TL;DR:
- Ubitus, a Taiwanese cloud gaming provider, boasts a significant number of datacenter-grade GPUs, positioning itself as a prominent player in the “GPU-rich” category.
- SemiAnalysis categorizes computing power demand into GPU-Rich and GPU-Poor groups, with a few companies, including Ubitus, owning over 20,000 high-end GPUs.
- Ubitus collaborates with National Taiwan University to leverage its substantial GPU resources for technological contributions.
- Ubitus shifts resources towards developing localized Chinese language models, contributing to the growing GenAI landscape.
- Nvidia’s data center revenue experiences remarkable growth, driven by data center GPUs, while gaming revenue sees only marginal increases.
- The price gap between gaming GPUs and data center GPUs is substantial, leading Ubitus to focus on datacenter-grade GPUs.
- Ubitus chooses PCIe GPUs over server-based solutions for easier maintenance.
- Taiwan emerges as the fifth-largest mobile game market globally, with impressive revenue growth.
- Taiwan-based startup Wonders.ai introduces a virtual character application with voice dialogue functionality and a substantial model parameter count, independent of ChatGPT’s trend.
Main AI News:
As the field of Generative AI continues to gather momentum, it’s attracting a wave of enthusiastic newcomers. Yet, for many, the lack of computing resources and the challenges of data collection have acted as formidable entry barriers. Ubitus, a Taiwanese cloud gaming service provider, has recently unveiled a formidable advantage – a vast arsenal of tens of thousands of datacenter-grade GPUs. This revelation places Ubitus in the coveted “GPU-rich” category and opens up exciting possibilities for major gaming companies to become influential players in the advancement of Generative AI research.
In a recent article published by SemiAnalysis, the computing power demand has been categorized into two distinct groups based on GPU availability: GPU-Rich and GPU-Poor. The former includes a select few companies, such as OpenAI, Google, Anthropic, X, Inflection, and Meta, each owning more than 20,000 A100 or H100 GPUs. Some of these entities even boast GPU counts exceeding 100,000. This surge in GPU resources is particularly pronounced in the San Francisco Bay Area, where numerous teams proudly display their growing GPU holdings, signaling a burgeoning trend.
On November 8th, Ubitus took a significant stride forward by forging a collaboration with the Department of Computer Science and Information Engineering at the National Taiwan University. Wesley Kuo, the founder and CEO of Ubitus, revealed that the company likely holds the title of the largest GPU holder in Taiwan. These GPUs were originally intended to support cloud gaming computations, but their potential extends far beyond.
Kuo harked back to a visit from former Google Taiwan managing director Chien Lee-Feng and others who suggested that Ubitus, with its abundant GPU resources, should consider making substantial contributions to Taiwan’s technological landscape. This proposal emerged in early 2023, shortly after the meteoric rise in popularity of OpenAI’s ChatGPT.
Not merely content with their expertise in cloud gaming software development and deployment for both PC and mobile applications, Kuo’s team is strategically allocating surplus resources toward the development of large language models (LLMs). Their specific focus is on creating localized and traditional Chinese LLMs tailored for the Taiwanese market, with plans to extend support for other local languages in the near future.
While Nvidia chips have predominantly catered to data center servers and the gaming industry, the second quarter of 2023 saw their data center revenue skyrocket to an impressive US$10.32 billion, marking a remarkable 141% increase on a quarter-to-quarter basis and a staggering 171% growth compared to the previous year. In stark contrast, gaming revenue experienced only modest growth during the same period.
Experts within the Taiwanese industry, well-versed in Nvidia’s offerings, have highlighted substantial differences in specifications and unit prices between gaming GPUs and data center GPUs. For example, the RTX 4090 chip is a top-tier gaming card, while the H100 and A100 chips find their primary application in data center servers. Gaming cards typically start at NT$10,000, while data center GPUs span a wide price range from NT$100,000 to NT$1 million, resulting in a significant price gap.
In response, Kuo clarified that his team primarily possesses datacenter-grade GPUs, including H100, A100, and L40S, rather than consumer-oriented gaming cards. This decision is rooted in the inherent risk of gaming cards overheating when subjected to intensive computing demands.
Moreover, the team opted for GPUs with PCIe specifications instead of servers featuring sets of 8 GPUs. This choice was guided by the practicality of individual GPU repair, as opposed to the more complex and time-consuming maintenance required for servers with eight soldered-together GPUs.
As the tech landscape in Taiwan continues to evolve, Google Taiwan highlighted the nation’s ascent as the fifth-largest mobile game market globally. In 2022, only South Korea and Taiwan displayed growth in revenue, with Taiwan witnessing the fastest expansion rate.
Adding to this burgeoning ecosystem is Wonders.ai, a Taiwan-based startup and AWS cloud service user. In February, the company unveiled an impressive virtual character application boasting voice dialogue functionality and approximately 10 billion model parameters tailored for mobile use. Remarkably, Wonders.ai embarked on this venture nearly three years ago, independently of the ChatGPT trend, showcasing the diversity of innovation within Taiwan’s tech community.
In conclusion, the convergence of GPU-rich gaming companies and the burgeoning field of Generative AI in Taiwan holds great promise for the future of technology and innovation. With abundant resources and a strong focus on localized language models, these companies are poised to play a pivotal role in shaping the GenAI landscape, both domestically and internationally.
Conclusion:
The integration of GPU-rich gaming companies into the Generative AI landscape in Taiwan signifies a significant shift in the market dynamics. With abundant GPU resources and a focus on localized language models, these companies are poised to shape the future of GenAI innovation, paving the way for new opportunities and advancements in both the local and global tech markets.
