Skip to main content
scalenta

Ada Lovelace architecture

NVIDIA L40S

NVIDIA L40S is the universal Ada Lovelace accelerator: strong AI compute plus best-in-class graphics and media engines, in a 48 GB dual-slot PCIe card built for mixed data center workloads.

Request GPU ConsultationView specifications
GPU memory
48 GB GDDR6 ECC
Memory bandwidth
864 GB/s
FP8 tensor
1,466 TFLOPS

Specifications as published by NVIDIA; Tensor Core ratings shown are sparsity-accelerated figures.

L40SL40SADA LOVELACE ARCHITECTURE

Overview

Why L40S fits

Ada Lovelace for generative AI and graphics

Many real environments are not pure AI or pure graphics: a design team renders, a product team fine-tunes an image model, a media pipeline transcribes and captions video. L40S is built for exactly that mix, combining fourth-generation Tensor Cores with RT cores and AV1-capable media engines in a standard PCIe form factor.

It is our first recommendation where visual computing and AI coexist, and where the 350 W power envelope and passive cooling fit standard server designs more easily than flagship data center parts. Capacity is partner-sourced and configurations are confirmed per engagement.

Specifications

L40S technical specifications

Published figures for the accelerator, followed by the variant and configuration caveats that apply.

Architecture
NVIDIA Ada Lovelace
GPU memory
48 GB GDDR6 with ECC
Memory bandwidth
864 GB/s
FP8 Tensor Core
1,466 TFLOPS (with sparsity)
FP16 Tensor Core
733 TFLOPS (with sparsity)
TF32 Tensor Core
366 TFLOPS (with sparsity)
FP32
91.6 TFLOPS
RT Core performance
212 TFLOPS
Cores
18,176 CUDA · 568 Tensor · 142 RT
Interconnect
PCIe Gen4 x16 (64 GB/s bidirectional)
Form factor
Dual-slot, passive
Max power
350 W
Media engines
3x NVENC / 3x NVDEC (AV1)
Virtualisation
vGPU supported · no MIG, no NVLink

Variant & configuration note

Single card configuration without NVLink or MIG; scale is achieved across multiple cards and servers. The physical card includes 4x DisplayPort 1.4a, and server integration varies by OEM chassis. 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 L40S fits

Workload shapes this platform suits — the scoping conversation maps them to a concrete configuration.

  • Generative AI and LLM serving

    Image, video and language model inference and fine-tuning where FP8 throughput and 48 GB of memory per card matter.

  • Rendering and visualisation

    Real-time ray tracing, product design, simulation visualisation and virtual production with third-generation RT cores.

  • Video and media pipelines

    Transcode, caption and analyse large video libraries with AV1-capable encode/decode engines.

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

L4L4ADA LOVELACE ARCHITECTUREAda Lovelace

NVIDIA L4

The efficiency play: dense video processing and AI inference in a low-profile card built for many-servers, many-streams deployments.

GPU memory
24 GB GDDR6
Memory bandwidth
300 GB/s
Max power
72 W
View specifications
RTX 6000 AdaRTX 6000 AdaADA LOVELACE ARCHITECTUREAda Lovelace

NVIDIA RTX 6000 Ada

The current-generation professional graphics flagship: Ada throughput, 48 GB ECC memory, and strong AI capability in one card.

GPU memory
48 GB GDDR6 ECC
Memory bandwidth
960 GB/s
CUDA cores
18,176
View specifications
A100AMPERE ARCHITECTUREAmpere

NVIDIA A100

A proven training-and-HPC accelerator with an 80 GB memory pool per GPU and mature tooling across the AI ecosystem.

GPU memory
80 GB HBM2e
Memory bandwidth
2,039 GB/s
FP16 tensor
624 TFLOPS
View specifications
Browse all GPU instances

FAQ

L40S — common questions

When should I pick L40S over H100?
When the workload is mixed rather than purely AI training, or when graphics, rendering and video matter as much as model throughput. L40S fits standard PCIe server designs and spreads well across many users. For large-scale training, Hopper-class capacity is usually the better fit.
Can L40S handle both a design team and an AI workload?
That is its design brief. The combination of RT cores, media engines and tensor compute lets a single server profile serve rendering, video and AI tasks — with vGPU support for virtual workstations when that fits your licensing model.
How is L40S deployed at Scalenta?
As partner-sourced capacity in standard servers — usually multi-card nodes for AI density, or virtualised hosts for workstation-style use. Exact card counts, CPU and storage are confirmed per engagement before anything is committed.

Request a GPU consultation

Talk to us about NVIDIA L40S — workload-first scoping, written findings and a proposal with clear scope. No obligation.

Request Infrastructure AssessmentTalk to our team