GPU Kernel Engineer for High-Performance ML

MakerMaker.AI; San Francisco (CA); On-site; USD 180,000 - 230,000; Full time

Benefits offered by this job

Open-source collaboration

Hybrid/On-site flexibility

Competitive compensation

Job summary

MakerMaker.AI in San Francisco is seeking a kernel engineer to write and optimize GPU kernels (CUDA, ROCm, Triton) for training and inference workloads. You will work on performance counters, memory bandwidth, and warp-level scheduling to accelerate our models.

You will profile workloads, collaborate with researchers, and integrate optimized kernels into training and serving stacks while maintaining kernel quality as hardware evolves.

Qualifications

• 4+ years writing performant GPU kernels (CUDA, ROCm, Triton, or production-grade equivalent).
• Hardware-level fluency: memory hierarchy, occupancy, register pressure, tensor cores, warp scheduling.
• Profiling fluency with Nsight/ncu or comparable tools and measuring before changing.
• Track record of shipping kernel-level optimizations that moved a measurable metric in a real system.
• Strong systems expertise: kernels inside larger frameworks and end-to-end performance impacts.

Responsibilities

• Write and optimize GPU kernels for training and inference workloads (CUDA, ROCm, Triton).
• Profile workloads with hardware counters and translate findings into kernel optimizations.
• Co-design kernels with research teams when algorithms change.
• Integrate optimized kernels into training and serving stacks; benchmark and verify end-to-end wins.
• Maintain kernel quality over time as hardware/frameworks evolve.
• Spread kernel-level fluency across the team.

Skills

GPU kernel development

Performance profiling

C++/Python

System-level optimization

Warp scheduling

Tools

Nsight

ncu

CUDA

ROCm

Triton

Job description

MakerMaker.AI in San Francisco is seeking a kernel engineer to write and optimize GPU kernels (CUDA, ROCm, Triton) for training and inference workloads. You will work on performance counters, memory bandwidth, and warp-level scheduling to accelerate our models.

You will profile workloads, collaborate with researchers, and integrate optimized kernels into training and serving stacks while maintaining kernel quality as hardware evolves.