Production of Nvidia's GB200 models increases subsequent to suppliers addressing AI server overheating issues and leaks in the liquid cooling system.
It Seems Nvidia's GB300 Hype Was A Bit Premature
Looks like the AI community will have to wait a bit longer for the GB300, as Nvidia's Taiwanese partners have allegedly faced some technical hurdles. According to the Financial Times, suppliers like Foxconn, Inventec, Dell, and Wistron have hit some roadblocks which delayed shipments. These issues emerged at the end of last year, causing a production hold-up.
However, all's not lost! Nvidia partners announced at Computex 2025 that GB300 shipments commenced at the end of Q1 2025. Production is said to be ramping up rapidly, as these suppliers have been working tirelessly to overcome various obstacles.
One supplier told FT that internal testing revealed connectivity problems, necessitating collaboration with Nvidia two to three months ago. The sources also cited overheating, leaks in the liquid cooling systems, software bugs, and inter-chip connectivity problems due to the complexity of synchronizing so many processors.
An analyst cautioned that Nvidia had not allowed the supply chain sufficient time to be fully prepared, suggesting that inventory risk for the GB300 will ease in the second half of the year.
Interestingly, the Financial Times claims that Nvidia has reverted back to the older "Bianca" board design for the GB300, instead of the new "Cordelia" design. The Bianca layout powers the GB200 AI servers and is more reliable and easier to maintain. While this move will preclude the replacement of individual GPUs in the system, it will help expedite GB300 deployment and eliminate further delays caused by the problematic Cordelia board.
This news aligns with an earlier report from May, suggesting that Nvidia was delaying the introduction of SOCAMM memory tech originally planned for the Blackwell Ultra GB300. Yet, Nvidia still plans to implement the Cordelia board layout for its next-generation Rubin chips.
In other words, the GB300 will launch sooner, but with fewer cutting-edge features and lower complexity. On the bright side, the advanced memory subsystem, SOCAMM, and other innovative features will likely debut with Rubin, positioning Rubin as Nvidia's next major leap in AI hardware.
Stay tuned for more updates on this developing story! For all your tech news, be sure to follow Tom's Hardware on Google News.
Data-and-cloud-computing advancements have been instrumental in enabling Nvidia's partners to collaborate remotely and address the technical hurdles faced in the production of the GB300. Moreover, technology solutions such as the cloud have facilitated continuous testing and optimization, assisting partners in resolving issues like connectivity problems, overheating, software bugs, and inter-chip connectivity problems more efficiently.
