Chris Buckminster’s Post

We at DeepSig, Inc. have just published the details behind the CUDA-accelerated OCUDU L1. It is in the open GitLab today, with the dApp/E3 framework close behind. → Several-fold to ~10x speedups on the heaviest shared-channel blocks; ~100x on fronthaul compression → BLER within 0.06 dB of the CPU implementation. Zero accuracy traded for throughput → One codebase from a laptop CPU to a GB10 or GH200 rack → A neural equalizer, AI scheduler, and spectrum/ISAC sensor running concurrently over the air under live traffic AI has always been kept out of the RAN's real-time core. This is what it looks like when that changes, using open, portable interfaces any vendor can build against. I am proud of what the team and the OCUDU community shipped here! The next AI-RAN Working Group call is August 5 if you want in. https://lnkd.in/egzjBWVf #AIRAN #OpenRAN #AI #6G