Q: Where did you go to school and when did you know you wanted to be an engineer?
A: I went to Stanford for undergrad and CMU for graduate school. In between undergrad and grad degrees, I worked for a few years at Intel.
I’ve always been interested in science and engineering but knowing the specific area took a few years of exploration through course work, research in graduate school, and practical industry experience. I’ve loved working as a performance engineer straight out of graduate school; to me it’s a great blend of the theoretical and practical ‘hackery’.
Q: What attracted you to Cerebras and what do you love the most about working here?
A: What attracted me to Cerebras is the innovative system and all the new performance issues and optimizations that would come with a novel, massively parallel system.
What I love most about working here is the non-stop learning and so many chances for being the first to come up with novel ways to solve a new problem.
Q: What are you working on today?
A: I am working on making the CS-1 perform faster! My focus is on figuring out the bottlenecks in our system due to how software interacts with the hardware, coming up with visualization tools to pinpoint bottlenecks easily, work with microcode writers to improve the algorithms, and with hardware architects so the next-gen hardware can avoid some of these bottlenecks.
Q: What has been your most rewarding and challenging project?
When I first came to Cerebras there weren’t any tools to help diagnose performance issues in our system, certainly not at the scale at which we operate! I wrote a suite of tools which help analyze and visualize simulator traces, microcode, and output from hardware to pinpoint bottlenecks and help us focus on the issues with the most impact.
In the process I have enjoyed working with many different groups and of course seeing the performance improve is super exciting!