Video Card Nvidia A30 Tensor Core 24GB HBM2 900-21001-0140-030

Nvidia A30 Tensor Core 24GB HBM2 Graphics Card

The NVIDIA A30 is a server compute accelerator based on the Ampere architecture, not a display graphics card. It combines third-generation Tensor Cores with 24GB of HBM2 memory and 933GB/s of bandwidth, and is designed for AI inference, model training and high-performance computing (HPC).

The card is a dual-slot, full-height, full-length module that is passively cooled by the airflow of the server and draws up to 165W. It has no video outputs and connects through PCIe Gen4 and a single 8-pin power connector.

Key features

• Third-generation Tensor Cores - Support for TF32, FP64 Tensor Core and FP16, BF16, INT8 and INT4 precisions, together with a full software stack for AI workloads.
• 24GB of HBM2 memory - Up to 933GB/s of bandwidth and native ECC memory, with no overhead in capacity or bandwidth.
• Hardware partitioning with MIG - The accelerator can be split into up to four separate 6GB instances, two 12GB instances or a single 24GB instance, with guaranteed quality of service.
• HPC performance - 5.2 TFLOPS of FP64 and 10.3 TFLOPS of FP64 Tensor Core for research and simulation workloads.
• PCIe Gen4 and NVLink - A PCIe Gen4 interface with 64GB/s of bandwidth and third-generation NVLink at 200GB/s through a bridge for two cards.
• Built for servers - Passive cooling, up to 165W of power draw and support for NVIDIA AI Enterprise and Virtual Compute Server software.

Specifications