Employee Spotlight: Mark Khait

Mark Khait, Senior Computational Scientist, joined SRT in Spring 2022 and works from our Milan office. Mark gained his PhD in Petroleum Engineering from TU Delft in 2019

Emily: You started working with SRT as an Intern in 2019. What was your first project when you arrived?

Mark: Firstly, it is important to mention that Stone Ridge Technology (SRT) was the only company commercially promoting full GPU reservoir simulation at that time. I was writing a reservoir simulation code on GPU for my PhD so I hoped to be involved in source code development.

However, my task as an intern was not about reservoir simulation but geological simulation so I essentially ported the so-called ‘Sequential Gaussian Simulation’ to GPU that Spring. During my tenure I realized that this algorithm is not as sequential as it seems. The work resulted in a conference paper ‘Exposing Fine-Grained Parallelism in Sequential Gaussian Simulation’ and I was honored to represent both my university and SRT at a conference (The fourth EAGE workshop in HPC 2019) in the UAE. Being an intern at SRT was a big deal for me.

Emily: You mention your PhD, please tell me more about it.

Mark: I started my PhD back in 2016. I was convinced that GPUs should be part of it; and I was pretty lucky because the topics suggested by my supervisor, Dr. Denis Voskov, complimented the GPU context. Essentially we wrote, almost completely from scratch, a reservoir simulator code called DARTS which stands for Delft Advanced Research Terra Simulator. Like many modern machine learning frameworks, it is implemented as a Python module, and therefore very flexible. On the other hand, it runs on both GPU and CPU quite efficiently.

A novel example of how flexible it is, despite being called a ‘terra’ simulator, it was recently used for modeling water extraction from lunar soil. I just recently heard it from my supervisor. DARTS now serves as a platform for many other PhD and postdocs at Delft University.

Emily: What led you to focus on GPUs? When did you first hear about them and feel they were something that you would like to work with?

Mark: In 2010, I was working in the Rosneft oil company as a part of the team developing a proprietary reservoir simulator. Back then, we were looking at different hardware platforms as candidates for future simulators - and GPUs were among them. We tried it out with a very basic core algorithm for the linear solver called Sparse Matrix Vector Multiplication. As the difference between CPU and GPU performance was about an order of magnitude, we were pretty happy with the results. What makes it really fascinating, despite both CPU and GPU industries having gone a long way since then, the difference is still about an order of magnitude.

At that time, I was convinced that the GPU platform was worth going all in. SRT was already miles ahead, because they already had a black oil simulator, ECHELON! But Rosneft management was not so sure about GPUs, so we continued our traditional development. However, whenever I got an opportunity to make choices myself, I opted for GPUs with no regrets.

Emily: Where do you see the reservoir engineering industry in 10 years?

Mark: When I was a kid, I imagined that robots would take over the most boring part of my everyday (future) job. Right now, when looking at the pace of AI development, I am pretty sure that in 10 years the typical reservoir engineer won’t have to deal with text files. One wouldn't have to know every keyword, and maybe wouldn't even have to think about which simulator would take on the job required, generative AI will do that part. I imagine, then, the rest of the role would be an engineer using voice commands, formulating the task and how the data should be interpreted, or choosing the type of optimization which needs to be undertaken. Future reservoir engineers will no longer be figuring out which keyword to use and how but will likely be further educated on the physics side of the role, they will (hopefully) have the luxury to prioritize what actually happens in the subsurface. Therefore I am positive about the role of reservoir engineer in the future because it will have more emphasis on fundamentals and less on the details of a particular implementation.

Emily: Would you like to share some thoughts about how the energy industry will be in 10 years?

Mark: I don't know about the whole energy industry. New ways of energy generation are being adopted, maybe in 10 years we will start thinking about not only how the energy should be produced, but also to where? To minimize the distance between production points and utilization points of energy. So there will be more importance on ‘how do we optimize energy transport across the globe?’

Emily: And to finish our interview today… What do you have on your desk right now?

Mark: Right now, thanks to the size of my desk, I have a desktop and a laptop but also my passport, my mobile phone, a bunch of paper to write on and a few nice gifts from my kids.