Devtech Compute Engineer

at NVIDIA · AI Frontier · Beijing, China +1

NVIDIA is seeking a Devtech Compute Engineer to develop performance-critical code for deep learning applications, focusing on accelerating model training and inference on GPUs, particularly for recommender systems. The role involves optimizing CUDA kernels, integrating solutions into open-source libraries, and collaborating with hardware teams to define future solutions across various domains like LLM, Recsys, Robotics, and Assisted Driving.

What you'd actually do

In your role as Devtech Compute Engineer or CUDA Performance Engineer you will be primarily for the development of performance critical code of our deep learning applications with the goal of establishing world class performance for our customers.
This includes investigating the current performance and exploring optimization opportunities together with the global developers.
Important part of the work is that once optimal performance has been demonstrated that these solutions are integrated into our open source software libraries like ACCV-Lab, Recsys-Example .
With the knowledge to the requirements from customers and performance bottleneck, you will also work with our GPU, CPU, Network team to define the next generation hardware and software solutions.

Skills

Required

2+ years of experience of c++ code development in collaborative software development projects
Skilled at writing CUDA kernels and optimizing code
Basic knowledge of ML algorithms and deep learning
Basic knowledge and understanding of mathematical topics including linear algebra, calculus and statistics
Experience with algorithms and optimization
Python and jupyter notebook for analysis, algorithm exploration and processing
High standard for code quality and rigorous testing practices
Conversational level English proficiency
Some experience with Linux, openMP and MPI

Nice to have

Experience in c++ HPC code development / PhD in related fields
Able to perform in-depth performance analysis, can demonstrate to model the performance with mathematical and statistical considerations
Linear algebra, calculus and statistics as second nature and this is reflected in your background of mathematics, physics, applied science or HPC related field
Demonstrate the ability to write CUDA kernels with the purpose of utilizing the hardware to its full potential.
Write unit tests and validate the correctness of the optimizations as well as strive for and propose optimal solutions and ambitious goals, convince and help others to do the same

What the JD emphasized

performance critical code
CUDA kernels
deep learning applications
GPU-based recommender systems
LLM
Recsys
Robotic
Assisted Driving

Other signals

performance critical code
deep learning applications
GPU-based recommender systems
LLM
Recsys
Robotic
Assisted Driving

Read full job description

We’re working on the next generation of recommendation tools and pushing the boundaries of accelerating model training and inference on GPU. You’ll join a team of ML, HPC and Software Engineers and Applied Researcher developing a framework designed to make the productization of GPU-based recommender systems as simple and fast as possible.

What you'll be doing:

In your role as Devtech Compute Engineer or CUDA Performance Engineer you will be primarily for the development of performance critical code of our deep learning applications with the goal of establishing world class performance for our customers.
This includes investigating the current performance and exploring optimization opportunities together with the global developers.
Important part of the work is that once optimal performance has been demonstrated that these solutions are integrated into our open source software libraries like ACCV-Lab, Recsys-Example .
With the knowledge to the requirements from customers and performance bottleneck, you will also work with our GPU, CPU, Network team to define the next generation hardware and software solutions.
Our coverage is wide including: LLM, Recsys, Robotic, Assisted Driving.

What we need to see:

2+ years of experience of c++ code development in collaborative software development projects
Skilled at writing CUDA kernels and optimizing code
Basic knowledge of ML algorithms and deep learning
Basic knowledge and understanding of mathematical topics including linear algebra, calculus and statistics
Experience with algorithms and optimization
Python and jupyter notebook for analysis, algorithm exploration and processing
High standard for code quality and rigorous testing practices
Conversational level English proficiency
Some experience with Linux, openMP and MPI

Way to stand out from the crowd:

Experience in c++ HPC code development / PhD in related fields
Able to perform in-depth performance analysis, can demonstrate to model the performance with mathematical and statistical considerations
Linear algebra, calculus and statistics as second nature and this is reflected in your background of mathematics, physics, applied science or HPC related field
Demonstrate the ability to write CUDA kernels with the purpose of utilizing the hardware to its full potential.
Write unit tests and validate the correctness of the optimizations as well as strive for and propose optimal solutions and ambitious goals, convince and help others to do the same

#deeplearning