Advantech SKY-TESL-L4-24P NVIDIA L4 24GB Low-Profile Passive AI Inference GPU with 72W Slot Power

Video analytics and AI inference at the edge often have to fit a short server, a compact chassis or a tight power budget. The SKY-TESL-L4-24P is the NVIDIA L4 from Advantech, an Ada Lovelace AI inference GPU in a single-slot, low-profile card that draws up to 72 W. It suits video, AI and graphics applications in the enterprise, in the cloud and at the edge, where slot height and power are limited.

The GPU has 7,680 CUDA cores, 240 fourth-generation Tensor Cores and 60 third-generation RT Cores, rated at 30.3 TFLOPS FP32, with DLSS 3 support. Memory is 24 GB of GDDR6 with ECC at 300 GB/s. Two NVENC and four NVDEC engines include AV1 encode and decode, which suits decoding many camera streams for inference on the same card. NVIDIA vGPU support lets the SKY-TESL-L4-24P serve virtual desktops too, while Multi-Instance GPU and NVLink are not supported.

The card connects over PCI Express 4.0 x16 and has no separate power connector, so its 72 W comes through the slot. It is headless, with no display outputs, and its passive heatsink relies on the airflow of the server or edge system it sits in. The low-profile, single-slot format lets it go into compact chassis with half-height slots.

We supply the SKY-TESL-L4-24P tested, and fit, configure and burn-in test it in the GPU computers and servers we build, with airflow checked across the passive heatsink, and we can advise on how many cards suit a given chassis and power budget. For heavier graphics and AI loads, the SKY-TESL-L40S-48P offers 48 GB of GDDR6 and 4x DisplayPort, and the SKY-TESL-L40-48P is the 48 GB data centre graphics card at 300 W.