Infrastructure atlas
NVIDIA/Ada Lovelace

L40S

A PCIe data-center GPU for generative AI inference, rendering, video, and multimodal workloads without NVLink or MIG.

GPUAdainferencevideorendering
L40S hardware

01 / Memory

48 GB GDDR6

ECC enabled

02 / Bandwidth

864 GB/s

Memory bandwidth

03 / Power

350 W

Maximum consumption

Precision & peak compute

PrecisionPublished peakBasis
FP8 Tensor733 TFLOPSDense
FP8 Tensor1,466 TFLOPSSparse
FP16 / BF16 Tensor362 TFLOPSDense
INT8 Tensor733 TOPSDense
FP3291.6 TFLOPSDense

Peak FLOPS and TOPS are theoretical hardware figures. Sparse and dense modes, chip and system totals, and form factors are not interchangeable.

NVIDIA L40S specifications

Technical specification

Silicon & memory
Architecture
Ada Lovelace
Compute
18,176 CUDA cores · 568 Tensor Cores
FP8 Tensor
733 TFLOPS dense · 1,466 sparse
Video
3 NVENC · 3 NVDEC with AV1
System & software
Form factor
Dual-slot PCIe
Interconnect
PCIe Gen4 x16
Partitioning
No MIG; vGPU support
Best fit
Inference, graphics, video, multimodal