GPU Kernel Engineer | CUDA & High-Performance ML

Mohamed bin Zayed University of Artificial Intelligence; Sunnyvale (CA); Presencial; USD 150.000 - 450.000; Jornada completa

Ventajas ofrecidas por este puesto de trabajo

Medical, dental, vision benefits

Performance bonus

401K Plan

PTO and holidays

Parental Leave

Employee Assistance Program

Life insurance and disability

Descripción de la vacante

MBZUAI seeks a GPU Kernel Engineer to optimize machine-learning software stacks, focusing on CUDA kernels and distributed training. You’ll profile, benchmark, and build tools to maximize GPU utilization on state-of-the-art platforms, collaborating with researchers and engineers.

The role emphasizes hands-on system-level coding, parallel computing, and involvement in design reviews to drive efficient, scalable deep learning systems.

Formación

• Strong C++ and CUDA kernel development experience for deep-learning workloads.
• Understanding of GPU memory hierarchy and warp/block execution.
• Strong Python skills with PyTorch kernel integration and validation.

Responsabilidades

• Profile, optimize, and guide DL workloads on state-of-the-art hardware.
• Design benchmarks and testing methodologies for performance.
• Build tooling for workload analysis and optimization.
• Triage issues affecting GPU utilization and propose improvements.
• Develop kernels for new model architectures and algorithms.
• Lead design reviews and ensure code quality and documentation.

Conocimientos

C++ / CUDA kernel

GPU memory hierarchy

Python / PyTorch integration

Herramientas

Triton / CUTLASS / PTX

Distributed training systems

BF16 / FP8 precision

Descripción del empleo

MBZUAI seeks a GPU Kernel Engineer to optimize machine-learning software stacks, focusing on CUDA kernels and distributed training. You’ll profile, benchmark, and build tools to maximize GPU utilization on state-of-the-art platforms, collaborating with researchers and engineers.

The role emphasizes hands-on system-level coding, parallel computing, and involvement in design reviews to drive efficient, scalable deep learning systems.