Deep Learning Performance Architect jobs at Nvidia in India, Bengaluru

What you will be doing:

• Architect multi-GPU system topologies for scale-up and scale-out configurations, balancing AI throughput, scalability, and resilience.

• Define, modify and evaluate future architectures for high-speed interconnects such as NVLink and Ethernet co-designed with the GPU memory system.

• Collaborate with other teams to architect RDMA-capable hardware and define transport layer optimizations for GPU-based large scale AI workload deployments.

• Use and modify system models, perform simulations and bottleneck analyses to guide design trade-offs.

• Work with GPU ASIC, compiler, library and software stack teams to enable efficient hardware-software co-design across compute, memory, and communication layers.

• Contribute to interposer, package, PCB and switch co-design for novel high-density multi-die, multi-package, multi-node rack-scale systems consisting of hundreds of GPUs.

What we need to see:

• BS/MS/PhD in Electrical Engineering, Computer Engineering, or equivalent area.

• 8 years or more of relevant experience in system design and/or ASIC/SoC architecture for GPU, CPU or networking products.

• Deep understanding of communication interconnect protocols such as NVLink, Ethernet, InfiniBand, CXL and PCIe.

• Experience with RDMA/RoCE or InfiniBand transport offload architectures.

• Proven ability to architect multi-GPU/multi-CPU topologies, with awareness of bandwidth scaling, NUMA, memory models, coherency and resilience.

• Experience with hardware-software interaction, drivers and runtimes, and performance tuning for modern distributed computing systems.

• Strong analytical and system modeling skills (Python, SystemC, or similar).

• Excellent cross-functional collaboration skills with silicon, packaging, board, and software teams.

Ways to stand out from the crowd:

• Background in system design for AI and HPC.

• Experience with NICs or DPU architecture and other transport offload engines.

• Expertise in chiplet interconnect architectures or multi-node fabrics and protocols for distributed computing.

• Hands-on experience with interposer or 2.5D/3D package co-design.

What you'll be doing:

• Modeling and analysis of graphics and / or SOC algorithms and features

• Work in a matrixed environment, across the different modelling teams, to document, design, develop tools to analyze and simulate, validate, and verify models

• Familiarize with the different models (functional and performance) that are used at Nvidia and help with feature implementation as required.

• Develop tests, test plans, and testing infrastructure for newarchitectures/features.

• Mentor younger engineers and help sustain good coding practices.

• Learn about newer modelling techniques and frameworks, evaluate the best solution for our needs and work with your manager to drive the change

• Help develop AI based tools to increase efficiency.

What we need to see:

• Bachelors Degree (or equivalent experience) in a relevant discipline (Computer Science, Electrical Engineering or Computer Engineering)

• 8+ years of relevant work experience or MS with 5+ years of experience or PhD with 2+ years of experience

• Strong programming ability: C++, C along with a good understanding of build systems (CMAKE, make), toolchains (GCC, MSVC) and libraries (STL, BOOST)

• Computer Architecture background with experience in performance modeling with C++ and SystemC preferred

• Familiarity with Docker, Jenkins, Python, Perl

• Excellent communication and interpersonal skills and ability to work in a distributed team environment.

What you will be doing:

• Develop benchmarks, end to end customer applications running at scale, instrumented for performance measurements, tracking, sampling, to measure and optimize performance of important applications and services;

• Construct carefully designed experiments to analyze, study and develop critical insights into performance bottlenecks, dependencies, from an end to end perspective;

• Develop ideas on how to improve the end to end system performance and usability by driving changes in the HW or SW (or both).

• Collaborate with AI researchers, developers, and application service providers to understand internal developer and external customer pain points, requirements, project future needs and share best practice.

• Develop the necessary modeling framework and the TCO (total cost of ownership) analysis to enable efficient exploration and sweep of the architecture and design space

• Develop the methodology needed to drive the engineering analysis to Inform the architecture, design and roadmap of DGX Cloud

What we need to see:

• Expertise in working with large scale parallel and distributed accelerator-based system systems

• Expertise optimizing performance and AI workloads on large scale systems

• Experience with performance modeling and benchmarking at scale

• Strong background in Computer Architecture, Networking, Storage systems, Accelerators

• Familiarity with popular AI frameworks (PyTorch, TensorFlow, JAX, Megatron-LM, Tensort-LLM, VLLM) among others

• Experience with AI/ML models and workloads, in particular LLMs as well as an understanding of DNNs and their use in emerging AI/ML applications and services

• Bachelors/Masters in Engineering or equivalent experience (preferably, Electrical Engineering, Computer Engineering, or Computer Science)

• 10 years experience in the above areas

• Proficiency in Python, C/C++

• Expertise with at least one of public CSP infrastructure (GCP, AWS, Azure, OCI, …);

Ways to stand out from the crowd:

• PhD in the relevant areas

• Very high intellectual curiosity; Confidence to dig in as needed; Not afraid of confronting complexity; Able to pick up new areas quickly;

• Proficiency in CUDA, XLA

• Excellent interpersonal skills

What you'll be doing:

• Performance analysis/ bottleneck analysis of complex, high performance GPUs and System-on-Chips (SoCs).

• Work on hardware models of different levels of extraction, including performance models, RTL test benches and emulators to find performance bottlenecks in the system.

• Work closely with the architecture and design teams to explore architecture trade-offs related to system performance, area, and power consumption.

• Understand key performance usecases for the product. Develop workloads and test suites targeting graphics, machine learning, automotive, video, compute vision applications running on these products.

• Drive methodologies for improving turnaround time, finding representative data-sets and enabling performance analysis early in the product development cycle.

• Develop required infrastructure including performance simulators, testbench components and analysis tools.

What we need to see:

• BE/BTech or MS/MTech, or equivalent experience in relevant area, PhD is a plus.

• 3+ years of relevant experience dealing with system level architecture and performance issues.

• Strong understanding of System-on-Chip (SoC) architecture, graphics pipeline, CPU architecture, memory subsystem architecture and Network-on-Chip (NoC)/Interconnect architecture.

• Solid programming (C/C++) and scripting (Bash/Perl/Python) skills. Exposure to Verilog/System Verilog, SystemC/TLM is a strong plus.

• Strong debugging and analysis (including data and statistical analysis) skills, including use of RTL dumps to debug failures.

• Exposure to performance simulators, cycle accurate/approximate models or emulators for pre-silicon performance analysis is a plus.

• Excellent communication and organization skills.

• Ability to work in a global team environment.

Ways to stand out from the crowd:

• Strong background in System Level Performance aspects for Graphics and High Performance Computing.

• Exposure to GPU application programming interfaces like CUDA, OpenGL, DirectX.

• Expertise in data analysis and visualization.

What you'll be doing:

• Develop AI performance tools for large scale AI systems providing real time insight into applications performance and system bottlenecks.

• Conduct in-depth hardware-software performance studies

• Define performance and efficiency evaluation methodologies

• Automate performance data analysis and visualization to convert profiling data into actionable optimizations

• Support deep learning software engineers and GPU architects in their performance analysis efforts

• Work with various teams at NVIDIA to incorporate and influence the latest technologies for GPU performance analysis

What we need to see:

• Minimum of 8+ years of experience insoftware infrastructure and tools

• BS or higher degree in computer science or similar (or equivalent experience)

• Adept programming skills in multiple languages including C++ and Python

• Solid foundation in operating systems and computer architecture

• Outstanding ability to understand users, prioritize among many contending requests, and build consensus

• Passion for “it just works” automation, eliminating repetitive tasks, and enabling team members

Ways to stand out from the crowd:

• Experience in working with the large scale AI cluster

• Experience with CUDA and GPU computing systems

• Hands-on experience with deep learning frameworks (TensorFlow, PyTorch, JAX/XLA etc.)

• Deep understanding of the software performance analysis and optimization process

What you will be doing:

• Contribute to advancing GPU Streaming Multiprocessor (SM) Architecture, Simulators, compilers and testing infrastructure.

• Understand and model in architecture compiler, features for AI and graphics GPU SM.

• Develop test plans and testing infrastructure for next generation GPU SM.

• Develop tools to validate performance models and verify functional models.

What we need to see:

• A Masters/Bachelor’s degree in CS or Math, with 3+ years of relevant experience in Compilers, Parallel programming, computer architecture or related field.

• Strong programming ability in C, C++, Python.

• Strong understanding of Compiler and data structure is a must.

• Ability to work with teams spanning both hardware and software boundaries.

Ways to stand out from the crowd:

• Experience in GPU compilers and architectural exploration.

• Solid programming skills and CPU core architecture background.

• Understanding of ISA of CPU (GPU ISA will be a big plus)