Nvidia L4
Low-profile data centre GPU for AI and video
The Nvidia L4 is a single-slot graphics processor built on the Ada Lovelace architecture. It targets artificial intelligence inference, video processing and virtual workstations in servers from the edge to the cloud.
- Uses Ada Lovelace architecture with 24 GB memory
- Fits in low-profile server slots without extra power
- Delivers up to 120x better AI video performance than CPUs
- Supports 1,040 concurrent AV1 video streams at 720p
- Operates in a 72 W passive cooling envelope
What it is
The Nvidia L4 is a graphics processing unit designed for data centre servers. It uses the Ada Lovelace architecture to handle artificial intelligence tasks, video transcoding and graphical rendering. The card fits into standard server slots without needing extra power cables. It replaces the older Tesla T4 model, offering more memory and better efficiency for modern workloads like generative AI and real-time video streaming.
How it came about
This chip serves as the successor to the widely used Nvidia Tesla T4, which launched in 2019. The T4 established a popular form factor for server acceleration. Nvidia updated this design with the L4 to address growing demands for energy-efficient computing in cloud and edge environments. The new card retains the physical compatibility of its predecessor while adding support for newer software standards and higher performance metrics.
Inside the chip
The L4 relies on fourth-generation Tensor Cores for machine learning calculations. It includes third-generation RT Cores to accelerate ray tracing for virtual workstations. The processor holds 24 gigabytes of memory, which is fifty percent larger than the previous generation. This extra capacity allows for bigger image generation models. The card uses passive cooling, drawing air from the server chassis rather than using its own fans. It connects via the PCIe slot and does not require external power connectors.
How it was made
Nvidia designs the L4 but the specific manufacturing partner and process node are not detailed in the provided text. The physical card is a low-profile, half-height unit with a metal backplate for protection during service. Its compact size allows it to fit into dense server configurations. The thermal solution is improved over the T4, relying on system airflow to keep the chip within its 72-watt power envelope.
Where it went
Servers equipped with the L4 handle video pipelines, including decoding and encoding AV1 streams. It supports up to 1,040 concurrent mobile video streams at 720p. The chip accelerates generative AI inference for text and images. It also powers virtual desktops and cloud gaming through real-time rendering. Enterprises use it for content moderation, search relevance and smart-space analytics. The card runs certified AI software suites for streamlined deployment.
Versions and family
The L4 is the direct follow-up to the Tesla T4. While the T4 was branded under the Tesla line, the L4 drops that name. Both cards share a similar physical footprint, though the L4 has slightly different ribbing and mounting holes. The L4 offers double the memory of the T4 and supports newer features like DLSS 3.
Why it matters
The card aims to reduce the carbon footprint of data centres by replacing CPU-based video processing. Nvidia claims switching to L4s in a two-megawatt facility saves enough energy to power nearly 2,000 homes for a year. It enables smaller servers to handle heavy AI loads, reducing rack space requirements. The card makes high-memory AI inference accessible in compact form factors like tiny mini micro nodes.
Problems and reception
Reviewers note the card is not the fastest available but expect it to be popular due to its proven format. The lack of display outputs is intentional for dense server packing. Some hobbyists find the price high for home use but value the memory capacity for local machine learning experiments. Cooling outside of a proper server chassis is cited as a challenge for non-standard deployments.
Questions people ask about Nvidia L4
What is the main purpose of the Nvidia L4?
The L4 is a data centre GPU designed for AI inference, video transcoding and virtual workstations. It handles tasks like generative AI, real-time rendering and video analytics in servers from the edge to the cloud.
How much memory does the L4 have?
The card comes with 24 gigabytes of GPU memory. This is fifty percent more than the previous Tesla T4 generation, allowing for larger image generation models and more complex data processing.
Does the L4 need a separate power cable?
No, the L4 is powered entirely through the PCIe slot. It operates within a 72-watt power envelope, which eliminates the need for external power connectors and simplifies installation in various server types.
Can the L4 be used for gaming?
The L4 is designed for servers and lacks video output ports. While it supports ray tracing and rendering for cloud gaming and virtual workstations, it is not intended for direct connection to a monitor in a personal computer.
Good reading and sources
There is no English Wikipedia article on Nvidia L4 yet, so the facts on this page come from the sources listed above: nvidia.com/en-us/data-center/l4/, servethehome.com/nvidia-l4-review-the-versatile-ai-inference-card-pny/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.