72× B200 GPUs + 36× Grace CPUs · 13.5 TB/s · Liquid-Cooled · NVLink 72
72× B200 (Blackwell)
36× NVIDIA Grace
Up to 13.8 TB
13.5 TB/s
1.44 EFLOPS FP4 (inference)
720 PFLOPS FP4
NVLink 72, all-to-all
Direct liquid cooling
42U rack-scale system
The NVIDIA GB200 NVL72 is the rack-scale Grace Blackwell platform that first brought NVLink 72 all-to-all connectivity to a single cabinet. It interconnects 72 B200 GPUs and 36 Grace CPUs into a coherent NVLink domain, exposing 13.5 TB/s of aggregate memory bandwidth and 1.44 EFLOPS of FP4 inference compute.
GB200 NVL72 is the direct predecessor of GB300 NVL72 and remains a high-volume deployment choice for frontier AI training and inference, offering a proven software and interconnect stack with liquid cooling and 720 PFLOPS of FP4 training throughput.
QS Compute can source GB200 NVL72 cabinets, sub-rack GB200 NVL36, and GB200 HGX configurations for AI infrastructure builders across APAC.
Need NVIDIA GB200 NVL72?
Enterprise hardware — configured, tested, deployed. Contact us for pricing and availability.
Request Quote