Senior Deep Learning Compiler Engineer - XLA

NVIDIA•Redmond, WA

1d•$152,000 - $287,500

About The Position

NVIDIA is seeking versatile software engineers for its XLA team to build high-performance, production-grade software at the core of next-generation AI systems. This role involves developing compiler optimization algorithms for deep learning workloads, optimizing inference and training performance for the JAX framework and the OpenXLA compiler on NVIDIA GPUs at scale. The engineer will collaborate with deep learning framework teams and hardware architecture teams to accelerate next-generation deep learning software.

Requirements

Bachelors, Masters or Ph.D. in Computer Science, Computer Engineering, related field (or equivalent experience).
4+ years of relevant work or research experience in performance analysis and compiler optimizations.
Ability to work independently, define project goals and scope, and lead your own development effort adopting clean software engineering and testing practices.
Excellent C/C++ programming and software design skills, including debugging, performance analysis, and test design.
Strong foundation in architecture of CPU, GPUs or other high performance hardware accelerators.
Knowledge of high-performance computing and distributed programming.
Strong interpersonal skills are required along with the ability to work in a dynamic product-oriented team.

Nice To Haves

CUDA or OpenCL programming experience is desired but not required.
Experience with the following technologies is a huge plus: XLA, TVM, MLIR, LLVM, OpenAI Triton, deep learning models and algorithms, and deep learning framework design.
A history of mentoring junior engineers and interns is a bonus.
Experience working deep learning frameworks such as JAX, PyTorch or TensorFlow.
Extensive experience with CUDA or with GPUs in general.
Experience with open-source compilers such as XLA, LLVM, MLIR or TVM.

Responsibilities

Crafting and implementing compiler optimization techniques for deep learning network graphs.
Designing novel graph partitioning and tensor sharding techniques for distributed training and inference.
Performance tuning and analysis.
Code-generation for NVIDIA GPU backends using open-source compilers such as MLIR, LLVM and OpenAI Triton.
Designing user facing features in JAX and related libraries and other general software engineering work.
Working closely with GPU hardware engineering teams to design AI compiler software features for next-generation GPUs.