Tensordyne
Systems Performance Modeling Engineer
Sunnyvale, CA +1 · Full-time · Posted Sep 23, 2026
- Location
- Sunnyvale, CA
San Jose, California, United States - Education
- MS or higher in Computer Science, Computer Engineering, Electrical Engineering, or a related field.
About the job
Build and validate performance models and tools that predict generative AI inference performance on Tensordyne systems, from individual accelerators to clusters.
What you'll do
- Develop simulators and analytical models for compute, memory, collective communication, and network fabrics; validate predictions against hardware measurements.
- Build trace-capture and replay tools, model serving strategies and collective algorithms, and run design-space experiments.
- Produce performance analyses and projections for architecture, system configuration, product, and customer decisions.
What they ask for
- Hands-on experience building performance models, simulators, or analytical tools for ML workloads, distributed systems, or computer architecture.
- Understanding of distributed ML execution, parallelism strategies, and collective communication.
- Working knowledge of compute, memory, interconnect, and networking architecture; experience comparing predictions with real measurements and debugging discrepancies.
- Strong C++ and Python programming skills; ability to independently plan experiments and communicate results.
Nice to have
- Familiarity with LLM inference serving and experience modeling or benchmarking collective communication libraries such as NCCL or RCCL.
- Data center or HPC networking, accelerator profiling, hardware/software co-design, or relevant publications and open-source contributions.
Tools
C++, Python, NCCL, RCCL, RDMA/RoCE
AI focus
Models and tools for predicting, analyzing, and improving multimodal generative AI inference performance and serving strategies.
Model-written summary · the original posting is authoritative