Kalray Speeds Up Its Data Processor

Kalray's Coolidge v2 chip targets network processing for the storage market. It’s based on the company’s massively parallel processor array (MPPA) architecture; it’ll enter production in 2024.

Anand Joshi

Kalray’s latest data-processing unit, Coolidge v2, improves matrix performance over the previous generation. It more than doubles INT8 speed and increases floating-point execution eightfold, enabling data operations such as digital signal processing near storage.

The company performs computing using its proprietary massively parallel processor array (MPPA). This architecture allows for higher programmability and predictable response time compared with chip architectures that use a small number of Arm processor cores combined with hardware accelerators.

CEA, the French equivalent of the US Department of Energy, spun Kalray out in 2008. It is now a public company listed on the Paris stock market, having raised €48 million via an initial public offering in 2018. The company has been shipping products based on its MPPA processor since 2012. It reported revenue of €15 million in 1H23. In April 2023, the company announced a design win with a US-based customer worth €10s of millions. It generates over 90% of its revenue from the storage market.

Data-processing units (DPUs) enable in-line network-function acceleration, relieving the CPU load. They’ve found widespread success in servers, with industry stalwarts such as Nvidia and Marvell offering them. Kalray primarily targets data-center storage, with AI and 5G as secondary markets. The company expects chip samples by the end of 2023 and plans to qualify the chip for production in 2024.