Generative AI is enabling transformative outcomes, but model complexity is growing exponentially. Traditional interconnects create a bottleneck for data, forcing GPUs to remain idle a majority of the time. We need a new approach to generative AI architecture: optically connect nodes at scale so they effectively work like a single, giant GPU.
- Sivers Photonics demonstrates next-generation DFB laser arrays supporting 4 Terabit GPU to GPU communication for Generative AI applications at ECOC 2023 in Glasgow, Scotland
- It’s Photons Over Electrons in Effort to Eke Out Better Chip Performance
- Road to Commercialization for Optical Chip-to-Chip Interconnects
- The FPGA Turns to Optical Interconnects
- Ayar Labs + Intel