NVIDIA L40 48GB GDDR6 — Ada Lovelace PCIe Gen4 Data Center GPU Accelerator
Technical Identity
This listing covers an NVIDIA L40 48GB GDDR6 PCIe GPU accelerator whose unit label reads “L40 CNX” (SKU L40-CNX). “CNX” is not a documented NVIDIA L40 variant. NVIDIA’s published converged-accelerator line consists of the A100X and A30X (GPU plus BlueField-2 DPU) and the H100 CNX (H100 plus ConnectX-7 SmartNIC); no public NVIDIA documentation describes an L40 CNX. T.E.S IT-SOLUTIONS makes no claim of an on-board ConnectX network interface on this unit. The exact model designation must be confirmed from the unit label and regulatory marking before ordering. Everything specified below is the verified NVIDIA L40 as published in the NVIDIA L40 Datasheet. Condition: Refurbished.
Specifications
Brand: NVIDIA
Model / MPN: NVIDIA L40 (unit label reads “L40 CNX”; SKU L40-CNX)
NVIDIA ordering P/N: not stated — confirm from unit label
Product type: Data center GPU accelerator
GPU architecture: NVIDIA Ada Lovelace (AD102-class GPU)
Cores: 18,176 CUDA; 568 fourth-generation Tensor; 142 third-generation RT
GPU memory: 48GB GDDR6 with ECC at 864GB/s
Interconnect: PCIe Gen4 x16, 64GB/s bi-directional
FP32: 90.5 TFLOPS; RT Core: 209 TFLOPS
Tensor Core (dense | with sparsity): TF32 90.5 | 181 TFLOPS; BFLOAT16 and FP16 181.05 | 362.1 TFLOPS; FP8 and INT8 362 | 724; INT4 724 | 1448 TOPS
Form factor: 4.4in (H) x 10.5in (L), dual slot
Display ports: 4x DisplayPort 1.4a
Max power / connector: 300W / 16-pin
Thermal: Passive
vGPU support: Yes; NVENC / NVDEC: 3x / 3x (includes AV1 encode and decode)
Secure Boot with Root of Trust: Yes; NEBS Ready: Level 3
MIG support: No; NVLink support: No
Condition: Refurbished
NVIDIA marks the L40 datasheet figures as preliminary specifications, subject to change.
Engineering Context
The L40 is the Ada Lovelace generation’s combined graphics-and-compute data center card: third-generation RT Cores and fourth-generation Tensor Cores with FP8, paired with 48GB GDDR6 with ECC at 864GB/s rather than HBM. Four DisplayPort 1.4a outputs and three NVENC / three NVDEC engines with AV1 place it in rendering, virtual-workstation and streaming roles as well as inference. With no NVLink and no MIG, scale-out is over PCIe and the network.
Use Cases
Rendering, 3D graphics and NVIDIA Omniverse Enterprise workloads.
Virtual workstations with NVIDIA RTX vWS software (licensed separately).
AI inference and data science on a 48GB memory footprint.
Streaming and video pipelines using the AV1-capable NVENC / NVDEC engines.
What This Is NOT
NOT a documented NVIDIA converged accelerator. NVIDIA documents the A100X, A30X and H100 CNX only; there is no public NVIDIA documentation for an “L40 CNX”.
NOT claimed to carry an on-board ConnectX NIC or BlueField DPU. Confirm from the unit label and regulatory marking.
NOT an L40S. The L40S is a separate NVIDIA SKU with different published figures.
NOT MIG-capable and NOT NVLink-capable. The datasheet lists MIG Support: No and NVLink Support: No.
NOT an HBM part. Memory is 48GB GDDR6 with ECC at 864GB/s — not HBM2e or HBM3.
NOT actively cooled. The thermal solution is passive; the chassis must supply directed airflow.
NOT supplied with power cables or adapters, display cables, brackets or software licences.
Compatibility & Hard Constraints
Identity verification is mandatory. Confirm the model designation from the unit label and regulatory marking before ordering — “CNX” matches no documented NVIDIA L40 variant.
Typical platforms: OEM and NVIDIA-Certified servers with a dual-slot, full-height full-length PCIe Gen4 x16 slot rated for a 300W passively cooled accelerator.
Power and cooling: 300W maximum via a 16-pin connector; passive thermal solution, so directed chassis airflow is mandatory.
Software: an NVIDIA data center driver branch supporting Ada Lovelace. vGPU profiles require separately licensed NVIDIA vGPU software.
Pre-deployment verification: model designation on the label, slot clearance, power delivery and connector type, chassis thermal rating, driver branch.
Frequently Asked Questions (FAQ)
Q. Is "L40 CNX" a documented NVIDIA product?
No. No public NVIDIA documentation describes an L40 CNX. NVIDIA's documented converged accelerators are the A100X and A30X (GPU plus BlueField-2 DPU) and the H100 CNX (H100 plus ConnectX-7 SmartNIC). The CNX marking on this unit must be confirmed from the unit label and regulatory marking before ordering. T.E.S IT-SOLUTIONS makes no claim of an on-board ConnectX network interface on this unit.
Q. What are the verified NVIDIA L40 specifications?
Per the NVIDIA L40 Datasheet: Ada Lovelace architecture, 18,176 CUDA cores, 48GB GDDR6 with ECC at 864GB/s, PCIe Gen4 x16, 90.5 TFLOPS FP32, 209 TFLOPS RT Core, 4x DisplayPort 1.4a, 300W over a 16-pin connector, passive cooling, 4.4in x 10.5in dual slot.
Q. Does the NVIDIA L40 support MIG or NVLink?
No. The NVIDIA L40 Datasheet lists MIG Support: No and NVLink Support: No.
Q. Does the NVIDIA L40 have display outputs and video engines?
Yes. The NVIDIA L40 Datasheet lists 4x DisplayPort 1.4a, three NVENC and three NVDEC engines including AV1 encode and decode, and vGPU software support.
Q. Is this an L40 or an L40S?
This listing describes the NVIDIA L40, 48GB GDDR6 with ECC at 864GB/s. The L40S is a separate NVIDIA SKU with different published figures. Because the label reads L40 CNX rather than a standard NVIDIA designation, confirm the exact model from the unit label before ordering.
Q. What power and cooling does the NVIDIA L40 require?
300W maximum via a 16-pin power connector with a passive thermal solution, so the host chassis must provide directed airflow.
Q. What condition is supplied by T.E.S IT-SOLUTIONS?
Refurbished. Each unit receives a cosmetic inspection and, where a matching test platform is available, a power-on and detection check. Because the CNX marking is not a documented NVIDIA L40 variant, pricing is set conservatively and final pricing is subject to confirmation of the exact model. EU stock, global B2B shipping.
Why T.E.S IT-SOLUTIONS
Functional verification where a test platform is available, capture of the unit label and regulatory marking, cosmetic inspection, anti-static packaging, global B2B shipping from EU stock.