Ankan Banerjee
Distinguished System Software Engineer at NVIDIA
Karnataka, India
Summary
🤩
Rockstar🎓
Top SchoolAnkan Banerjee is a Distinguished System Software Engineer with 13 years of experience designing high-performance GPU drivers and compute systems, currently leading prototype efforts at NVIDIA. He has a strong track record across GPU generations—from Fermi to Ampere-era projects—shipping features like ML inference support via Direct3D, MFAA, DSR and custom HLSL extensions, and mentoring teams to deliver production driver capabilities. His passion for squeezing performance from hardware extends into open source: he implemented a cuDNN/CUDA-based neural network evaluator for Leela Chess Zero that outperformed TensorFlow on consumer GPUs. Based in Karnataka, India, Ankan blends low-level driver expertise with practical ML acceleration know-how, making him adept at bridging architecture, driver internals, and applied machine learning.
13 years of coding experience
19 years of employment as a software developer
Kendriya Vidyalaya SangathanBE, Computer Science, BE, Computer Science at AIT, Pune
English, Hindi, Bengali
Github Skills (13)
neural-network10
kernel10
cuda10
machine-learning10
c-language10
deep-learning10
gpu-acceleration10
gpu10
cprogramming-language10
performance-optimization10
cudnn10
chess-engine10
deep-reinforcement-learning9
Programming languages (5)
Github contributions (5)
LeelaChessZero/lc0
Apr 2018 - Dec 2022
Open source neural network chess engine with GPU acceleration and broad hardware support.
Role in this project:
Backend Developer & Performance Engineer
Backend Developer & Performance EngineerContributions:36 reviews, 65 commits, 48 PRs in 4 years 8 months
Contributions summary:Ankan focused on optimizing the Leela Chess Zero engine by implementing a CUDA-based neural network evaluation method. They replaced the TensorFlow implementation with a cuDNN-based approach, leading to significant performance improvements, specifically benchmarking 50% faster on a GTX 970. The user also made subsequent changes by refactoring and refactoring the implementation by integrating Batch Normalization into convolutions.
pythonrewrittenucichess-enginebackends
glinscott/leela-chess
Apr 2018 - May 2018
**MOVED TO https://github.com/LeelaChessZero/leela-chess ** A chess adaption of GCP's Leela Zero
Role in this project:ML Engineer
ML EngineerContributions:33 commits, 6 PRs, 16 comments in 9 days
Contributions summary:Ankan implemented and optimized a CUDA-based network evaluation system for Leela Chess Zero, significantly improving performance compared to a TensorFlow implementation. Their work involved modifying the `network_cudnn.cu` file to leverage cuDNN, including creating and configuring various cuDNN objects and kernels. The user's commits highlight their focus on utilizing GPU acceleration for neural network inference within the chess engine. This optimization resulted in substantial speed improvements.
chess-enginegcpzerouci
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.