Highlights:

  • In 2021, Tenstorrent, a company based in Toronto, achieved a valuation of USD 1 billion following a successful funding round that garnered over USD 200 million from investors.
  • Tenstorrent has partnered with LG to collaborate on chiplets, which are modular semiconductor components utilized in the construction of processors.

Semiconductor startup Tenstorrent Inc., helmed by renowned engineer Jim Keller, has partnered with LG Electronics Inc. to create a series of innovative chiplets.

Recently, the companies unveiled their partnership, aiming to leverage the chiplets resulting from their collaboration in developing processors for data centers, smart TVs, and vehicles.

In 2021, Tenstorrent, a company based in Toronto, achieved a valuation of USD 1 billion following a successful funding round that garnered over USD 200 million from investors. The company is headed by CEO Jim Keller, a renowned chip engineer who previously served as Intel Corp.’s vice president of silicon engineering. Before that, he held senior technical positions at major tech companies, including Apple Inc. and Advanced Micro Devices Inc.

Tenstorrent has created Grayskull, a specialized chip designed to run artificial intelligence models. The company claims that Grayskull is simpler to program than graphics cards and provides additional advantages. In addition to Grayskull, Tenstorrent offers a dedicated central processing unit (CPU) tailored for AI server applications.

Tenstorrent has partnered with LG to collaborate on chiplets, which are modular semiconductor components utilized in constructing processors. These chiplets consist of CPU circuits designed for general-purpose computing and specialized optimizations for specific use cases, such as AI model execution.

In the semiconductor industry, it is common for companies to integrate multiple chiplets into a single processor. Major market players like Intel, AMD, and several others have already adopted this technology as the foundation for their latest product offerings.

In collaboration with LG, Tenstorrent intends to create chipsets catering to three applications. The partnership aims to develop compute modules optimized for running AI models and video codecs essential for video file format conversion. Furthermore, the companies plan to manufacture CPU chiplets based on the increasingly popular RISC-V processor architecture, known for its open-source nature.

Chief Technology Officer of LG, Byoung-hoon Kim, stated, “This collaboration is just a beginning. Tenstorrent’s market leading AI and RISC-V CPU technologies will strengthen SoC competitiveness of LG’s future products while our long-time proven video codec technology will help Tenstorrent take control of data center high-performance processor markets.”

As a leading TV manufacturer, LG will examine the feasibility of integrating the jointly developed chiplets into its displays. Additionally, LG will assess the potential utilization of these chiplets in the automotive sector. With its automobile division, LG Mobility, specializing in developing vehicle components like infotainment systems, the company aims to explore the applicability of the chiplets within this domain.

Under the partnership agreement, LG will benefit from Tenstorrent’s AI and RISC-V CPU technology, while Tenstorrent will gain access to LG’s video codec technology. Video codecs are instrumental in modifying the format of multimedia files to optimize storage space and bandwidth utilization.

Tenstorrent states that incorporating LG’s technology will enhance the video processing capabilities of its upcoming data center products.

Tenstorrent introduced its most advanced data center AI module, the Grayskull e300 accelerator card. The company claims to achieve impressive speeds of up to 1700 TOPS (trillion operations per second) with specific software optimizations.

The Grayskull e300 card incorporates two of Tenstorrent’s AI-optimized Grayskull chips, built on the company’s internally-developed Tensix core design. Each Tensix core comprises five RISC-V CPU modules and circuits optimized for vector and tensor operations. These optimized circuits enable the efficient processing of data by AI models.

Tenstorrent has incorporated a technology called conditional computing into its chips. This technology enables specific AI tasks to be executed by activating only relevant parts of a neural network rather than running the entire code. As a result, infrastructure requirements are reduced, leading to improved efficiency and resource utilization.

Keller added, “It is increasingly important for industry leaders to own their silicon future. LG is a giant in our industry, and this collaboration will strengthen their portfolio of technologies for their future chip solutions, providing greater flexibility to differentiate their products.”