NVIDIA hits major roadblocks with Blackwell AI GPU: revised B200A AI GPU is now in the works

NVIDIA is reportedly experiencing some major issues with its new Blackwell AI GPUs, with the new B100 and B200 AI GPU shipments heavily impacted, with design flaws causing major headaches not just for NVIDIA.

3

In a new report published by SemiAnalysis, the outlet reports that there are issues with the Blackwell die needing a possible redesign or that the bridge dies inside TSMC's new CoWoS-L advanced packaging needs a redesign.

The result? NVIDIA is now reportedly working on the B200A AI GPU, which will be released towards the second half of 2025 and won't have the issues that are now plaguing the fastest AI GPU before it could hit the market.

B200 were meant to begin shipping in Q4 2024, but now SemiAnalysis reports that the roadmap has been updated and that B200 will begin shipping in "low volume" on the tail end of 2024, while B200A arrives in Q2-Q3 2025 (a year or so from now).

• Read more: NVIDIA's new Blackwell AI GPUs 'encountering major issues' which require redesign, big delays
• Read more: NVIDIA's next-gen Blackwell AI GPUs delayed, rumor has it 'design flaws' are to blame

The new B200A is based on the B102 die, which is also used on the China version of Blackwell: B20. The new B102 is a single monolithic compute die with 4 stacks of HBM, and it can be packaged on CoWoS-S instead of CoWoS-L or even NVIDIA's other 2.5D packaging suppliers.

3

B200A will come in both 700W and 1000W HGX form factors, with up to 144GB of HBM3E memory, and up to 4TB/sec of memory bandwidth (less memory bandwidth than H200).

NVIDIA is reportedly dealing with the issues by extending the Hopper AI GPU series and introducing a new Blackwell family GPU: the B200A. The new B200A AI GPU would be packaged using TSMC's CoWoS-S and rival packaging technologies from other suppliers, including Amkor, ASE, and Samsung.