Nvidia Reportedly Skipping TSMC N2 Node for A16 in Future GPU Generations

Nvidia, a company not typically known for being at the forefront of adopting the newest silicon fabrication technologies, is reportedly making a significant shift. New information suggests that Nvidia intends to be the inaugural client for TSMC's highly anticipated A16 manufacturing node. This strategic acceleration in process technology adoption is believed to be a direct response to intensifying competition, particularly from rivals such as AMD, which is also developing high-performance chips on TSMC's upcoming N2 node. This potential leap to A16 for its future GPU generations, specifically the Feynman architecture slated for 2028, indicates a proactive approach to overcoming current limitations and pushing the boundaries of chip design and performance.

Historically, Nvidia has often utilized slightly more mature fabrication processes. For instance, its current generation of GPUs, including the Blackwell series, relies on a variant of TSMC's N4 node, which is an enhancement of the N5 technology originally introduced in 2020. This contrasts with Apple, which launched its A17 Pro SoC for iPhones on TSMC's N3 technology as early as 2023. The consequence of Nvidia's current approach is that its chip designs are encountering the reticle limit, a fundamental physical constraint on the maximum size of a single chip that can be produced on a wafer. This limit directly impacts the achievable transistor density, thereby restricting potential performance improvements. For example, Nvidia's powerful GB200 AI GPU combines two reticle-sized GPUs on a single interposer, and the RTX 5090 (codenamed GB202) is nearing the N4 reticle limit at 750mm².

The continued use of N4 production technology, largely carried over from the previous RTX 4090 family to the current RTX 50 GPUs, explains why certain models like the RTX 5070 and RTX 5080 have offered only incremental performance gains over their predecessors. This stagnation underscores the necessity for a substantial technological upgrade to facilitate more significant advancements.

For the forthcoming Rubin generation of GPUs, Nvidia is expected to transition to TSMC's N3 node. However, if the recent reports regarding TSMC A16 prove accurate, Nvidia might bypass the N2 node entirely, jumping directly to A16 for the generation following Rubin, codenamed Feynman, which is anticipated around 2028. This move holds considerable implications for both artificial intelligence and gaming GPU sectors.

Nvidia’s long-standing strategy involves employing a unified GPU architecture across both AI and gaming products, with all chips fabricated on the same node. While AI and gaming GPUs serve distinct purposes, they share common elements, such as Tensor cores, making it more efficient to design them for a single manufacturing process. If Nvidia does indeed transition its gaming GPUs to the A16 node, it could lead to a dramatic increase in transistor count within a few generations. Such an increase would unlock substantial performance boosts and enable the integration of innovative new features, benefiting PC gaming enthusiasts significantly.

While Nvidia has faced criticism for delivering incremental upgrades in some recent GPU generations, often attributed to the use of older TSMC nodes, the potential adoption of A16 for the Feynman generation could mark a turning point. This advanced node promises to provide the necessary technological foundation for more substantial performance gains and architectural innovations, potentially reinvigorating the PC gaming hardware landscape. However, as with all future technology announcements, the ultimate impact will only be clear once these products are released and tested in real-world scenarios.