To give a quick spiel, my name is Matthew Zegar and I’m a third-year Software Systems student at Simon Fraser University. My previous internship experiences included working on a Marvel mobile game at Kabam and a mobile app in Flutter at HomeX. In my free time I spend quite a lot of time contributing to open-source which can all be checked out at my GitHub profile.
Towards the end of 2020 I was searching for the last place I wanted to do my final internship. It somehow ended up being at NVIDIA...
NVIDIA has all their internship roles posted on their Workday website. There are plenty of internships available across multiple countries and a wide variety of teams to apply for. As the SFU co-op coordinators always suggest, the secret is to just apply, apply, and apply.
Applying through Workday I managed to get a role as a PerfLab Intern which tasked me with testing the performance of NVIDIA GPUs (Graphics Processing Units). I was able to automate and improve the internal benchmarking software which assisted with testing NVIDIA GPU performance. There is quite a lot more to mention, however, I’m unfortunately unable to elaborate to avoid leaking any sensitive information.
I was encouraged by my manager to investigate and apply to the various other intern roles NVIDIA had to offer, so I did!
I ended up finding a Data Visualization role which worked on the open-source NVIDIA project RAPIDS. You can think of RAPIDS as a publicly available data science library that is GPU accelerated. It exposes data science libraries to a friendly Python interface which allows data scientists to pick it up with ease.
node-RAPIDS is essentially an extension of RAPIDS, however, instead you can interact with the data science libraries with Node.js bindings instead of Python. node-RAPIDS is the primary project I worked on during the summer and fall term of 2021.
One of the larger projects I owned was developing a separate SQL (Structured Query Language) module we’ve included inside of node-RAPIDS. We were able to hook up Node.js bindings to the open-source GPU accelerated SQL library BlazingSQL. You can think of BlazingSQL as essentially just SQL queries sped up massively by utilizing NVIDIA GPUs. This resulted in the ability to call into our SQL module from Node.js and process queries on multiple NVIDIA GPUs.
To showcase the full potential of this tech a simple demo was built. I downloaded the entirety of English Wikipedia and loaded it up into our SQL module. We were now able to search through English Wikipedia blazing fast. To give an example of usage, we could scan every English Wikipedia page for the word "Shakespeare" and see how many pages mention the playwright (in only ~40 seconds using only two NVIDIA GPUs!).
If anyone ever has the opportunity to work at NVIDIA, I’d say take it. NVIDIA is one of the top tech companies in the world working on some bleeding edge technology across multiple different subject areas. It was amazing being able to collaborate with talented people on a daily basis to build out some really interesting projects. I honestly didn’t think that I would be spending 2021 working at NVIDIA let alone working at NVIDIA on a completely open-source project.
