NVIDIA’s foray into the AI server market with its new 2U air-cooled MGX GB200 NVL2 has hit a snag, according to a recent investment note from Morgan Stanley. The server, packing a punch with two Grace CPUs and two Blackwell B200 AI GPUs all crammed onto a single PCB board, appears to be battling a troublesome foe – heat.
The note dives deeper into the challenges, explaining that the server’s design, with the GPU module connected to the main board via an SXM7 module, is facing thermal bottlenecks within the 2U form factor. While all servers showcased at the Open Compute Project (OCP) were initially designed for the compact 2U form factor, Morgan Stanley’s discussions with supply chain partners suggest a potential switch to a larger 4U form factor to mitigate these thermal issues.
This revelation casts a shadow over NVIDIA’s initial plans for the MGX GB200 NVL2, potentially delaying its rollout and raising questions about the server’s overall efficiency and long-term stability. The move to a 4U form factor would undoubtedly impact the server’s footprint and potentially introduce new challenges in terms of power consumption and deployment in data centers.
As the AI race intensifies, NVIDIA’s MGX GB200 NVL2 was expected to be a significant contender, offering powerful processing capabilities to fuel the next wave of AI innovation. However, the thermal hurdles pose a substantial obstacle for the company, raising concerns about its ability to meet the demanding requirements of AI workloads.
The evolving story of NVIDIA’s MGX GB200 NVL2 serves as a reminder that even the most innovative technology can face unexpected challenges, particularly when dealing with the complexities of power and cooling in high-performance computing environments. The industry will be watching closely to see how NVIDIA addresses these thermal concerns and if the switch to a 4U form factor will ultimately impact the server’s performance and market appeal.
