Deep Learning Performance Architect

As an NVIDIAN, you will be part of a company that consistently reinvents itself and is at the forefront of innovation in GPU technology. Your role will involve benchmarking and analyzing AI workloads in both single and multi-node configurations. You will contribute to high-level simulator and debugger development using C++ and Python. Evaluating PPA (performance, power, area) for hardware features and system-level architectural trade-offs will be a key aspect of your responsibilities. Collaboration with wider architecture teams, architecture, and product management will be essential to provide valuable trade-off analysis throughout the project. Staying updated with emerging trends and research in deep learning will also be part of your role. Key Responsibilities: - Benchmark and analyze AI workloads in single and multi-node configurations. - Develop high-level simulator and debugger in C++/Python. - Evaluate PPA for hardware features and architectural trade-offs. - Collaborate with architecture teams and product management for trade-off analysis. - Stay informed about emerging trends and research in deep learning. Qualifications Required: - MS or PhD in a relevant discipline (CS, EE, Math). - 2+ years of experience in parallel computing architectures, interconnect fabrics, and deep learning applications. - Strong programming skills in C, C++, and Python. - Proficiency in architecture analysis and performance modeling. - Curious mindset with excellent problem-solving skills. If you want to stand out from the crowd, having an understanding of modern transformer-based model architectures, experience with benchmarking methodologies, workload profiling, and the ability to simplify complex technical concepts for non-technical audiences will be beneficial. Join NVIDIA's Deep Learning Architecture team and be a part of driving success in the rapidly growing field of AI computing. Your contributions will help in building real-time, cost-effective computing platforms that are shaping the future of technology. (JR1990438),