23.5 billion, NVIDIA has invested in MediaTek

The two parties will carry out cooperation around AI infrastructure, on-device AI computing and automotive sectors.

August 31 report from AI Core: NVIDIA has just announced that it will jointly develop the next-generation AI computing platform with MediaTek, with cooperation covering three major sectors: AI infrastructure, on-device AI computing and automotive.

NVIDIA will also invest 3.5 billion US dollars (equivalent to about 235.19 billion yuan) to subscribe to the convertible bonds issued by MediaTek.

According to the cooperation agreement, MediaTek will adopt NVIDIA's NVLink Fusion platform to serve hyperscale cloud service providers, cloud service providers and cutting-edge model developers, helping them build custom XPU devices and integrate these devices into rack-scale AI systems connected via NVIDIA NVLink.

The two parties plan to integrate NVIDIA's accelerated computing, artificial intelligence, graphics and software platforms with MediaTek's strengths in custom chips, high-performance computing, SoC design, advanced packaging, interconnection technology and connectivity technology.

Specifically, NVIDIA and MediaTek will carry out cooperation in three areas:

1. AI Infrastructure: MediaTek will help enterprises develop customized AI infrastructure based on the NVIDIA NVLink Fusion ecosystem, making it compatible with NVIDIA rack-scale systems and AI computing factories.

In this sector, the NVLink Fusion platform can provide a prefabricated, pre-certified and system-level pre-verified foundation for multi-die XPU development, accelerating the implementation process from chip silicon wafers, advanced packaging to rack-scale systems.

The NVLink Fusion platform integrates key technologies required to build custom XPUs, including the NVIDIA NVLink Fusion die, which can connect the XPU to the NVIDIA NVLink horizontal scaling interconnection architecture via NVIDIA photonics interconnection or electrical interconnection; NVLink-C2C provides high-bandwidth, energy-efficient interconnection between XPUs, NVIDIA Rosa CPUs and other compatible processors; NVHBM integrates customized memory capabilities, increasing bandwidth and energy efficiency while leaving more chip area for computing units.

On this basis, enterprises do not need to design and verify all components around the custom XPU from scratch. The NVLink interconnection, memory architecture, packaging, manufacturing and rack-scale technologies required for mass production and implementation can be supported by NVIDIA and MediaTek.

Enterprises can submit their own XPU designs to MediaTek, and customize and optimize the interconnection, memory, packaging, performance and power consumption characteristics according to their own business load and infrastructure requirements.

In addition, MediaTek is also a member of the NVLink Fusion ecosystem, so hyperscale cloud vendors, cloud service providers and cutting-edge model developers can quickly connect custom XPUs to NVIDIA AI infrastructure.

2. On-device AI Computing: The two parties will cooperate to develop multiple generations of NVIDIA RTX Spark and DGX Spark PC chips. This series of chips integrates NVIDIA GPUs and MediaTek SoCs to provide computing power support for consumer-grade PCs, AI developer supercomputers and enterprise-level workstations.

MediaTek has previously cooperated with NVIDIA to develop the GB10 Grace-Blackwell superchip, which provides computing power for the NVIDIA DGX Spark.

GB10 interconnects the NVIDIA Blackwell GPU and Grace CPU via NVLink-C2C to provide AI capabilities for edge systems. This time, the two parties will jointly develop the NVIDIA RTX Spark in the future to provide computing power support for the next-generation consumer-grade PCs in the AI era.

3. Automotive: In the field of physical AI, the two parties plan to jointly develop platforms that support AI capabilities and software-defined vehicles.

The MediaTek Dimensity Auto platform integrates NVIDIA technologies to provide AI capabilities and NVIDIA RTX graphics rendering capabilities for smart vehicle cockpits, and can work in synergy with NVIDIA DRIVE AGX.

Jensen Huang, Founder and CEO of NVIDIA, said that from large-scale AI computing factories to personal computers and automobiles, AI is reshaping every type of computing platform. The platform jointly built by NVIDIA and MediaTek will expand NVIDIA's accelerated computing technologies to brand new markets, enabling customers to freely build differentiated competitive AI systems on a large scale.

This article is from the WeChat official account "AI Core", Author: Cheng Qian, Editor: Panken, published with authorization from 36Kr.