Powered by the Aurora2 Optical Processing Unit (OPU) reaching a maximum capacity of 1.5 PetaOPS at 30 W TDP (non-linear Random Projections), LightOn Appliance is the most advanced photonic AI/HPC co-processor on the market today.

LightOn Appliance enables Machine Learning at scale. By harnessing the unique capabilities of photonics is able to perform truly massively parallel computations, where million-size vectors can be processed at once.

LightOn Appliance is the first commercial photonic co-processor built upon a non-von Neumann computing paradigm. LightOn Appliance leverages photonics to process large-scale generic data into information-revealing “sketches”. This opens new avenues ranging from fundamental concepts in mathematics and machine learning to alleviating excruciating bottlenecks in production for data science, HPC, and AI alike.

Max Input size: 1 million (1 039 680) up to 6-bit

Max Output size: 2 million (2 267 392) 8-bit

Peak batch-mode performance (non-linear RPs): 50 TOPS/W* (up to 1 900 dense matrix-vector multiplications per second)

Peak batch-mode performance (linear RPs): 10 TOPS/W* (up to 360 dense matrix-vector multiplications per second)

Frameworks compatibility: PyTorch 1.0+, scikit-learn

Peak TDP: 30 W

Server link: External PCIe Gen2 x4

Form factor: 2U

Power supply: 125/250V 50/60Hz

*Future updates will improve performance