Discover B200 Blackwell: The Next Generation AI Chip Powered by Dual TSMC Silicon Reticles
The NVIDIA B200 GPU Blackwell product introduction image. Image credit: NVIDIA Corporation)
NVIDIA has consistently led the charge in the AI revolution with its advanced hardware and software stack. Now, with the introduction of their Blackwell hardware family, it's evident they're determined to maintain their position at the forefront. The seemingly insatiable demand in the market for hardware to fuel machine learning models has propelled NVIDIA's market value to skyrocket by over 240%, soaring past the 2 trillion mark in the last 12 months.
The B200’s predecessor used TSMC’s industry leading chip on wafer on substrate (CoWoS) 2.5D advanced packaging technology. Discover the intricate details showcased in our comprehensive packaging report on the H100 Hopper. Picture a single TSMC accelerator die enveloped by up to six high bandwidth memory (HBM) modules, all ingeniously mounted on a silicon interposer. TSMC's CoWoS emerges as a pivotal force in the AI realm, driven by the widespread acclaim for NVIDIA’s products and their adoption by competitors. Yet, its soaring popularity has inadvertently created a bottleneck in the industry, triggering TSMC's investment in new facilities to expand production.
While analysis of the B200's packaging and dies is slated for Q1 2025, dive into our insights on NVIDIA's strategies to bolster production capacity today.
