AI chip startup Taalas @taalas_inc is showing off a chip that can do 16,000 tps/user on Llama3.1-8B, many multiples of its nearest competitor. The catch? The chip ONLY runs Llama3.1-8B, and a model like DeepSeekR1-671B would need 30 separate tapeouts: https://t.co/IJuprQZqaE
South Korean AI chip startup HyperAccel @HyperAccel0112 is preparing to launch Bertha 500, an LLM inference accelerator aimed at cost-efficient, low-latency token generation. The company also plans an edge SoC version in partnership with Korean giant LG: https://t.co/TRgQrMNik2
Former Altera CEO Sandra Rivera has joined French AI chip startup @vsora_dsp as chair of the board. I had a chat with her about raising Vsora's profile, raising funding, and competing with Nvidia, AMD and the entire landscape of startups...