Principal / Senior Systems Performance Engineer

Our vision is to transform how the world uses information to enrich life for all.

Micron Technology is a world leader in innovating memory and storage solutions that accelerate the transformation of information into intelligence, inspiring the world to learn, communicate and advance faster than ever.

Micron Data Center Workload Engineering in Austin, Texas, is seeking a senior engineer to join our team. We build, performance tune, and test data center solutions using innovative DRAM, SSD, and emerging memory hardware! Understanding key data center workloads is a Micron imperative, in order to improve current products in the deep-memory hierarchy (HBM, CXL, DDR) and bring about a total value proposition to customers based on efficiently applying several Micron products in concert!

Particularly, with the proliferation of generative AI, there is the urgent need to better understand how large language model training is impacted by HBM and GPU characteristics. To this end, the successful candidate will primarily supply to the HBM program by analyzing how AI/ML, HPC, data analytics and other compute-intensive workloads perform on the latest MU-HBM3e / NVIDIA H100 GPU / AMD GPU, conduct competitive analysis of SS / SK HBM3 products, showcase the benefits that workloads see with MU-HBM3e’s capacity / bandwidth / thermals, give to marketing collateral, and extract AI/ML and HPC workload traces to help optimize future HBM designs.

Job Responsibilities:

These include but are not limited to the following:

• Analysis and characterization of data center workloads in several areas such as AI/ML, HPC, data analytics.

• Profiling AI training / inference models in generative AI, computer vision recommendation and NLP on CPU/GPU systems using HBM/DDR.

• Performance benchmarking of HBM using both microbenchmarks and the aforementioned data center applications and benchmarks.

• Particular focus on generative AI, large language models and how they are impacted by HBM and GPU features.

• Overlaying deep learning models on multi GPU-based system architectures to understand their interplay

• Understanding how HBM capacity, bandwidth and power/thermals impact DL models

• Competitive analysis of HBM power / thermals under workload behavior

• Detailed telemetry analysis of workloads on the HBM / GPU

• Understand how ML models in LLM, vision, recommendation and NLP are sensitive to additional HBM bandwidth and capacity

• Study the impact of model quantization and GPU / HBM resource usage / needs

• Profile DL model training / inference in a GPU + CPU-offload scenario, and in a cluster of GPU setting

• Build workload memory access traces from AI models and HPC applications

• Study system balance ratios for DRAM/HBM/NVMe in terms of capacity and bandwidth, memory expansion analysis vis-à-vis DRAM/HBM/NVMe, and study the interplay between these products and understand TCO

• Study memory/core, byte/FLOP and memory bandwidth/core/FLOP requirements for a variety of workloads to influence future products

• Study data movement between CPU, GPU and the associated memory subsystems (DDR, HBM) in heterogeneous system architectures via connectivity such as PCIe/NVLINK/Infinity Fabric to understand the bottlenecks in data movement for different workloads (particularly AI)

• Understand how additional DDR memory footprint or CXL memory footprint can help with AI training / inference occurring on the GPU-CPU complex

• Develop an automated testing framework through scripting

• Customer engagements to present findings

Preferred Qualifications:

• Strong AI/ML training/inference models, PyTorch/TensorFlow/DeepSpeed background

• Strong computer systems foundations

• Strong foundation in GPU and CPU processor architecture

• Familiarity with synthetic memory bandwidth testing applications

• Familiarity with and knowledge of server system memory (DRAM) and server system architecture

• Experience with performance analysis

• Understanding of memory and storage hierarchy including HBM

• Knowledge of AI, ML/DL frameworks and running them in CPU-GPU heterogeneous architectures

• Strong software development skills using leading scripting, programming languages and technologies (Python, CUDA, RoCm, C, C++)

• Strong systems software development skills by way of developing solutions for memory, storage, CPU and bigdata subsystems

• Familiarity with PCIe and NVLINK connectivity

• Knowledge of bigdata and data-intensive analysis tools (e.g., Spark)

• Be abreast with the state of the art in deep lear