VSORA is a French company providing high-performance, low-power silicon solutions for AI and signal processing. Our performance for very large LLMs used in Generative AI is untouched. However, our solutions can be used in other applications as well, for example, autonomous driving (AD) and ADAS, where they help reduce the entry barrier enabling L4/L5 autonomous driving; robotics, augmented / extended / virtual reality; drones; Industry 4.0 etc. Our solutions are configured to work as companion platforms and they are algorithm agnostic and host processor agnostic.

Our unique architecture provide a CUDA-free development and almost unlimited compute power for both AI and signal processing at a very low power consumption (with a small silicon footprint). In large LLMs, like GPT-3.5 or GPT-4, this means fast time to first token and low latency coupled with low investment in infrastructure.

In autonomous driving this solves the “power gap” issue that is plaguing the existing solutions on the market, which effectively limits them to L2 or parts of L3 functionality. Our solution is to provide a companion compute chip that offloads the heavy AI and signal processing algorithm computing from the existing (and new) vehicle compute systems – e.g. central compute or sensor hubs. A complementary solution that will allow the car makers to rapidly progress to L4/L5.

Our powerful multi-core architecture eliminates the need for other co-processors and hardware accelerators to provide a superior level of flexibility and by using high-level programming language any need to write low-level code is eliminated. Our state-of-the-art design process and architectures optimizes the computing power in a CUDA-free environment while simultaneously optimizing power consumption,  system performance and hardware cost.

Based near Paris, France, the company was founded in 2015 by a team of highly qualified and accomplished DSP experts and entrepreneurs.