AI Performance Engineer – GPU & ROCm

Sapphire Stream Technology; Toronto; On-site; CAD 90,000 - 140,000; Full time

Job summary

Sapphire Stream Technology in Toronto, Ontario, is seeking a GPU AI Performance Engineer to optimize and deploy AI workloads using ROCm. You will focus on increasing training speed, reducing memory usage, and improving multi-GPU scalability across Linux and cloud environments.

You will work with GPU software, AI frameworks like PyTorch, TensorFlow and ONNX Runtime, profile with ROCm tools, and collaborate with hardware, compiler, and infrastructure teams to deliver robust performance

Qualifications

• Bachelor’s or master’s degree in CS, CE, EE, or related field.
• Proficient in C++ and Python.
• Experience with Linux and GPU computing.
• Hands-on ROCm, HIP, CUDA, or similar GPU technologies.
• Understanding of GPU architecture and memory management.
• Experience with PyTorch, TensorFlow, ONNX Runtime, or similar AI frameworks.
• Experience profiling and optimizing GPU applications.
• Knowledge of AI model training, inference, and performance optimization.

Responsibilities

• Improve training speed, inference latency, memory usage, and multi-GPU scalability.
• Work with large language models, vision models, multimodal AI, and generative AI.
• Develop and troubleshoot applications using ROCm and HIP.
• Work with PyTorch, TensorFlow, ONNX Runtime, or MIGraphX (and related frameworks).
• Profile GPU applications and identify performance bottlenecks.
• Apply memory optimization, mixed precision, and kernel tuning techniques.
• Support AI deployment on Linux, cloud, edge, and containerized platforms.
• Develop performance benchmarks and automated validation tests.
• Collaborate with hardware, compiler, runtime, framework, and infrastructure teams.

Skills

C++

Python

Linux

ROCm

CUDA

GPU architecture

PyTorch

TensorFlow

ONNX Runtime

Profiling

Performance optimization

Education

Bachelor's or Master's degree in CS/CE/EE

Tools

ROCm

HIP

CUDA

PyTorch

TensorFlow

ONNX Runtime

Job description

We are looking for a GPU AI Performance Engineer to optimize and deploy AI and machine-learning workloads using ROCm.

You will improve the speed, memory efficiency, and scalability of AI training and inference workloads. You will also work with GPU software, AI frameworks, profiling tools, and cloud or containerized environments.

Key Responsibilities

• Improve training speed, inference latency, memory usage, and multi-GPU scalability.
• Work with large language models, vision models, multimodal AI, and generative AI.
• Develop and troubleshoot applications using ROCm and HIP.
• Work with PyTorch, TensorFlow, ONNX Runtime, vLLM, SGLang, or MIGraphX.
• Profile GPU applications and identify performance bottlenecks.
• Apply mixed precision, quantization, kernel tuning, operator fusion, and memory optimization.
• Support AI deployment on Linux, cloud, edge, and containerized platforms.
• Develop performance benchmarks and automated validation tests.
• Collaborate with hardware, compiler, runtime, framework, and infrastructure teams.

Required Qualifications

• Bachelor’s or master’s degree in Computer Science, Computer Engineering, Electrical Engineering, or a related field.
• Strong programming skills in C++ and Python.
• Experience with Linux and GPU computing.
• Hands‑on experience with ROCm, HIP, CUDA, or similar GPU technologies.
• Understanding of GPU architecture, parallel programming, and memory management.
• Experience with PyTorch, TensorFlow, ONNX Runtime, or similar AI frameworks.
• Experience profiling and optimizing GPU applications.
• Knowledge of AI model training, inference, and performance optimization.

Preferred Qualifications

• Experience optimizing large language models or generative AI applications.
• Experience with vLLM, SGLang, or distributed inference.
• Familiarity with LLVM, MLIR, or compiler technologies.
• Experience migrating workloads between CUDA and HIP.
• Knowledge of Kubernetes, containers, and distributed AI systems.
• Contributions to ROCm or other open-source GPU projects.