NVIDIA L2 24GB - a compact accelerator for inference and video in dense servers
The L2 is built on the Ada Lovelace architecture (AD104 chip) and carries 24 GB of GDDR6 memory with ECC support. This is the most compact data center accelerator of the L series: a single-slot low-profile board with passive cooling that is powered solely by PCIe Gen4 x16. In this form factor many cards fit into one rack without additional heat output from fans.
The card is designed for serving small and medium neural networks, video transcoding and desktop virtualization. 4th generation Tensor Cores support FP8, INT8, BF16 and TF32: 193 TOPS on INT8, 96.5 TFLOPS on BF16/FP16 and 48.3 TFLOPS on TF32 are declared. A separate media block (2x NVENC with AV1 encoding, 4x NVDEC, 4x NVJPEG) offloads video encoding and decoding from the main compute cores.
- Ada Lovelace architecture, 4th generation Tensor Cores (FP8) and 3rd generation RT Cores
- 24 GB GDDR6 with ECC, 300 GB/s bandwidth
- PCIe Gen4 x16 interface
- Single-slot low-profile form factor, passive cooling (LP and ATX brackets included)
- MIG support - partitioning into isolated GPU instances
- NVIDIA vGPU support (vWS, vApps, vPC) for virtualization
The L2 has no video outputs and no NVLink - this is a purely compute card for servers and edge nodes. ECC memory and Enterprise drivers provide predictable operation under production load. ETE.UA is an official PNY/NVIDIA supplier in Ukraine: distributor warranty, B2B documents with VAT and engineer consultation at tel. +380 93,594 00 77.
Not sure this card fits your system?
We will check compatibility with your platform, calculate the power draw for your configuration and tell you whether the memory is enough for your models.
The model you need is out of stock? ETE is listed among official PNY partners, so we bring professional NVIDIA cards to order with the manufacturer warranty.
Message us on Telegram or leave a request.
