AMD Acquires Taalas to Etch Model Weights Into Silicon, 48x Speedup on Llama
petrusenko_max · x · 2026-08-08
AMD has acquired AI chip startup Taalas to significantly boost AI inference efficiency by etching model weights directly into silicon.
Reportedly, this approach achieves a 48x speedup when running the Llama 3.1 8B model. Taalas's new HC2 chip is designed to handle a single 20-billion-parameter model, requiring only 50 such chips to support trillion-parameter models, making large-scale AI infrastructure far more space- and power-efficient.
More from Companies & People
- ML researcher Emtiyaz Khan moves to TU Darmstadt, hiring PhD students via ELISS — EmtiyazKhan · 2026-10-16
- Tripo hosts first world-building hackathon in SF with World Labs and Runway talks — mhdfaran · 2026-10-16
- Dev warns enterprises: OpenAI and Anthropic products are an IP security risk — examachine · 2026-10-16
- a16z podcast: Can AI run the systems after writing the code? — a16z Podcast · 2026-10-16
- Zuckerberg backs electrician training school as US datacenter construction outpaces offices — emmanuelvivier · 2026-10-16
- awesome-ai-engineering: a free GitHub repo packing the full AI engineering learning path — bigaiguy · 2026-10-16