ECOC 2023: Sivers Photonics showcases DFB laser arrays for 4T GPU-GPU communication
Sivers Photonics at ECOC 2023
Sivers Photonics, a subsidiary of Sivers Semiconductors, is hosting a joint live demonstration at ECOC this year with Ayar Labs.
The companies will showcase the 8-wavelength distributed feedback (DFB) laser array from Sivers Photonics, integrated into the Ayar Labs SuperNova multi-wavelength optical source.
The demonstration is located at Sivers’ booth, #615 and the joint technology between both companies is designed to support generative AI applications. As foundational AI models have become larger and more complex, they require more GPU processing and memory capacity. Integrating optical I/O within the GPU package to communicate with the rest of the GPU cluster, can allow for efficient scaling of infrastructures to handle growing AI demands. By using optical connections instead of electrical, power consumption can be lowered by up to 90% while allowing for 5-10 times better communication within the AI cluster to increase use of all GPUs.
Andy McKee, Sivers Photonics Interim MD and CTO says: “We are excited to have this prestigious conference take place in Glasgow, offering us an incredible platform to showcase Sivers’ expertise on our doorstep. We have an exciting week planned, continuing our long-standing partnership with Ayar labs, with the next generation of our advanced arrays demonstration taking place on our booth, as well as showcasing our recent work on our advance arrays with a talk at the Market Focus stage.”
Anders Storm, Sivers Semiconductors Group CEO adds: “Electrical communication in generative AI clusters is getting to its limit, with regards to speed, heat and capacity. With Sivers laser array we will be able to allow for 5-10x better performance with up to 90% less power consumption to give generative AI clusters the performance needed. With Sivers lasers integrated in Ayar Labs SuperNova multi-wavelength optical source, we are working together to solve one of the main challenges for GPU to GPU communication to support the future revolution of generative AI applications.”