth token generation and responsive inference for chat and coding tasks.

Is NVIDIA L40 good for AI?

The NVIDIA L40 has 48 GB of GDDR6, making it excellent for running local LLM models. You can run most popular 7B-30B models at good quality.

How many parameters can NVIDIA L40 handle?

With 48 GB, the NVIDIA L40 can handle models up to approximately 30-70B parameters depending on quantization. Using Q4_K_M quantization (the typical sweet spot), you can fit roughly 80B parameters.

What quantization should I use on NVIDIA L40?

For the best balance of quality and speed on 48 GB, Q4_K_M is the recommended starting point. If you have headroom, try Q5_K_M for better quality. For larger models that barely fit, Q3_K_M or Q2_K can squeeze them in at the cost of some output quality.

How fast is NVIDIA L40 for AI inference?

Speed depends on the model size and quantization. With 864.0 GB/s memory bandwidth, the NVIDIA L40 can typically achieve 30-50+ tokens per second on 7B models at Q4_K_M quantization, which is comfortable for interactive chat.