Ada Lovelace architecture
NVIDIA L4
NVIDIA L4 packages Ada Lovelace AI and media acceleration into a 24 GB, single-slot, 72 W card — high-density video and inference throughput per watt, per rack unit.
- GPU memory
- 24 GB GDDR6
- Memory bandwidth
- 300 GB/s
- Max power
- 72 W
Specifications as published by NVIDIA; Tensor Core ratings shown are sparsity-accelerated figures.
Overview
Why L4 fits
Ada Lovelace, 72 W efficiency
L4 is engineered for density. Its low-profile single-slot form factor and 72 W power envelope make it possible to pack substantial inference and video acceleration into racks that could never host flagship GPUs. For streaming, video analytics and AI-heavy media pipelines, it is often the most cost-effective compute per watt.
It also works as a universal development target where workloads need a real GPU but not a large memory pool. Configurations are partner-sourced and confirmed per engagement.
Specifications
L4 technical specifications
Published figures for the accelerator, followed by the variant and configuration caveats that apply.
- Architecture
- NVIDIA Ada Lovelace
- GPU memory
- 24 GB GDDR6
- Memory bandwidth
- 300 GB/s
- FP8 Tensor Core
- 485 TFLOPS (with sparsity)
- FP16 / BF16 Tensor Core
- 242 TFLOPS (with sparsity)
- TF32 Tensor Core
- 120 TFLOPS (with sparsity)
- FP32
- 30.3 TFLOPS
- Media engines
- 2x NVENC / 4x NVDEC / 4x JPEG
- Interconnect
- PCIe Gen4 x16 (64 GB/s)
- Form factor
- 1-slot, low-profile
- Max TDP
- 72 W
Variant & configuration note
Single low-profile card without NVLink or MIG; server options span 1–8 cards per system. Host configuration is scoped per engagement.
Specifications describe the GPU itself as published by NVIDIA — not a fixed server configuration. Scalenta scopes complete instances (GPU count, interconnect, CPU, memory, storage) per engagement and states them in writing.
Use cases
Where L4 fits
Workload shapes this platform suits — the scoping conversation maps them to a concrete configuration.
Video processing at scale
Live streaming, transcode and content analysis with hardware AV1 encode and decode engines on every card.
AI inference density
Many parallel inference services — vision, speech, summarisation — packed into standard low-profile servers.
Virtual workstations and graphics
Cloud desktop and light visualisation workloads where power and density matter more than peak rendering throughput.
Delivery model
Partner-sourced, scoped in writing
How GPU capacity is delivered at Scalenta — the same three commitments on every model page.
Partner-sourced capacity
Accelerator capacity is sourced through verified data center and hardware partners — no facility-ownership claims, no stock promises.
Scoped per engagement
Model, variant, GPU count, interconnect and host configuration are confirmed in writing before anything is committed.
Operated with 24/7 support
Deployments are covered by 24/7 support and SLA-backed managed services scoped to your environment.
Related GPU models
Nearby options in the catalogue
NVIDIA L40S
One card for generative AI, 3D graphics, rendering and video — the pragmatic choice where workloads refuse to stay in one lane.
- GPU memory
- 48 GB GDDR6 ECC
- Memory bandwidth
- 864 GB/s
- FP8 tensor
- 1,466 TFLOPS
NVIDIA A2
Entry-level, single-slot Ampere acceleration for inference and edge workloads that do not justify flagship hardware.
- GPU memory
- 16 GB GDDR6
- Memory bandwidth
- 200 GB/s
- Max TDP
- 40–60 W
NVIDIA A30
A right-sized Ampere accelerator for teams stepping up from CPU inference or low-end GPUs into real AI capacity.
- GPU memory
- 24 GB HBM2
- Memory bandwidth
- 933 GB/s
- Max power
- 165 W
FAQ
L4 — common questions
How does L4 differ from L40S?
Can L4 run LLMs?
What does a video-processing deployment look like?
Request a GPU consultation
Talk to us about NVIDIA L4 — workload-first scoping, written findings and a proposal with clear scope. No obligation.