Login to Continue Learning
While the world looks towards silicon photonics for AI computing products, it seems that NVIDIA’s CEO isn’t too optimistic about this technology. Instead, he is betting on copper for now.
**NVIDIA’s CEO Wants to Rely On Copper For Several Years To Come, Avoiding Silicon Photonics**
For those unaware, silicon photonics refers to a next-gen adaptation of traditional copper transmission cables. It combines laser and silicon technology to provide high data transfer speeds. NVIDIA is using this technology to improve interconnect latency and create high-bandwidth connections between the CPU and GPU. However, Jensen isn’t too confident with silicon photonics packaging yet, as he reportedly wants to stick to traditional copper technology for years to come.
Jensen does say that they are working with TSMC to develop silicon photonics, but the technology seems several years away. NVIDIA is developing technologies around interconnects such as Quantum-X Photonics, a networking switch platform claimed to be released this year, and Spectrum-X Photonics, Ethernet switches expected to debut by 2026.
This shows that NVIDIA has positioned its portfolio around silicon photonics, but it remains hesitant to take the leap right now. Given the industry’s demand for computing, this approach makes sense. For Team Green, integrating silicon photonics within individual GPU tiles could prove effective if managed well, allowing for higher scalability and performance. However, this would require a ground-up architectural revision.
It will be interesting to see how NVIDIA and others in the industry utilize silicon photonics in actual architectures since this technology could bring massive computing power. We estimate full-scale solutions arriving on the market by the end of the decade, but for now, NVIDIA wants to rely heavily on copper interconnects.
