Andrew Feldman: Cerebras and AI Hardware

Pharma Revolution

AI accelerators are transforming pharmaceutical research by enabling rapid data analysis and experimentation. Andrew Feldman highlights the collaboration with GlaxoSmithKline (GSK), where Cerebras Systems utilized a version of BERT called Ebert to analyze genetic sequences, significantly speeding up the research process 1. This advancement allows scientists to explore questions previously deemed too time-consuming, as Feldman notes, "Taking 25 days on a 16 node cluster is too long to ask many interesting questions."

What Galaxy showed was that our system was for work they used to do on a 16 node GPU cluster in 20 days we did in about two days.

--- Andrew Feldman

The ability to process large datasets quickly opens new avenues in life sciences, potentially revolutionizing how pharmaceutical companies approach drug discovery and development 1.

   

Industry Impact

The impact of AI accelerators extends beyond pharmaceuticals, influencing various industries through rapid iterations and advancements. Andrew Feldman discusses the challenges faced by Cerebras Systems in competing with established players like Nvidia and AMD, emphasizing the importance of patient customers who believe in their vision 2. He explains, "You need to find some customers who have vision and who want to bet on your trajectory, not just on where you are today."

Many of the guys who'd been with us at C Seamicro, we started meeting, and as computer architects do, we say, is there a new workload? Could we build a better machine for it?

--- Andrew Feldman

Feldman also highlights the serendipitous role of GPUs in AI development, noting that their architecture, initially designed for graphics, inadvertently suited AI workloads 3. This realization led to the creation of a new machine, capable of handling complex AI tasks more efficiently, thus paving the way for further innovation across scientific fields 3.