AI Performance Modeling Engineer

Advanced Micro Devices, Inc.; Bengaluru Urban; On-site; INR 2,800,000 - 4,600,000; Full time

Job summary

Advanced Micro Devices, Inc. (AMD) is seeking an experienced AI performance modeling engineer to project workloads on GPU architectures for next-generation AI systems. You will build analytical and cycle-accurate models, guiding architectural decisions and optimizations across the stack.

The role requires deep GPU knowledge, strong C++ and Python skills, and the ability to translate simulation results into actionable engineering insights for pre-silicon and post-silicon workloads.

Qualifications

• Experience with LLM and Transformer performance on GPUs.
• Ability to build analytical and cycle-accurate models.
• Strong C++ and Python development skills.
• Experience with GPU cache/memory bandwidth modeling.

Responsibilities

• Project AI workload performance projections for GPU architecture.
• Develop analytical and cycle-accurate models and simulators.
• Identify bottlenecks in compute engines, memory, and interconnects.
• Collaborate with compiler, runtime, and framework teams.
• Validate projections with pre/post-silicon data.

Skills

AI workload modeling

Analytical performance models

Cycle-accurate simulators

Modern C++

Python

GPU architecture depth

Data/Model parallelism

Tools

Performance simulators

Linux tooling

Job description

WHAT YOU DO AT AMD CHANGES EVERYTHING

At AMD, our mission is to build great products that accelerate next-generation computing experiences-from AI and data centers, to PCs, gaming and embedded systems. Grounded in a culture of innovation and collaboration, we believe real progress comes from bold ideas, human ingenuity and a shared passion to create something extraordinary. When you join AMD, you'll discover the real differentiator is our culture. We push the limits of innovation to solve the world's most important challenges-striving for execution excellence, while being direct, humble, collaborative, and inclusive of diverse perspectives. Join us as we shape the future of AI and beyond. Together, we advance your career.

The Role

We're building GPUs to run the most demanding AI in the world - Transformers, LLMs, and large vision models - and we need an experienced hands-on engineer who can tell us how fast those workloads will run before the silicon exists . This is an AI performance projection/modeling role: your job is to model AI workloads on our GPU architecture, predict their performance with confidence, and use those projections to shape the hardware we build.

You'll work across the full modeling spectrum - analytical models for fast architectural what-ifs, and architectural and cycle-accurate simulators for precision - choosing the right level of fidelity for the question at hand. Because these workloads live on the GPU, you'll need real GPU architecture depth to do it.

If you think in rooflines, arithmetic intensity, tensor/pipeline/data parallelism, and KV-cache bandwidth budgets - and you want your models to change the microarchitecture, not just describe it - this is the role.

What You'll Do

• Project AI workload performance (your core mission)
• Build and use models - analytical, architectural, and cycle-accurate - to project the performance of generative AI and LLM workloads (attention, GEMMs, collectives, training and inference, vision models etc) on our GPU architecture.
• Choose the right fidelity for the question: analytical bounds and rooflines for rapid trade-off exploration, architectural and cycle-accurate simulation for precise answers.
• Find where performance will be won or lost - compute-engine occupancy, L1/L2 hierarchy, memory bandwidth, SoC fabric - and quantify the impact of each.
• Characterize workloads & shape the hardware
• Characterize AI workloads end to end - trace and profile execution, extract structural footprints, and translate them into microarchitectural bottlenecks and performance bounding boxes.
• Run architecture experiments to evaluate GPU configurations, scaling limits, and trade-offs against real AI workloads, backing every recommendation with sensitivity analysis and data.
• Extend and maintain the cycle-accurate / cycle-approximate simulators for the GPU subsystems AI workloads depend on - compute engines, cache hierarchies, memory subsystems, interconnects.
• Prove it on silicon (post-silicon)
• Run AI workloads on early silicon, capture telemetry, and correlate results back to your pre-silicon models to sharpen fidelity.
• Root-cause AI performance mismatches on physical silicon using low-level performance counters, register dumps, and Linux tracing.
• Collaborate across the stack
• Partner with compiler, runtime, and framework teams to turn model insights into optimizations, and to ground your projections in how workloads actually execute.
• Act as the bridge between architecture/RTL and software, translating AI workload requirements into concrete architectural constraints.

What You’ll Bring

• AI workload expertise: Deep, hands‑on understanding of LLM and Transformer performance - scaling techniques, parallel training/inference strategies (tensor, pipeline, data parallelism), kernel execution graphs - and how they map to hardware.
• Performance modeling breadth: Hands‑on experience with analytical performance models and execution-driven / architectural / cycle‑accurate simulators, and sound judgment about which to reach for. Track record of turning models into decisions.
• GPU architecture depth: Strong command of GPU execution pipelines, SIMD/SIMT models, cache hierarchy, memory technologies, and high‑bandwidth interconnects (PCIe, custom fabrics).
• Engineering craft: Expert modern C++ and production‑grade Python.
• Execution: Excellent structured problem-solving; able to turn noisy simulation data into crisp, actionable findings for engineering stakeholders.
• Nice to have: Experience with NPU / AI-accelerator architecture, or modeling AI workloads on dedicated inference/training accelerators.

Benefits offered are described: AMD benefits at a glance .

AMD and its subsidiaries are equal opportunity, inclusive employers and will consider all applicants without regard to age, ancestry, color, marital status, medical condition,