Trainium Kernel Performance Specialist

Obsidian; San Francisco (CA); Remote; USD 120,000 - 180,000; Full time

Job summary

Obsidian is seeking an evaluator to assess the quality and hardware-appropriateness of Neuron Kernel Interface (NKI) development tasks used to train and evaluate a frontier AI lab's models. You will review CUDA→NKI migration fidelity, Trainium-specific performance details, and cross-platform numerical-correctness standards, providing rubric-based feedback.

Ideal candidates bring hands-on NKI kernel experience on Trainium/Inferentia2, familiarity with Neuron SDKs, and demonstrated ability to

Qualifications

• 2+ years hands-on kernel development with NKI on Trainium/Inferentia2
• Deep understanding of NKI patterns: tile-based computation, SBUF/PSUM/HBM memory hierarchy, partition-dimension constraints, and DMA orchestration
• Experience assessing CUDA→NKI migration fidelity
• Familiarity with Trainium performance profiling: NeuronCore throughput and memory bandwidth bottlenecks
• Experience defining or evaluating cross-platform numerical-correctness standards (GPU vs Trainium accumulation order, rounding behavior, mixed-precision semantics)

Responsibilities

• Provide rubric-based written feedback on NKI tasks used to train and evaluate frontier AI models
• Assess hardware-appropriateness of NKI kernel development
• Evaluate CUDA→NKI migration fidelity and Trainium-specific performance
• Identify cross-platform numerical-correctness issues and suggest improvements

Skills

NKI kernel

CUDA migration

Trainium profiling

Memory hierarchy

DMA orchestration

Tools

AWS Neuron SDK

Neuron Compiler

Triton kernel

Profiling tools

Job description

Obsidian is seeking an evaluator to assess the quality and hardware-appropriateness of Neuron Kernel Interface (NKI) development tasks used to train and evaluate a frontier AI lab's models. You will review CUDA→NKI migration fidelity, Trainium-specific performance details, and cross-platform numerical-correctness standards, providing rubric-based feedback.

Ideal candidates bring hands-on NKI kernel experience on Trainium/Inferentia2, familiarity with Neuron SDKs, and demonstrated ability to