On March 18th local time, the GTC (GPU Technology Conference) officially opened in 2024. From CPUs, GPUs to large models, Nvidia has launched multiple "dazzling" works. Along with the continuous launch of high-performance products, how to "cool down" massive computing devices has become a hot topic of concern in the industry.
On March 19th, a reporter from the Daily Economic News learned from Lenovo Group that during the GTC conference, Lenovo Group launched a new 5U system using Lenovo Neptune liquid cooling technology - the all-new Lenovo ThinkSystem SR780a V3, with a power efficiency (PUE) of 1.1. By using direct water cooling with CPU and GPU, as well as NVIDIA NVSwitch technology, the system can maintain maximum performance without reaching any thermal limits.
Public information shows that direct water cooling solutions reduce power consumption by circulating warm water to cool data center systems, which is more efficient than traditional air-cooled systems. According to the Uptime Institute's 2021 Global Data Center Annual Survey, PUE and power consumption are one of the most focused sustainable development methods as industry indicators for determining data center energy efficiency. Due to liquid cooling providing a more energy-efficient alternative than air, the system can achieve higher sustained performance while consuming less energy. In this context, with the rapid iteration of AI technology and the rapid expansion of data centers, water cooling technology has become a new round of competition hot spot for various manufacturers.
In addition to water-cooled products, according to Lenovo Group, in terms of large-scale and efficient enhancement of artificial intelligence workloads, Lenovo Group has launched an expansion of the ThinkSystem artificial intelligence product portfolio, which includes two new 8-way NVIDIA GPU systems. Lenovo Group stated that these two systems are designed specifically to provide massive computing power, generative artificial intelligence, natural language processing (NLP), and large-scale language modeling (LLM) development, to support the entry of NVIDIA HGX artificial intelligence computing platform into the market.
In addition, Lenovo Group also showcased the all-new PG8A0N, the ultimate 1U node designed specifically for AI. According to Lenovo, customers can run NVIDIA AI Enterprise on Lenovo's NVIDIA certification system product portfolio; NVIDIA NIM inference microservices included in NVIDIA AI Enterprise can also be run on Lenovo's enterprise level infrastructure to perform high-performance inference on popular AI models in NVIDIA and its ecosystem.