UALink delivers 2.0 spec before v. 1.0 silicon ships
UALink Consortium aims to create an open standard for GPU networking to rival Nvidia's offerings, with new specs released but silicon shipping months away.
Copackaged optics have officially found their killer app
Co-packaged optical (CPO) switching is being adopted to cut power and enable extremely high-speed GPU network ports, solving AI datacenter scaling and cabling limits.
Perplexity's open-source tool to run trillion-parameter models without costly upgrades
TransferEngine enables full-speed GPU-to-GPU communication across AWS and Nvidia hardware, letting trillion-parameter models run on older H100 and H200 systems.