HPE 870919-001

NVIDIA Tesla P40 24 GB PCIe GPU Accelerator

The HPE 870919‑001 features an NVIDIA Tesla P40 GPU with 24 GB of GDDR5 memory and a PCI Express 3.0 ×16 interface. Designed for AI inference,high-performance computing, and virtual workstation acceleration, it supports compute workloads across CUDA, OpenCL, and TensorRT. With 3,840 CUDA cores and up to ~12 TFLOPS of single-precision performance, this Accelerator excels at parallel processing and inference tasks at scale. The large memory capacity and memory bandwidth (~346 GB/s) make it well-suited for large neural network inference, graphics virtualization, and simulation workloads in HPE ProLiant and Synergy server environments.

Features

Based on NVIDIA Pascal architecture (GP102 GPU), the Tesla P40 delivers up to ~47 TOPS of INT8 inferencing throughput (ideal for quantized deep-learning models), along with ~11.7 TFLOPS FP32 compute performance. The board is passively cooled and rated at ~250 W TDP, requiring sufficient chassis airflow for thermal management. Equipped with 24 GB GDDR5 memory on a 384‑bit bus, it sustains ~346 GB/s memory bandwidth for heavy workloads. The card supports hardware virtualization via NVIDIA GRID/vDWS profiles, up to 24 vGPU instances (1 GB profile) for remote desktops, plus GPU compute APIs like CUDA, OpenCL, and TensorRT. Compatible with HPE ProLiant Gen9/Gen10 servers including DL380, DL360, Apollo, XL, and Synergy platforms.

Advanced Options

The Tesla P40 is optimized for enterprise-class AI inference and GPU virtualization workloads. Powered by passive cooling, it relies on server airflow (bidirectional supported) to maintain thermal stability. With high single-precision and INT8 throughput, it accelerates inference models drastically—for example, a single server with eight P40s can replace over 140 CPU-only servers in inference tasks.