Advantech SKY-TESL-H200N-141 NVIDIA H200 NVL 141GB HBM3e Data Centre GPU with NVLink Bridge Support and PCIe 5.0

Generative AI models and HPC datasets often run out of GPU memory before they run out of compute. The SKY-TESL-H200N-141 is the NVIDIA H200 NVL from Advantech, a compute-optimised data centre GPU on the Hopper architecture with 141 GB of HBM3e. It comes as a passive, dual-slot PCIe card for air-cooled enterprise rack servers, where it suits generative AI and high-performance computing (HPC) workloads.

The GPU has 14,592 CUDA cores and 456 fourth-generation Tensor Cores, rated at 60 TFLOPS FP32 and 30 TFLOPS FP64, with fast FP64 for double-precision work. Its 141 GB of HBM3e with ECC runs at 4.8 TB/s, and second-generation Multi-Instance GPU can split the SKY-TESL-H200N-141 into up to seven instances, each with partitioned hardware and guaranteed quality of service. Seven JPEG decoders and seven video decoders handle image and video ingest on the card.

The card connects over PCI Express 5.0 x16 and draws up to 600 W through a 16-pin PCIe connector. It is headless, with no display outputs, and its passive heatsink relies on the server’s airflow. For multi-GPU jobs, the optional SKY-TESL-NVL-2A two-way and SKY-TESL-NVL-4A four-way NVIDIA NVLink bridges connect cards at 900 GB/s, both at dual-slot pitch. NVIDIA’s NGC catalog provides software, libraries and optimized AI models.

We supply the SKY-TESL-H200N-141 tested, and fit, configure and burn-in test it in the GPU servers we build, including NVLink bridging, power and airflow for multi-card builds, and we can advise on how many cards a given chassis can power and cool. Where 94 GB is enough, the SKY-TESL-H100N-94P is the H100 NVL with HBM3 at 400 W, and the SKY-TESL-H100-80P is the H100 PCIe card with 80 GB of HBM2e at 350 W.