DeepSeek, the Chinese AI company, experienced significant delays in the development of its next-generation R2 AI model due to issues with Huawei's Ascend chips. Despite pressure from government authorities to utilise homegrown silicon, the project faced numerous hurdles including unstable hardware, sluggish interconnects, and immature software. These challenges ultimately prevented DeepSeek from completing a single successful training run.
According to sources cited by the Financial Times, DeepSeek enlisted the assistance of a dedicated team of Huawei engineers in an attempt to overcome these technical obstacles. However, the inherent flaws in Huawei's Ascend chips proved insurmountable, leading to a substantial setback in DeepSeek's development timeline. The company had previously made waves with the launch of DeepSeek R1 earlier in the year, increasing pressure to deliver a worthy successor.
Ultimately, DeepSeek decided to pivot and utilise Nvidia's H20 GPUs for training purposes. While the Huawei Ascend accelerators have reportedly been relegated to inference tasks, the initial failure represents a significant challenge for Huawei's aspirations in the high-performance computing and AI chip market. The company also faced challenges with data labelling during the development process. The delay of DeepSeek R2 underscores the critical importance of reliable and efficient hardware in the rapidly evolving field of artificial intelligence.
Artículos relacionados de LaRebelión:
- AI2s MolmoAct A 3D Robotics AI Model Challenging Nvidia and Google
- DeepMinds Genie 3 Creating Real-Time Interactive Simulations with a New World Model
- What is the tech stack for the shortern future?
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario