技術サポートE-mail:meixj@inspur.com お問い合わせE-mail:meixj@inspur.com
Inspur GX4 GPU Box A GPU Box which supports 4 GPUs
購入する

製品概観:

GX4 has inherited the pioneering design of the Inspur SR AI full cabinet solution. Decoupled CPU and GPU. Stand-alone device supports 4 GPUs. Equipped a Dual-Socket Server as head node. Ensuring efficient GPU cross-node communication efficiency. Effectively reduces IO redundancy and system purchase costs. Perfectly suitable for deep learning model training, scientific computation, engineering computation and research, among other such fields of application.

Product Features:

High-efficiency Parallel Processing, Powerful Performance

2U space supports 4 GPUs. 4 GX4 platforms can be equipped with any Dual-Socket server to compose a 16 GPU system, creating high-efficiency parallel computational processing capacity.

CPU/GPU Decoupling, Resource Pooling

Flexibly adjustable GPU topology. Responds to different applications.

Balanced: compatible with public cloud services and small-scale model training

Common: appropriate for Deep Learning model training

Cascaded: appropriate for Deep Learning model training and P2P performance optimization

Minimum Lag, Maximum Communication

Given a 16-GPU system composed of 4 GX4 platforms, model training can be conducted without requiring a network protocol for data exchange. Maximum distance of only 1 QPI conversion. Latency reduces 50%

仕様:

Model

GX4

Height

2U

GPU expansion

NVIDIA-certified. Supports 4 GPU accelerators, such as V100, P100, P40, M40, K80, M60, M10, etc.

Storage

16 2.5-inch U.2 hard drives

I/O expansion slots

1 PCI-E3.0 expansion slot, 4 mini PCIe 4-bit lines

Power supply

1600W 1 + 1 redundant power supply

会社概要

Inspurグループ パートナー ニュース

サポート

ダウンロードセンター

購入方法

購入方法

連絡先

連絡先 当社と提携する

INSPURをフォローする

Facebook Instagram Twitter

Copyright © 2018 Inspur. All Rights Reserved.

inspur logo
  • お問い合わせ:

    09042006344