As an Amazon Associate, we earn from qualifying purchases. TweakTown may also earn commissions from other affiliate partners at no extra cost to you.
NVIDIA's beefed-up B300 AI chip production has been reportedly pulled forward to May, and will be fabbed on TSMC's new 5nm (N4P) process node and will use CoWoS-L advanced semiconductor packaging.
In a new post by Ctee, we're hearing that NVIDIA's new B300 AI chips will use Bianca compute boards with 1 x CPU and 2 x GPUs, with B300 to enter mass production before the end of the year. Analysts estimate that NVIDIA's new B300 AI GPU will boost the related supply chains including TSMC, Machtech, Inventec, Chipset, and assemly plants Quanta, Wistron, and Foxconn.
NVIDIA's new B300 using TSMC's new 5nm family (N4P) process node was used to fill the production capacity gap left wide open by the now-banned H20 AI GPU in China. Ctee reports that the supply chain noted a shipment of AP8 from Nanya Advanced Packaging in early April, in order to follow the CoWoS-L advanced packaging to be used for B300.
The strong customer demand has pushed TSMC to quickly ramp production capacity, with NVIDIA Chief Scientist Bill Dally saying at TSMC's recent North American Technology Symposium that the B200 AI chip uses CoWoS advanced packaging to package two GPUs, breaking the limitations of the single reticle limit size.
Semiconductor analysts believe that TSMC is pushing various advanced semiconductor packaging technologies, and that they're breaking through the limitations of Moore's Law by increasing the package size and stacking in more and more transistors.
- Read more: Apple becomes NVIDIA customer, orders $1B+ of new GB300 NVL72 AI servers
- Read more: NVIDIA GB300 'Blackwell Ultra' AI GPU: 288GB HBM3E, 1.4kW power, 50% faster than GB200
We should expect to hear all about NVIDIA's new B300 AI GPU at Computex 2025 next month, where NVIDIA is also expected to launch new robotics technology and other products -- with the NVIDIA x MediaTek Arm-based AI PC processor teased for a possible Computex 2025 reveal.