HP推出的ZGX Fury AI工作站现已开放订购1，该系统集成了NVIDIA GB300 Grace Blackwell超芯片，配置252GB HBM3e显存、72核Grace CPU和496GB LPDDR5X内存1。系统总统一内存容量达748GB1，可支持万亿参数模型推理和千亿参数模型微调1，最高提供20 petaFLOPS的FP4计算性能1。
这套工作站采用1400W液冷系统，满载运行时GPU温度可维持在71°C1，配备两个QSFP112 400Gbps端口以及10GbE和1GbE管理网络1。HP与Red Hat和NVIDIA合作在该系统上部署Red Hat AI Factory，将其定位为部门级或分支机构的共享推理设备1。系统支持Red Hat Enterprise Linux和Red Hat AI Factory with NVIDIA1，提供塔式和5U机架两种形态1，用户可选配NVIDIA RTX PRO GPU驱动显示器1。
HP has opened orders for its ZGX Fury AI workstation, featuring NVIDIA's GB300 Grace Blackwell superchip and delivering up to 20 petaFLOPS of FP4 compute performance 1. The system is equipped with 748GB of unified memory, comprising 252GB of HBM3e GPU memory, a 72-core Grace CPU, and 496GB of LPDDR5X memory 1. This configuration enables inference workloads for trillion-parameter models and fine-tuning of hundred-billion-parameter models 1.
The ZGX Fury has been designed as a shared inference device for departments or branch offices rather than as an individual workstation 1. HP is collaborating with Red Hat and NVIDIA to deploy Red Hat AI Factory on the system, providing enterprise-grade AI capabilities 1. The workstation features robust cooling through a 1400W liquid cooling system that maintains GPU temperatures at 71°C under full load 1. Connectivity options include two QSFP112 400Gbps ports alongside 10GbE and 1GbE management networks 1.
The system is offered in both tower and 5U rack form factors, with optional NVIDIA RTX PRO GPU display drivers available 1. It supports Red Hat Enterprise Linux and Red Hat AI Factory with NVIDIA integration 1.
评论
还没有评论，欢迎留下第一条。