Remote Kernel Engineer for AI Hardware & ML Kernels

United States Digital Space LLC; United States; Remote; USD 140,000 - 200,000; Full time

Job summary

Cerebras is seeking a Kernel Engineer to design, optimize, and scale deep learning kernels for its massively parallel WSE architecture. You will contribute to a library of parallel algorithms and collaborate with hardware teams to maximize compute utilization and training efficiency.

The role focuses on developing low-level kernels, optimizing assembly and a custom CSL language, and applying performance modeling to guide design decisions.

Qualifications

• Bachelor’s, Master’s, or PhD (or foreign equivalents) in Computer Science, Computer Engineering, Mathematics, or related fields.
• Strong understanding of hardware architecture concepts and ability to learn new hardware details.
• Proficient in C++ and Python programming languages.
• Experience with library and API development best practices.
• Excellent debugging skills and ability to diagnose complex software stacks.

Responsibilities

• Develop design specifications for ML and linear algebra kernels for the Cerebras WSE System.
• Build and debug kernel libraries using low-level assembly and a C-like DSL to target Cerebras hardware.
• Develop high-performance kernel routines in assembly and CSL, optimizing for hardware performance.
• Use models and analysis to measure performance and influence design decisions.
• Develop and integrate unit and system tests to verify kernel functionality and performance.
• Study ML trends and evolve kernel library architecture to address challenges of state-of-the-art neural networks.
• Collaborate with chip and system architects to optimize instruction sets, microarchitecture, and I/O for future systems.

Skills

C++

Python

Hardware concepts

Debugging

API development

Education

BS/MS/PhD in CS/CE/Math

Job description

Cerebras is seeking a Kernel Engineer to design, optimize, and scale deep learning kernels for its massively parallel WSE architecture. You will contribute to a library of parallel algorithms and collaborate with hardware teams to maximize compute utilization and training efficiency.

The role focuses on developing low-level kernels, optimizing assembly and a custom CSL language, and applying performance modeling to guide design decisions.