Procero: Pioneering Efficiency in AI and Graphics Processing
Procero is an innovative software technology company based in Montreal, a city renowned for its deep learning ecosystem and vibrant AI research community. Drawing from this tradition, Procero has developed a unique technology that currently manifests as software, with plans for silicon implementations in the future. This technology algorithmically accelerates key processing classes that support Generative AI (GenAI) workloads, including large language models (LLMs) and various graphics applications.
Technology Foundation
Richard Reiner, a board member of Procero, told the audience at the Canada-Taiwan Co-Innovation Forum in Hsinchu on September 6 that the foundation of Procero's technology consists of more than 40 person-years of research conducted by three PhD-led teams, both within the company and in collaboration with two research universities. With four issued patents and three more pending, the company demonstrates a strong commitment to innovation.
The core offering, Procero for GenAI, focuses primarily on enhancing LLM inference, with applications in training currently under development.
At its core, Procero's technology is founded on advanced mathematical principles. The system transforms data structures into a format that allows for more efficient computation. This process includes a preprocessing step that optimizes data structures, followed by a runtime module that applies inference on these modified structures.
Expanding Applications
Procero's technology is versatile, powering various applications including AI inference on devices, where customers seek improved user experiences with lower latency and faster throughput. In cloud environments, while providers achieve necessary performance levels, they often do so at high costs—both financial and in energy consumption. Procero's solutions can significantly reduce energy expenditures, enhancing efficiency in AI training and inference.
In gaming and XR applications, both on-device and in the cloud, Procero's technology delivers significant performance enhancements, including faster graphics rendering and efficient simulation workloads. Notably, the technology operates effectively on standard CPUs, GPUs, and NPUs, requiring no special hardware modifications. This makes it suitable for immediate integration with existing systems and applications.
Ready for Market
Currently, Procero's deployment consists of a purely software-based proof-of-concept implementation, featuring a preprocessing component and a runtime component. However, the company foresees a significant shift toward silicon-based production for the runtime component, partnering with manufacturers to embed this technology into various chipsets.
Procero stands out from competitors by operating on existing processors, ensuring compatibility without necessitating new hardware designs. Users can leverage current LLMs, such as downloading Llama 3, to benefit from Procero's acceleration capabilities. The focus is pragmatic; Procero aims to enhance performance on existing devices rather than developing custom models or untested inference engines.
Partnering for Progress
Procero is committed to delivering tangible benefits to its customers and partners, significantly reducing costs and energy consumption for LLM inference in cloud environments. The technology enables partners to enhance the performance of current hardware ahead of new releases, improving both efficiency and power budgets.
In addition to partnerships for existing products, Procero is actively seeking strategic collaborators poised to make an impact in the machine learning sector. The company aims to expand its technology's reach by optimizing more families of CPUs, GPUs, and NPUs, as well as supporting additional LLM configurations and gaming engines.
Future Outlook
As Procero continues to evolve, the focus will remain on both graphics and LLM markets, adapting rapidly to the fast-paced developments within these sectors. The company is dedicated to advancing its pioneering technology to address diverse workloads efficiently, solidifying its position as a leader in power-efficient processing solutions for AI and graphics applications.
Procero's ongoing collaboration with the Montreal AI community, along with its commitment to innovation, positions the company to play a significant role in the future of machine learning and graphics processing. With a clear vision and strong research foundation, Procero is ready to seize the opportunities that lie ahead.