The industry revealed that Nvidia will launch a new HPC GPU platform for data center, artificial intelligence and gaming applications in 2022. Its manufacturing partner TSMC is reportedly expected to triple orders with packaging substrate and thermal material suppliers to support large-scale CoWoS packaging of new chips.
DigiTimes quoted the source as saying that Nvidia estimates that its data center HPC chip shipments will grow by 200%-250% year-on-year in 2022, and that its data centers based on the new Hopper architecture, as well as AI chip solutions, will greatly increase drive this growth. These solutions will be available as early as July .
According to reports, the new HPC chips will be manufactured by TSMC using 4nm (N4) process technology (an enhanced version of the 5nm node) and CoWoS packaging solutions. TSMC's purchases of thermal, underfill and flux materials, as well as substrates for CoWoS applications, are likely to triple this year to meet strong orders from Nvidia, the sources added.
Nvidia is also expected to launch its next-generation Blackwell-based HPC chips in the first half of 2023, but it remains to be seen whether TSMC's HPC-specific N4X node-set will be used for trial production by then, the person noted.
Currently, AMD also uses CoWoS technology to process most of its HPC and server chips, while packaging some of its products through ASE's FOEB 2.5D IC packaging solution to improve cost performance.
0 Comments