GPU Software Engineer

OHO Group (Lithuania); San Francisco (CA); On-site; USD 150,000 - 190,000; Full time

Job summary

OHO Group in San Francisco is seeking a GPU Software Engineer specializing in kernels and runtime to advance our high-performance AI compute platform.

You will optimize GPU kernels, bring models to new hardware, and collaborate with compiler and hardware teams to push performance and reliability across the stack.

Join a fast-growing startup with strong equity and a competitive benefits package.

Qualifications

• Proven track record in performance-sensitive GPU software.
• Strong background in C++ and GPU programming.
• Experience with AI workloads or accelerator systems preferred.

Responsibilities

• Develop and optimize GPU kernels for AI operations, including attention and matrix multiplication.
• Bring models onto new hardware and resolve performance or correctness issues.
• Improve memory access, parallel execution, scheduling and hardware utilization.
• Contribute to runtime software and integration with AI frameworks.
• Profile workloads, investigate bottlenecks and validate improvements through reproducible benchmarks.
• Work directly with compiler and hardware engineers to solve problems across the stack.

Skills

C++

CUDA

GPU programming

Performance optimization

Parallel programming

Tools

HIP

Triton

Profiling tools

Job description

GPU Software Engineer – Kernels & Runtime

A well-funded AI accelerator startup is building a new compute platform for demanding AI workloads.

The team is looking for engineers who enjoy getting models running, understanding what limits their performance and making the underlying software faster. This role will sit within the architecture team, contributing to key decisions and requires a high level of ownership.

What you’ll work on

• Develop and optimize GPU kernels for AI operations, including attention and matrix multiplication.
• Bring models onto new hardware and resolve performance or correctness issues.
• Improve memory access, parallel execution, scheduling and hardware utilization.
• Contribute to runtime software and integration with AI frameworks.
• Profile workloads, investigate bottlenecks and validate improvements through reproducible benchmarks.
• Work directly with compiler and hardware engineers to solve problems across the stack.

What we’re looking for

• Strong C++ and practical GPU programming experience using CUDA, HIP, Triton or comparable tools.
• Understanding of GPU execution, memory hierarchies, synchronization and parallel programming.
• Evidence of building, debugging or optimizing performance-sensitive software.
• Ideally, experience with data‑centre GPUs, AI workloads or accelerator systems.

Successful applicants will enjoy strong salary, benefits and meaningful equity.