TapTechNews October 11th news, Trendforce, a consulting firm, released a blog post today (October 11th), reporting that after Nvidia ships GB200 in the fourth quarter of this year, it is considering using a standalone GPU slot design in the next-generation AI GPU product to replace the current on-board solution, which is beneficial to supply chain companies such as Foxconn and interconnection component supplier Lotes.
Industry sources say that Nvidia plans to switch from on-board solution (OAM) to socket type, and it is estimated to be applied to the GB300 product line at the fastest.
The slot design is usually used in the CPU server market. Nvidia extends it to the GPU, which is beneficial to the replacement and upgrade of the GPU, realizes modularization, and will simplify the after-sales service and the maintenance of the server board, while optimizing the yield of the computing board manufacturing.
TapTechNews quoted the media report that the B300 series may become Nvidia's flagship product in the second half of 2025, and the main attraction of the B300 series is that it adopts FP4, which is very suitable for the inference scenario.