Source: Nvidia

As the advances of Artificial Intelligence requires an unprecedented coordination between software and hardware industries, Taiwan’s strong hardware, especially semiconductor, industry as well as its proximity to the Chinese market can play a key role. 

At Computex 2021, a closer strategic alignment between Nvidia and Arm is taking shape, and this alignment includes some key Taiwanese players. 

Arm’s long road into data center

Following Arm’s announcement in March regarding its latest architecture, Armv9, Arm used the opportunity at Computex to reveal that Taiwan’s MediaTek would adopt the new architecture and deliver Armv9-based chips by the end of 2021. Simon Segars, Arm CEO, said that Arm’s Total Compute solutions have helped MediaTek to expand the reach of its mobile devices and into new markets, such as data centers. 

MediaTek has already made a few moves to enter the data center business, and the adoption of Armv9 also signified the architecture’s potential future in the Chinese market. 

Nvidia and Arm have been eyeing the data center market for a while, which cumulated into Nvidia’s ongoing attempt to acquire Arm. The market is currently dominated by Intel. 

Until the release of Arm’s Neoverse core in 2018, the company never had an IP core catered to the server market. Its Cortex series of cores had been used by data centers before, but the design wasn’t intended for such specific application. When Amazon acquired SoC maker Annapurna Labs in 2015, the resulting Cortex-based Graviton chip had been perhaps Arm’s most successful story in the data center market.

Amazon Web Services are currently the largest maker of Arm servers. If Microsoft Azure, the world’s 2nd largest cloud platform provider, adopted Arm design as rumored, it would expand Arm’s reach further. 

The release of Neoverse changed the trajectory and marked Arm’s dedication to be a provider of data center infrastructure. According to Arm, the Neoverse includes infrastructure-specific features that enable Arm’s ecosystem to build products for both cloud and edge computing. 

Armv9, the once-in-a-decade architecture update, accelerated Arm’s entry into the data center market. Neoverse N2 has become Arm’s first server CPU based on Armv9. 

Nvidia’s encompassing vision for data center

Neoverse N2 eventually enabled Nvidia to announce its first data center CPU, Grace, even though it won’t be released until 2022. In an interview during Computex 2021, Nvidia CEO Jensen Huang explained that Grace would be primarily used for two AI applications: natural language processing and cognitive computing. 

However, Huang believes that no single processor type will dominate the server market. Instead, he perceives a future of computing characterized by diverse processors, and that view is also reflected in Nvidia’s data center strategy to accommodate such diversity. 

Just like Arm, Huang considers Nvidia to be an IP company rather than a chip company. Therefore, Arm’s ecosystem can provide Nvidia a licensing channel, while extending Arm’s own reach as well. Such combination might just suit a data center future driven by multiple processor types.

Heading to that direction, Nvidia announced its “Nvidia-Certified Systems” under which enterprises can run their AI workloads on the hardware and software conformed to Nvidia’s design best practices. The Nvidia-Certified Systems were introduced with other products, such as Nvidia’s Ampere-based A100 Tensor Core GPU and BlueField-2 DPU. At Computex 2021, Nvidia revealed that its Certified Systems would even expand to support Arm-based CPUs as well. 

Enters Taiwan’s Gigabyte Technology 

In fact, Arm’s architecture has always been a missing puzzle in Nvidia’s data center grand vision, and Huang ultimately pursues AI’s democratization through the Nvidia-certified system. That’s where a Taiwanese company, Gigabyte Technology, entered the picture. 

Primarily a manufacturer dedicated to motherboard business, Gigabyte has also been making graphic cards for Nvidia, and it intends to expand into the data center market. At Nvidia’s GPU Technology Conference (GTC) this year, Gigabyte, together with Wistron Corporation’s cloud infrastructure-focused subsidiary Wiwynn Corp., announced its cooperation with Nvidia in developing a server platform that would incorporate one Arm’s Neoverse-based CPUs (Ampere Altra) and two Nvidia’s Ampere-based GPUs (A100) as well as two DPUs (BlueFeld-2).

The sever platform, named GIGABYTE G242, will be released in 2022. Of course, it will be Nvidia-certified, as revealed at Computex, and it will become a part of the Nvidia Arm HPC Developer Kit – a computing system for creating HPC, AI and other applications. At Computex, Gigabyte also elaborated on these new servers

Taiwan to have more role? 

It will be interesting to see if the data center vision of Jensen Huang, a Taiwanese-American, will further leverage Taiwan’s hardware prowess and eventually drive the country’s industry onto the road of hardware-software integration, a path deemed by the country’s technology industry to be the necessary path forward into an AI future defined by both hardware and software. 

In an age when data is perceived as the new oil, Taiwan might become a center of Asia-Pacific’s data-based economy, especially when both Google and Microsoft have committed to build their data centers in the country. 

Furthermore, in the face of China’s authoritarian interpretation of AI, Taiwan as a democracy should have a say in AI’s democratization, and Nvidia can play a role in this.


Source: ChinaTimes, StormMedia, EET-Taiwan, EET-China

More Related Articles