Summer 2027 PhD AI Machine Learning Engineer Intern

Advanced Micro Devices, Inc•San Jose, CA

9h•Onsite

About The Position

AMD is seeking a PhD intern for its AI Model and Application Engineering team. The intern will work on projects based on candidate experience and company needs, optimizing end-to-end AI training and inference pipelines on AMD Instinct GPUs using the ROCm software stack. Work may involve model and kernel design tradeoffs, numerical techniques for balancing accuracy and performance, reproducible benchmarking, profiling, and performance improvements. The intern will also help create experimental reports and code artifacts that may be contributed to internal repositories or open-source projects. The role is onsite in San Jose, California, for 40 hours per week during the Summer 2027 internship term. Semester-school dates are May 24, 2027–August 13, 2027; quarter-school dates are June 21, 2027–September 10, 2027.

Requirements

Currently pursuing a PhD in Computer Science, Computer Engineering, Electrical Engineering, Applied Mathematics, or a related technical field.
Demonstrated research or advanced project experience in AI systems, machine learning infrastructure, high-performance computing, computer architecture, or large-scale deep learning.
Hands-on experience training, optimizing, or serving large-scale AI models with PyTorch or JAX, including distributed training technologies such as DeepSpeed, FSDP, or Megatron-LM.
Strong experience in GPU performance engineering, kernel optimization, and accelerator programming using CUDA, HIP, Triton, or related technologies.
Solid understanding of GPU architecture concepts, including memory hierarchies, tensor cores/MFMA, operator fusion, and scheduling techniques.
Experience with profiling, benchmarking, and performance modeling methodologies, including roofline analysis, to evaluate utilization, identify bottlenecks, and improve AI training and inference performance.
Expertise in AI model efficiency techniques such as PTQ/QAT quantization, low-precision formats (FP8, FP4, INT8, or Ternary), mixed-precision training, speculative decoding, and energy-efficient AI training and inference.
Ability to build rigorous and reproducible experiments that produce impactful optimization outcomes.

Responsibilities

Optimize end-to-end training and inference pipelines on AMD Instinct GPUs using ROCm and related software.
Explore model and kernel-level design tradeoffs, including attention variants, operator fusion, and scheduling, and assess their impact through benchmarking.
Develop and evaluate numerical techniques such as quantization-aware and mixed-precision training.
Build reproducible benchmarks and experiment harnesses, profile applications, identify bottlenecks, and deliver measurable performance improvements.
Produce clear experimental reports and code artifacts that may be contributed to internal repositories or open-source projects.