Leading Chinese cloud service provider Alibaba Cloud held its annual developer and tech event Apsara Conference on Nov. 3, revealing a series of products and services powered by its cloud services.
Three notable products were revealed during the conference, each demonstrating Alibaba Cloud’s move to develop open-source tech in AI and chip design. The firm also displayed its ambition to extend its cloud services to power more consumer devices such as laptops and desktops.
Established in 2009, Alibaba Cloud is the cloud arm of the e-commerce giant Alibaba. It has been a top three public cloud infrastructure as a service (laaS) service provider since 2018, according to IDC.
Wuying Cloudbook
Alibaba Cloud revealed a series of new products based on Wuying desktop-as-a-service (DaaS) architecture. One of the highlights was the Wuying Cloudbook, an ultra-thin laptop that can run multiple operating systems, including Linux, Windows, and Android. The laptop features a 14-inch touchscreen and weighs 1.29 kilograms, a similar setup to the new 14-inch MacBook Air. The display has a 2k resolution with a 16:10 ratio.
All apps on the laptop are deployed remotely on the cloud, lowering the requirements for the device itself and saving battery life.
Alibaba claims the Cloudbook can last up to 20 hours when handling lightweight work and 12 hours when rendering heavy work or playing high-quality videos.
For performance, it can obtain up to 512-core virtual CPUs from Alibaba Cloud, with a Cinebench mark of 120,000. Theoretically, the laptop can acquire unlimited storage volume. It also comes with the support of 4G LTE and Wi-Fi 6. It is equipped with an HDMI video port and two USB Type-A and Type-C ports, separately.
Power-efficient processor
T-Head Semiconductor, Alibaba’s chip-design arm, released a new processor called Xuantie C908 during the conference. The chip enriches the Xuantie family with a focus on power efficiency.
The new processor is based on open-source architecture RISC-V and is targeting the mid-end market, hoping to plug the gap between the C910 and C906.
The Xuantie C908 adopts the RISC-V Vector 1.0 standard and adds support that optimizes AI computation. The computation capacity is about 50% better than older models in AI application scenarios like image classification and keyword spotting, according to data provided by Alibaba.
The new chip can run at up to 2 GHz frequency, thanks to the efficient pipeline design. Built with TSMC 12nm process, the processor’s dynamic power consumption can be 52.8 mW per GHz for each core. The energy efficiency ratio of XuanTie C908 in typical scenarios can be improved by over 20% compared with that of XuanTie C906 under the same frequency and process constraints.
For applications such as intelligent interaction, AR, and VR, the processor also has an AI acceleration engine.
Founded in 2018, T-Head Semiconductor is a wholly-owned semiconductor chip business entity of Alibaba Group, according to their official website. The firm focuses on products including AI chips and CPU Processor IP, covering end-to-end chip design processes.
Open-source AI models platform
ModelScope, an open-source AI modeling platform, was also launched by Alibaba DAMO Academy, the firm’s research and AI arm.
Over 300 “ready-to-deploy” AI models will land on the platform, covering all fields, from computer vision to natural language processing. The platform already has over 150 well-recognized models. Alibaba also offers some of its own pre-trained AI models, such as Tongyi, which can generate images from text prompts.
Profitability is not the platform’s first priority, according to Alibaba. Instead, the tech giant intends to make AI models more accessible and effective through the open-source platform and community.
On ModelScope, developers and researchers can test models online, customize them, and deploy them on Alibaba Cloud, in a local environment, or on other platforms.
With the support of multiple mainstream AI frameworks, the platform will stay neutral and open to all developers, Alibaba has stated. Users can decide how to train and develop these models.
Correction: An earlier version of this article misfiled Xuantie C908 as Xuantie C098.