NVIDIA L4 Tensor Core GPU: 24GB, 72W

Home
Accelerators
Nvidia
NVIDIA L4 Tensor Core

Low-Profile Inference GPU

Overview

Fits Anywhere. Accelerates Everything.

The NVIDIA L4 is the GPU you can deploy in servers where no other accelerator will fit. At just 72 watts in a single-slot, low-profile PCIe card, it draws power entirely from the motherboard slot. No auxiliary power cables. No special riser cards. No cooling upgrades. If your server has a free PCIe Gen4 x16 slot, it can run an L4.

Built on Ada Lovelace with 24GB of GDDR6, fourth-generation Tensor Cores (with FP8), and hardware AV1 encode/decode, the L4 handles AI inference, video transcoding, virtual desktops, and lightweight graphics. It delivers up to 2.5X the generative AI performance of the T4 it replaces, in the same power and form factor envelope.

The NVIDIA T4 became one of the most widely deployed data center GPUs in history. The L4 is the natural upgrade path: 2-3X better performance, same simplicity.

Running T4s Today? Here's What Changes.

The L4 is the direct successor to the NVIDIA T4. Same slot-powered, low-profile form factor. Same drop-in simplicity. But 2-3X the performance across inference, video, and graphics workloads.

	T4	L4
Architecture	Turing	Ada Lovelace
Memory	16GB GDDR6	24GB GDDR6
FP8 Tensor	N/A	485 TFLOPS
TDP	70W	72W
Form Factor	Low-profile	Low-profile
AV1 Encode	No	Yes

Compare All GPUs in Our Matrix

72W

Max TDP (Slot-Powered)

24GB

GDDR6 Memory

485

FP8 TFLOPS (Sparse)

Slot, Low-Profile

Compatible Servers

Dell PowerEdge Servers That Support the L4

The L4's single-slot, low-profile, slot-powered design fits into servers that can't accommodate full-height GPUs, including 1U and edge platforms.

Dell PowerEdge R760

Mainstream Inference

16th Gen Intel Xeon 2U server
Multi-L4 capable for production inference

View Server

Dell PowerEdge R660

1U Deployment

16th Gen Intel Xeon 1U server
L4 fits where full-height GPUs can't

View Server

Dell PowerEdge XR7620

Edge / Rugged

16th Gen edge-optimized server
Rugged, compact, ideal for edge AI

View Server

Dell PowerEdge R760xa

Dense Multi-GPU

16th Gen GPU-optimized 2U server
Pack multiple L4s for high-throughput inference

View Server

Dell PowerEdge R750

Refurbished Value

15th Gen Intel Xeon 2U workhorse
Cost-effective L4 retrofit platform

View Server

Dell PowerEdge R7625

AMD Platform

16th Gen AMD EPYC 2U server
High core count pairs well with multi-L4

View Server

View Full GPU Compatibility Matrix

Use Cases

What the L40 Does Best

Better Energy Efficiency

8x L4 vs. 2S Intel 8362 CPU server TCO comparison: end-to-end video pipeline with CV-CUDA pre- and postprocessing, decode, inference (SegFormer), encode, TRT 8.6 vs. CPU-only pipeline using OpenCV 4.7, PyTorch inference.

AI Inference at Scale

The L4 is built for production inference: recommendation engines, fraud detection, NLP, image classification. Its 72W power draw means you can pack more GPUs per rack without upgrading power distribution or cooling. For organizations running hundreds of inference endpoints, the L4 delivers the best performance per watt of any current-gen NVIDIA data center GPU.

Up to 120X Higher AI Video Performance

Measured performance: 8x L4 vs 2S Intel 8362 CPU server comparison, end-to-end video pipeline with CV-CUDA® decode, preprocessing, inference (SegFormer), postprocessing, encode, NVIDIA® TensorRT™ 8.6 vs CPU-only pipeline using OpenCV 4.7, PyTorch inference.

AI Video Pipelines

Servers equipped with L4 GPUs can host over 1,000 concurrent AV1 video streams at 720p30. For broadcast, live streaming, video surveillance, and content moderation, the L4 replaces dedicated encoding hardware with GPU-accelerated pipelines that also run AI inference on the same video frames in real time.

Virtual Desktops & Cloud Gaming

With vGPU software support, the L4 powers virtual PC sessions and lightweight cloud gaming instances. Its low power draw and small form factor make it ideal for VDI deployments in space-constrained environments like branch offices, retail locations, and edge data centers where full-size GPUs aren't an option.

Edge AI & Compact Deployments

The L4's low-profile, slot-powered design was built for edge scenarios. It fits into short-depth, 1U, and ruggedized servers that can't accommodate dual-slot GPUs. For smart retail, industrial inspection, autonomous systems, and telecom edge workloads, the L4 brings data center-class AI to locations where power and space are at a premium.

Specifications