Contrary’s Post

After 41 hours of researching & writing my 1st memo for Contrary Research on Cerebras Systems, it is now officially published! At 12,000+ words with 800+ reference links, it is the most detailed memo EVER published by Contrary 🤯 As a semiconductor researcher, I chose to write about Cerebras – an AI chip company that's gunning for NVIDIA. Here's some INSANE feats accomplished by Cerebras that blew my mind: 🧠 Cerebras' AI chips are 56x larger in size & contain 50x more transistors than an Nvidia GPU. 🕔 Cerebras' systems can train a Llama2-70B model from scratch in < 1 day. 🖥️ A Cerebras cluster of 48 CS-3s exceeds the performance of the U.S. Frontier (the world's fastest supercomputer). ⚡ Cerebras' inference is the fastest in the industry, running Llama3.1 8B at 1,800+ tokens/s! If that sounds insane, here's the full memo that dives into the narrative around Andrew Feldman and his company: https://lnkd.in/e5QRH6hs Thanks to the Contrary Research team (especially Sachin Maini and Kyle Harrison) for giving me this incredible opportunity to write and publish this memo and to Patrick Hulce for helping me understand ML better as a non-ML scrub. #MachineLearning #AI #Supercomputing