ML Inference Performance Visibility Engineer

Etched; San Jose (CA); On-site; USD 150,000 - 210,000; Full time

Benefits offered by this job

Housing subsidy

Relocation support

Medical benefits

Daily lunch

Job summary

Etched in San Jose is seeking a highly skilled engineer to design and develop a performance analysis tool for our ML accelerator hardware. You will create data collection, tracing, and analysis capabilities to understand workload behavior and unlock the full potential of our hardware.

Join a cross-disciplinary team collaborating with hardware, firmware, drivers, compilers, and ML applications to deliver insights and optimizations for cutting-edge ML workloads.

Qualifications

• Strong proficiency in C++ or Rust
• Proficiency in Python is a plus
• Deep understanding of computer architecture, memory hierarchies, PCIe interconnects
• Proven experience in low-level performance analysis and bottleneck identification
• Experience with performance analysis tools like Nsight, VTune, perf

Responsibilities

• Lead the design and architecture of a performance analysis suite for a custom ML accelerator
• Capture performance data from hardware via drivers and performance counters
• Implement tracing of host API calls and system events
• Correlate events across CPU, drivers, PCIe, accelerators, and hosts with precise timing
• Build analysis modules to identify bottlenecks and compute/ memory bound conditions
• Develop visualizations such as timelines and graphs to communicate performance
• Collaborate with hardware, firmware, driver, compiler, and ML teams to define tool needs

Job description

Etched in San Jose is seeking a highly skilled engineer to design and develop a performance analysis tool for our ML accelerator hardware. You will create data collection, tracing, and analysis capabilities to understand workload behavior and unlock the full potential of our hardware.

Join a cross-disciplinary team collaborating with hardware, firmware, drivers, compilers, and ML applications to deliver insights and optimizations for cutting-edge ML workloads.