Senior Deep Learning Frameworks CUDA Software Engineer

NVIDIA•Austin, TX

5d•$184,000 - $356,500

About The Position

NVIDIA is seeking a motivated Deep Learning engineer to bring advanced CUDA features and distributed runtime technologies into AI stacks, including PyTorch, TRT-LLM, vLLM, SGLang, and JAX. The role works with the team responsible for core CUDA features and runtimes that scale Deep Learning and HPC applications. Customers have multi-GPU needs ranging from training at scales of up to 100K GPUs to inference at microsecond latency. The work helps improve the productivity and performance of AI applications and makes CUDA features available through AI toolkits. The engineer will integrate CUDA features and runtime abstractions into AI frameworks, analyze workloads and frameworks, and collaborate with teams developing AI models. The role also drives improvements to the AI compiler-runtime interface, designs fault-tolerant and elastic solutions, influences the CUDA roadmap, and develops tools and runtime systems to profile and accelerate new deep learning paradigms. Clean, maintainable code should support transitions from exploratory prototypes to open-source releases, upstream integrations, internal tools, or commercial products. The role involves close collaboration with researchers and hardware and software specialists across teams and time zones. The position requires a relevant degree or equivalent experience, at least eight years of relevant industry or equivalent academic experience after degree completion, and development experience with deep learning frameworks or inference engines. The expected base salary is $184,000–$287,500 for Level 4 or $224,000–$356,500 for Level 5. The employee is also eligible for equity and benefits.

Requirements

Bachelor’s, master’s, or Ph.D. degree in Computer Science, Computer Engineering, Electrical Engineering, or a related field, or equivalent experience.
At least 8 years of relevant industry experience or equivalent academic experience after completing a degree.
Development experience with deep learning frameworks such as PyTorch and JAX, and inference engines such as TRT-LLM, vLLM, or SGLang.
Rapid prototyping and development experience with Python, C++, CUDA, or related domain-specific languages.
A solid grasp of AI models, parallelism, and/or compiler technologies such as torch.compile.
Experience conducting performance benchmarking on AI clusters.
Familiarity with at least one performance profiler toolchain, such as PyTorch profiler or NVIDIA Nsight Systems.
Understanding of HPC/AI communication concepts.
Good understanding of computer system architecture, hardware-software interactions, and operating system principles.
Adaptability and willingness to learn new frameworks and tools.
Ability to work and communicate effectively across teams and time zones.

Nice To Haves

Deep expertise in performance internals and execution graphs of major deep learning autograd, training, and inference frameworks, such as PyTorch, JAX, TensorRT, vLLM, SGLang, Nemo, Megatron, or MaxText.
Hands-on experience with CUDA, communication libraries such as NCCL, MPI, or UCX, and distributed machine learning techniques such as pipeline or tensor parallelism.
Expertise in one or more of training, distributed inference, mixture of experts (MoE), reinforcement learning, or kernel authoring with CUDA, Triton, cuTe, or similar technologies.
Background in deep learning compilers at the graph level and for code generation, such as Triton, XLA, or torch.compile.
Experience programming for compute and communication overlap in distributed runtimes.

Responsibilities

Integrate new CUDA features and runtime abstractions into AI frameworks, from proof of concept through performance analysis and production.
Analyze AI workloads and frameworks to identify requirements and opportunities for innovation in lower layers of the stack; work hands-on with teams developing AI models.
Own and drive improvements to the AI compiler-runtime interface to build high-performance multi-GPU, multi-node solutions.
Design fault-tolerant and elastic solutions for large-scale or dynamic AI workloads.
Influence the roadmap of core CUDA to support next-generation deep learning frameworks.
Collaborate across a dynamic team spanning multiple time zones.
Work with AI researchers, hardware and software architects, kernel and compiler authors, and CUDA driver experts to co-design performant and programmable systems and frameworks.
Develop exploratory tools and runtime systems to profile and accelerate emerging deep learning paradigms.
Write clean, effective, maintainable code and help exploratory prototypes transition into open-source releases, upstream framework integrations, internal tools, or commercial products.