Senior Software Engineer, CUDA Deep Learning Systems

NVIDIA•Austin, CA

About The Position

We are looking for an experienced and highly motivated software professional to work on pioneering initiatives and projects at the intersection of CUDA and Deep Learning Systems. As the complexity and scale of artificial intelligence continue to grow, the intersection of advanced deep learning architectures, massive-scale distributed computing, and low-level hardware optimization has never been more critical. Our team is dedicated to exploring and prototyping next-generation ideas that bridge the gap between deep learning algorithms and CUDA, pushing the boundaries of what is possible on modern accelerator architectures. Join our dynamic, research-oriented team to help unlock maximum hardware performance for emerging AI workloads. You will be a crucial member of a highly technical group exploring uncharted territories in model optimization, custom kernel development, and cluster-scale AI systems design. If you are passionate about the fundamentals of deep learning and thrive on squeezing every ounce of performance out of advanced computing systems from a single GPU to supercomputer clusters, we want you on our team!

Requirements

BS, MS, or PhD degree in Computer Science, Computer Engineering, Electrical Engineering, or related field (or equivalent experience).
8+ years of relevant industry experience or equivalent academic experience after degree achievement.
Strong proficiency in C++ and Python programming.
Solid background in the fundamentals of Deep Learning with a focus on transformers.
Strong understanding of distributed computing principles, multi-node scaling, and the unique performance challenges of cluster-scale execution.
Proven experience in systems programming, computer architecture, and low-level systems performance optimization.
Familiarity with deep learning accelerator architectures such as the GPU and hands-on experience with CUDA programming and kernel optimization.
A strong analytical approach with experience using profiling tools to deeply understand software performance on hardware.
Experience profiling and optimizing innovative vision models, generative AI architectures, or diffusion models.
Background in deep learning compilers, both graph-level and codegen (e.g., Triton, XLA, torch compile)

Nice To Haves

Deep expertise in the performance internals and execution graphs of major deep learning autograd, training and inference frameworks (e.g., PyTorch, JAX, TensorRT, vLLM, sgLang, Nemo, Megatron, MaxText, etc.).
Hands-on experience with CUDA, communication libraries (e.g., NCCL, MPI, UCX) and distributed machine learning techniques (e.g., pipeline parallelism, tensor parallelism).
Knowledge of numerical methods, low-precision arithmetic (e.g., NVFP4, MXFP4, FP8, INT8), and their implications on deep learning model accuracy and performance.
Familiarity with systems requirements for Reinforcement Learning (RL) or highly parallel simulation environments and/or research background in machine learning systems or adjacent fields.
Experience with machine learning, especially agentic systems, applied to systems problems.

Responsibilities

Explore, research, and prototype novel systems optimizations for advanced deep learning models at the intersection of high-level DL frameworks and low-level CUDA through modeling, simulation, and silicon prototyping.
Architect and optimize distributed computing systems that scale seamlessly from a single node to massive, cluster-scale supercomputing environments.
Design, implement, and optimize custom high-performance CUDA kernels tailored to emerging neural network architectures and workloads.
Analyze complex hardware-software interactions to identify and resolve performance bottlenecks in both training and inference pipelines.
Collaborate closely with AI researchers, HW and SW architects, kernel and compiler authors and CUDA driver experts to co-design systems and algorithms that improve accelerator compute utilization, memory bandwidth, cross-node network communication efficiency and programmability.
Develop exploratory tools and runtime systems to profile and accelerate new paradigms in deep learning.
Write clean, effective, and maintainable code, ensuring exploratory prototypes can smoothly transition into open-source releases, upstream framework integrations, internal tools, or closed-source commercial products.