AMD just signed the deal to acquire Canada’s AI chip startup Taalas—aimed at supercharging AMD’s AI inference tech and product roadmap. No deal value was disclosed, and the acquisition still needs regulatory approval, but the strategic “landing spot” is crystal clear. With AI apps exploding, the real battlefield is no longer only training speed—it’s inference cost and energy efficiency. Taalas, founded in 2023 in Toronto by ex–Tenstorrent CEO Ljubisa Bajic and others, takes a refreshingly specific approach: instead of relying on general-purpose GPUs for many models, they build dedicated chips for particular AI models. Their method maps a model’s computation structure directly into chip circuits, reducing the performance hit from constant data movement between processors and memory. The payoff: lower inference costs and less power use. Taalas says it can analyze an already-trained model and convert it into an optimized custom chip design in about two months. A demo chip (HC1) was customized for Meta’s Llama 3.1 8B. AMD plans to bring this into future AI accelerators and pair it with its Instinct GPU ecosystem—software, memory, and now specialized inference hardware. #AMD #AIInference #Semiconductors #GenAI #HardwareAcceleration #EdgeAI
Want to learn more? Visit Explore the world, stay updated on travel insights and international affairs, and discover authentic stories from real life
评论
发表评论