This keynote speech by Dr. Rick Tsai, Vice Chairman and CEO of MediaTek, at Computex Taipei, focuses on the concept of "Intelligence Everywhere: The Power of Ubiquitous AI." Dr. Tsai outlines MediaTek's strategy to enable pervasive AI across various platforms, from mobile and automotive to the cloud, emphasizing a hybrid computing approach. He presents four key pillars: high-performance computing, cloud AI accelerators, seamless connectivity, and strong ecosystem partnerships. The speech details MediaTek's advancements in mobile chip performance and power efficiency (Dimensity series), highlights a strategic automotive partnership with Nvidia and developments in cloud AI accelerators, and underscores the company's expertise in connectivity. AI demonstrations showcasing on-device generative capabilities are presented, and guest appearances by ARM CEO Rene Haas and Nvidia CEO Jensen Huang reinforce the crucial role of collaboration in achieving the vision of ubiquitous AI.
Intelligence Everywhere: MediaTek's Strategy for Ubiquitous AI
0:00 Introduction to the keynote theme: "Intelligence Everywhere: The Power of Ubiquitous AI," emphasizing the rise of AI across all aspects of technology and life.
0:11 MediaTek, as a company established and grown in Taiwan, expresses pride in its achievements and commitment to contributing significantly to Taiwan's success.
0:36 Dr. Tsai highlights the ubiquitous nature of AI and MediaTek's ambition to be a major enabler of AI presence "all around you."
0:53 The current wave of investment in AI infrastructure and edge AI is compared to the scale of investment seen during the rise of the internet in the late '90s.
0:57 Computing and semiconductors are identified as being at the core of the AI experience.
0:59 MediaTek advocates for Hybrid Computing, believing computing should span everywhere from the edge (devices) to the cloud.
1:17 MediaTek aims to participate in and enable AI across multiple platforms: mobile devices (smartphones), living rooms (TVs), offices (PCs), and cars, all supported by cloud AI.
1:52 Four pillars are presented for enabling ubiquitous AI: 1) High-Performance Computing (Edge to Cloud), 2) AI Accelerator in the Cloud, 3) Seamless Connectivity, and 4) Closed Ecosystem Partnership.
4:05 Highlighting their foundation in mobile, MediaTek showcases the Dimensity series chips, particularly the 9300+, as offering ultimate performance and power efficiency for mobile devices.
4:51 MediaTek proudly notes its heritage in Taiwan and contributions to the semiconductor industry, alongside companies like TSMC.
5:58 Projections show about 1.2 billion mobile devices with AI capability by 2027, collectively offering massive computing power (>50,000 EOps), potentially exceeding the computing power of all AI GPUs shipped in 2023.
7:54 This edge computing power is seen as complementary to the cloud, working together rather than against each other.
8:28 MediaTek details its computing capability progress over 5 years (2019-2024): CPU performance improved 2.6x, GPU 6x, and NPU (AI) a whopping 18x, while power consumption was reduced by 60%.
9:57 The Dimensity 9300+ smartphone chip achieves 68 TOPS AI performance, exceeding the 40 TOPS requirement for Co-pilot+ PCs, demonstrating high AI capability in a portable device.
10:56 MediaTek is expanding beyond mobile into Automotive, announcing a partnership with Nvidia a year prior to move into cockpit control and eventually ADAS, projecting significant performance gains (5.8x TOPS, 4.5x faster LLM inference, 5.1x lower latency).
13:53 MediaTek also aims to participate in the Cloud AI Accelerator market, identifying a niche opportunity projected to reach $45 billion by 2028.
15:06 Six leading-edge capabilities are identified as necessary for Cloud AI Accelerators: Computing, High-speed Interconnect (SerDes, PCIe, UCIe), Advanced Process Technology (4nm, 3nm, 2nm), Advanced Packaging (2.5D, 3D), High Bandwidth Memory (HBM), and High-density Low-power SRAM. MediaTek claims proficiency in all these areas.
17:09 MediaTek demonstrates its capabilities with examples like the Dimensity 9300+ (4nm, 22.7B transistors) and a custom networking chip (5nm, >70B transistors, 91x91mm package).
17:38 In-house development of high-speed interconnect technology (SerDes) from 112G to 224G is highlighted as a critical technology bottleneck for AI.
17:49 The third pillar is Seamless Connectivity, essential for Edge AI devices to connect to the cloud anytime, anywhere, covering wide area (5G, satellite), local (Wi-Fi, Bluetooth), and wired networks (EPON/GPON).
19:03 Benefits of Edge AI computing are listed: better power consumption (due to form factor limits), efficiency for inference applications, reduced latency (on-chip), and enhanced privacy/personalization.
22:20 Demonstrations of generative AI capabilities on MediaTek-equipped phones are shown: 22:38 AI Eraser for removing unwanted objects from photos; 24:26 AI Laura Fusion for creating real-time avatars from video.
25:17 The potential of AI in Automotive cockpits, including AI assistance and entertainment, is reiterated through the partnership with Nvidia.
26:31 The fourth pillar, Strong Ecosystem Partnership, is emphasized, listing partners like Google, Meta, Alibaba, and Tencent.
28:37 Jensen Huang, CEO of Nvidia, joins Dr. Tsai on stage, highlighting the importance of the automotive partnership due to MediaTek's leadership in consumer electronics and Nvidia's expertise in automotive grade technology and software.
35:14 Jensen Huang discusses the future of computing shifting from retrieval-based to generative-based experiences, requiring powerful AI capabilities (GPU, NPU) on edge devices to generate content locally.
37:50 Huang also expresses support for MediaTek's Cloud Accelerator strategy and the potential for further collaboration using each other's IP and networking technologies (Infiniband, Spectrum-X).
40:01 Rene Haas, CEO of ARM, joins the stage with Dr. Tsai, discussing the long-standing and mutually beneficial 20-year partnership, the collaborative design process, and the goal of delivering great products together.
42:45 Haas stresses the critical need for a consistent software environment that allows models trained in the cloud to run efficiently on edge devices (inference), achievable only through partnerships like the one between MediaTek and ARM.
44:45 Dr. Tsai, Rene Haas, and Jensen Huang gather on stage for photos, symbolizing the strong industry partnerships MediaTek is forging.
45:54 Dr. Tsai concludes by thanking the partners and reiterating MediaTek's ambition, capability, and determination to enable the AI future together.