Apple's M4 processors hold significant computing power for AI operations, but the tech giant traditionally keeps its components under tight control. In practice, this means that app developers can only use the Neural Engine to infer or run pre-trained models, but cannot directly develop and train new algorithms on it from scratch.<\/p>\n\n\n\n
A researcher who goes by the online alias \u201c0x0SojalSec\u201d has published source code on GitHub that details how to harness the full potential of the silicon and achieve 15.8 TFLOPS of hidden computing power for machine learning purposes. While that number isn\u2019t a record today, the feat is remarkable because it was performed entirely outside of Apple\u2019s official development environment.<\/p>\n\n\n\n
Since the company\u2019s security settings do not allow direct communication with the neural engine for such advanced tasks, the project author had to find a way without using official software tools like CoreML or Metal, and he also could not rely on the GPU. Instead, he built his own intermediate language from scratch. This custom software solution acts as a bridge, allowing full backpropagation of errors and learning of transformer models directly on the neural chip.<\/p>\n\n\n\n
Due to the factory-limited hardware design, some extremely ingenious tricks had to be implemented to maintain the stability of the operating system. If a process gets stuck or freezes during the intensive learning phase, the custom programming language uses a special execution command to restart the process. In this way, the system refreshes its current state and continues processing data without causing the entire application to crash.<\/p>\n\n\n\n
A significant challenge in running such demanding workloads was also the speed of operation. To make the entire learning process run as smoothly as possible, the developer configured the system to write all data directly to the system's working memory. By deliberately avoiding the significantly slower flash memory, the entire operation remained extremely fast.<\/p>","protected":false},"excerpt":{"rendered":"
Procesorji Apple M4 skrivajo precej\u0161en ra\u010dunalni\u0161ki potencial za operacije z umetno inteligenco, vendar tehnolo\u0161ki velikan svoje komponente tradicionalno dr\u017ei pod strogim nadzorom. V praksi to pomeni, da lahko ustvarjalci programov nevronsko enoto (Neural Engine) uporabljajo samo za sklepanje oziroma poganjanje \u017ee nau\u010denih modelov, ne morejo pa na njej neposredno razvijati in od za\u010detka trenirati novih […]<\/p>","protected":false},"author":2,"featured_media":0,"comment_status":"","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[36],"tags":[698],"class_list":["post-7713","post","type-post","status-publish","format-standard","hentry","category-procesorji","tag-racunalniski-procesorji"],"acf":{"subtitle":"","heading":"","summary":"Raziskovalec je z lastnim vmesnim jezikom zaob\u0161el platformi CoreML in Metal ter na \u010dipu M4 odklenil 15,8 TFLOPS mo\u010di. Projekt dokazuje, da lahko nevronski motor na napravah Apple neposredno trenira nove modele umetne inteligence skozi sistemski RAM.","thumbnail_small":"https:\/\/racunalniske-novice.com\/wp-content\/uploads\/2026\/06\/Apple-M4-chip-badge-240507_big.jpg.large_-560x315.jpg","thumbnail_large":"https:\/\/racunalniske-novice.com\/wp-content\/uploads\/2026\/06\/Apple-M4-chip-badge-240507_big.jpg.large_.jpg","thumbnail_caption":"","gallery":"","video_gallery":null,"author":"","links":null,"sources":[{"title":"notebookcheck","url":"https:\/\/www.notebookcheck.net\/Developer-bypasses-Apple-s-restrictions-to-unlock-the-M4-s-true-AI-potential.1323683.0.html"}],"skip_language":[]},"yoast_head":"\n