Shortly after an analyst reported a significant reduction in lead times for Nvidia’s Hopper-based H100 GPUs, utilized in artificial intelligence (AI) and high-performance computing (HPC), Nvidia itself announced expectations of supply constraints for its forthcoming Blackwell-based GPU products.
More About Nvidia Supply of Next-Gen GPU
During Nvidia’s call with analysts and investors Colette Kress, the company CFO disclosed that there are expected supply constraints, due to high demand for Nvidia’s upcoming Blackwell architecture, which will power the B100 products. These new products are projected to offer improved AI computing performance compared to the Hopper architecture. It is likely that Nvidia’s existing customers have already placed orders for some B100 products given the demand for high-performance AI processors in the market.
The main focus is on how Nvidia can ramp up production of B100 SXM modules B100 PCIe cards and DGX servers based on the architecture. These products utilize components. Represent a significant shift. Rumors are suggesting that Blackwell could be Nvidia’s venture into chiplet designs. If this turns out to be true it could simplify the production of GPUs at the silicon level by improving chip yields. However, incorporating chiplet solutions also introduces additional complexities.
Nvidia is getting ready to launch a B40 GPU for business and training purposes alongside the GB200 product that pairs the B100 GPU with an Arm-based Grace CPU. They are also working on the GB200 NVL designed specifically for training language models. In November Nvidia unveiled the H200 compute GPU suited for AI and HPC tasks, which is now, in the production phase.
Based on the Hopper architecture, the H200 essentially refreshes Nvidia’s existing product lineup with increased memory capacity and bandwidth. Given Nvidia’s refined Hopper supply chain since 2022, the product ramp-up should progress relatively swiftly. However, Nvidia’s CEO, Jensen Huang, acknowledges that meeting 100% demand for this new product immediately is not feasible.
Huang emphasized during the quarterly earnings call with analysts and investors that the transition from zero to significant production levels cannot occur overnight, especially with new product generations like the H200.
