PNY NVIDIA H100 NVL (TCSH100NVLPCIE-PB)

The PNY NVIDIA H100 NVL (TCSH100NVLPCIE-PB) is a server compute accelerator designed for data centers, AI inference, large language models, HPC, analytics, large dataset processing, and professional application acceleration. Built on the NVIDIA Hopper architecture, this card is engineered for PCIe-based servers with appropriate power delivery, cooling, and software infrastructure. The H100 NVL features 94 GB of HBM3 memory with 3.9 TB/s bandwidth, critical for parameter-intensive models and workloads where memory access speed is essential. Computational performance reaches 30 TFLOPS FP64, 60 TFLOPS FP64 Tensor Core, 60 TFLOPS FP32, 835 TFLOPS TF32 Tensor Core with sparsity, 1671 TFLOPS BFLOAT16/FP16 Tensor Core with sparsity, and 3341 TOPS INT8 Tensor Core with sparsity. The card supports NVLink via bridges for scaling performance across multi-GPU server configurations. The maximum thermal design power is adjustable between 350–400 W with passive cooling, requiring properly organized airflow within the server chassis. The PNY NVIDIA H100 NVL includes NVIDIA AI Enterprise and is designed for organizations deploying generative AI on standard PCIe servers.