Abstract:

This video discusses key announcements made by Jensen Huang of Nvidia at a recent keynote in Taipei. The primary focus is on the official announcement of NVLink IP licensing to partners including MediaTek, Marvell, Cadence, and Synopsys. This initiative allows other companies to integrate Nvidia's high-bandwidth NVLink SerDes technology, considered one of the fastest available, into their custom silicon designs. The discussion delves into who might benefit from this (positing it could be smaller players or those seeking customization rather than hyperscalers), the scope of the licensed IP (extending beyond just the physical SerDes layers), and the potential influence Nvidia might retain over its use. The video also briefly mentions the new Nvidia office in Taiwan and updates regarding RTX Pro servers utilizing PCIe cards and GDDR7 memory.

Nvidia's Taipei Keynote: NVLink IP Licensing and Hardware Updates

• 0:00 Jensen Huang's Keynote: Discussion of Jensen Huang's recent keynote at the Taipei Music Center.
• 0:16 NVLink IP Announcement: Official announcement of NVLink IP licensing, allowing partners to use Nvidia's high-bandwidth connectivity in custom silicon.
• 0:36 Partners: Partners for NVLink IP licensing include MediaTek, Marvell, Cadence, and Synopsys.
• 0:50 NVLink Technology: NVLink is described as Nvidia's custom, high-bandwidth SerDes (Serializer/Deserializer) connection, noted as potentially the fastest on the market due to Nvidia developing it for internal use first.
• 1:31 Target Users Speculation: Discussion with Dylan suggests that potential users of the NVLink IP might be smaller players or those seeking customization ("biggest losers" in the hyperscaler race) like Qualcomm or sovereign AI efforts, rather than major hyperscalers who might prefer independent solutions.
• 2:13 Scope of IP Licensing: The licensing covers the entire NVLink solution, including packet processing and the MAC layer, not just the SerDes itself.
• 2:21 Nvidia's Control: Speculation on how much control Nvidia will retain over which companies get to use the NVLink IP in their final products, possibly requiring Nvidia's approval for certain customers (e.g., Amazon).
• 3:20 EDA Partner Benefits: Cadence and Synopsys benefit from the increased verification and design complexity involved in integrating the NVLink IP into customer pipelines.
• 3:55 New Taiwan Office: Mention of Nvidia's new office in Taiwan, with a humorous note about its location relative to the city center and its perceived busyness during lunchtime due to subsidized food.
• 5:15 RTX Pro Servers: Brief mention of RTX Pro servers, which use PCIe cards and GDDR7 memory, offering an alternative to XXM modules and HBM.
• 5:26 RTX Pro 6000: The creator expresses interest in obtaining an RTX Pro 6000 card.