Recently, media reports have highlighted the research and development progress of NVIDIA's next-generation GB300AI server. According to information from Digitimes, the design work for the GB300 server is underway, with an official launch expected in the second quarter of 2023 and trial production set to begin in the third quarter. The introduction of this server will mark another significant breakthrough for NVIDIA in the field of AI computing.
One of the standout features of the GB300 server is its cooling design. Compared to previous models, the cooling requirements for the GB300 server have significantly increased, leading to a reduction in the number of fans used on the motherboard. This means that the new generation of servers will rely more on liquid cooling systems to better address the heat challenges posed by high-performance computing. The improvements in the cooling system will directly affect the server's performance and stability, allowing it to maintain good operational conditions even under high loads.
According to reports, NVIDIA plans to launch the GB300 server in mid-2025, which will undergo a comprehensive design upgrade covering all aspects from chips to peripherals, thereby unleashing more powerful AI computing capabilities. In terms of chips, the GB300 server will be equipped with a super chip based on the latest B300GPU, which will significantly enhance FP4 performance, with power consumption rising from 1000W in the B200 to 1400W, reaching twice that of the original B100. This change indicates that the GB300 server will have greater capabilities in handling complex computing tasks.
Regarding memory, the HBM specifications of the GB300 will also be upgraded, reaching 288GB with 8-stacked 12Hi HBM3E technology. This will further enhance data processing speeds and overall performance. Additionally, the B300GPU may adopt a slot design, potentially improving production yield and simplifying after-sales maintenance, while the Grace CPU will use LPCAMM memory modules to replace the existing LPDDR5 memory, thus enhancing performance.
In terms of interconnect technology, the GB300 server will be equipped with the next-generation ConnectX-8 SuperNIC and optical modules capable of up to 1.6Tbps, significantly boosting data transfer speeds and ensuring the server's efficiency when handling massive amounts of data. However, the introduction of a fully liquid cooling solution will also increase the server's cost. According to WccfTech, the top configuration price of the GB300 server is expected to far exceed the current approximately $3 million price tag of the GB200NVL72 server, further solidifying its position in the high-end market.
Key Points:
🚀 The GB300AI server is expected to launch in the second quarter of 2023, with trial production in the third quarter.
🌊 The new server will adopt a liquid cooling design, reducing the number of motherboard fans while significantly increasing cooling demands.
💲 The top configuration of the GB300 server is expected to far exceed the current GB200NVL72 server, targeting a higher-end market.
