NVIDIA L4 24GB GDDR6 graphics card — an energy-efficient accelerator for AI and media

NVIDIA L4 24GB is a specialised GPU based on the Ada Lovelace architecture architecture (AD104 chip, 4 nm), designed for data centres, cloud services, and compact edge servers. It combines energy efficiency, modern memory, and hardware acceleration of AI tasks, making it an excellent choice for deploying generative models, video processing, and virtualisation.

Key features of NVIDIA L4 24GB

• CUDA cores: 7680
• Tensor Cores (Gen4): 240
• RT cores (Gen3): 60
• Frequencies: 795 MHz (base) / 2040 MHz (boost)
• Memory: 24 GB GDDR6, 192-bit bus, 300 GB/s bandwidth
• PCIe: Gen 4.0 x16
• Media engines: 3× NVENC (9th generation) + 3× NVDEC (6th generation), AV1 and JPEG support
• Power consumption: 72 W (passive cooling)
• Form factor: Low-profile, single-slot card, length 169 mm, weight 0.49 kg
• Outputs: None (server accelerator)
• Software support: vGPU, NVIDIA AI Enterprise

Performance:

• • FP32: 30.3 TFLOPS
  • FP16/BF16: 242 TFLOPS
  • TF32: 120 TFLOPS
  • FP8/INT8: 485 TOPS

How NVIDIA L4 differs from other solutions

When choosing a server accelerator, many buyers look towards L40 or L40S, but it is important to understand their key differences.

• Performance.
  The L40S delivers up to 1466 TFLOPS FP8, making it ideal for inference and training large LLM models. The L4 delivers around 485 TFLOPS FP8, which is less in terms of raw power, but still more than enough for medium-sized models (32B–70B parameters) and real-time generative AI.
• Power consumption.
  The L40 requires up to 350 W, while the L4 is limited to 72 W. This means that dozens of L4s can be installed in a single server or rack without overloading the power supply and cooling system.
• Form factor and installation.
  The L40 is a full-size card, while the L4 is designed in a compact, low-profile form factor with passive cooling. This solution is for those who build edge servers, compact AI nodes, and distributed data centres.
• Purpose.
  The L40 is suitable for heavy-duty clusters where maximum performance is important. The L4 is the choice for companies that value scalability and energy efficiency, as well as fast video stream processing, medium LLM inference, and virtualisation.

What tasks is NVIDIA L4 24GB suitable for?

• AI inference: running medium-sized generative models (e.g., Llama, Mistral, Qwen) in cloud and edge nodes.
• Video analytics and media streams: Simultaneous processing of dozens of video streams thanks to three NVENC/NVDEC and AV1 support.
• Virtualisation: Ability to use vGPU for multi-user work in corporate infrastructures.
• Compact servers and edge solutions: low TDP (72 W) and passive cooling allow L4 to be installed in nodes with limited power and ventilation.

Why choose OsoDoso-Store

• Direct deliveries from NVIDIA, verified partners.
• 3-year warranty and original software support.
• Any form of payment. Card, bank transfer (w/ex VAT), USDT cryptocurrency.
• Consultation on configuration selection and integration for your tasks.

Place your order for NVIDIA L4 24GB today at OsoDoso-Store!