NVIDIA Data Center GPU L40S
The NVIDIA L40S is a versatile enterprise PCIe data center accelerator powered by the Ada Lovelace architecture, engineered as a universal workhorse for generative AI model fine-tuning, high-throughput LLM inference, 3D graphics rendering, and NVIDIA Omniverse metaverse workloads. Equipped with 48GB of high-speed GDDR6 memory with ECC and an 864 GB/s memory bandwidth, it features 18,176 CUDA cores, 568 4th-generation Tensor Cores with FP8 support, and 142 3rd-generation RT Cores within a 350W thermal envelope. Because it does not rely on HBM supply chains, it offers high availability and rapid cluster deployment in standard enterprise PCIe servers. Used buyers should inspect passive cooling airflow and 16-pin power connections.
Also written as: NVIDIA L40S PCIe, L40S 48GB, NVIDIA L40S Server GPU
Specifications
| Brand | NVIDIA |
|---|---|
| Family | Data Center GPU |
| Form factor | PCIe 4.0 dual-slot full-height full-length |
| Released | Aug 2023 |
| Gpu memory (GB) | 48 |
| Tdp (W) | 350 |
Generations
- Previous generation: NVIDIA Data Center GPU A40
Reference prices (last 14 days, updated 2026-10-27)
New: USD 10,500 (1 sighting, Oct 2026)
Used: USD 7,900 (1 sighting, Oct 2026)
All sightings
| Seller | Condition | Price | Note | Seen |
|---|---|---|---|---|
| cdw.com | new | USD 10,500 | New retail dual-slot PCIe card in factory packaging | 2026-10-27 |
| ebay.com | used | USD 7,900 | Tested working pull from cloud inference cluster | 2026-10-27 |
Prices seen on public listings on the dates shown, before tax and shipping. Not an offer.
Equivalent models from other brands
Buying a used L40S: what to check
Inspect the 16-pin 12V-2x6 power connector for heat stress, verify passive server chassis airflow (300 CFM required), test FP8 Tensor Core throughput via NVIDIA DCGM, and ensure VBIOS is updated to current enterprise branches.
FAQ
What makes the L40S popular for generative AI workloads?
It features 568 4th-Gen Tensor Cores with FP8 Transformer Engine support and 48GB of GDDR6 memory, offering immense compute throughput without HBM supply constraints.
Can the L40S be used for graphical rendering and VDI?
Yes, with 142 RT Cores and NVIDIA RTX vWS support, it is fully optimized for Omniverse, VFX rendering, and virtual workstations.
Does the L40S support NVLink bridges?
No, the L40S connects purely over high-speed PCIe 4.0 x16 without physical NVLink finger connectors.
Buying or selling L40S?
New, refurbished or used. Tell us the quantity and condition and we connect you with a vetted reseller in your region, usually within one working day.
Sources
- NVIDIA L40S Data Center GPU Overview, checked 2026-10-27