[SpaceX's Self-Developed C-Language AI Training Framework Nearing Completion, To Be Used for Training Grok v5] According to monitoring by Beating, Elon Musk revealed that SpaceX's internal team is nearing completion of version 1.0 of a self-developed, large-scale AI training framework written in C language. The new training stack is designed to precisely adapt to an ultra-large-scale computing cluster composed of 220,000 NVIDIA GB300 accelerator cards and 800G network cards, with optimized hardware mapping and parallel mechanisms. To fully exploit underlying computational power, the framework is designed to operate extremely close to the bare metal layer and deeply incorporates pipeline parallelism technology. Musk disclosed that for handling ultra-large-scale training tasks, the C-language-based architecture has the potential to achieve a runtime speed improvement of over an order of magnitude (i.e., more than 10 times) compared to Google's mainstream advanced AI framework JAX. The new self-developed training stack will run on SpaceX's Colossus supercomputing cluster, directly supporting the full-scale training and iteration of the next-generation large model Grok v5. [Original Link]