Performance Engineer
PDT Partners5 months ago
Responsibilities
- Analyze system performance to improve researcher throughput and velocity.
- Optimize infrastructure and system-level efficiency across Python, PyTorch, operating systems, networking, storage, CPU, and GPU layers.
- Read software layers and provide code changes or pull requests that improve performance.
- Improve computational efficiency to free capacity and reduce costs.
- Support scaling of core models through efficient implementations.
- Design and implement systems that improve performance telemetry.
- Conduct proof-of-concept evaluations and contribute to system design.
- Identify and implement optimization opportunities across the technology stack.
Requirements
- Strong proficiency in Linux and its performance engineering toolset.
- Experience optimizing PyTorch workloads using GPUs and CUDA.
- Deep understanding of the hardware-software interface.
- Coding, micro-optimization, and familiarity with multiple programming languages.
- Ability to analyze performance across the stack while adapting to diverse engineering challenges.
Benefits
- Hybrid position requiring at least 3 days per week in the New York City office.
About PDT Partners
PDT, which stands for Process Driven Trading, was founded by Peter Muller in 1993 as a proprietary trading group within Morgan Stanley. In 2013, after establishing a 19-year track record finding inefficiencies in financial markets, the group spun out of Morgan Stanley to become an independent investment management firm, PDT Partners. Today we are a leader in the field of quantitative finance, known for our rigorous scientific, systematic, and model-driven approach to investing, with multiple billions in assets under management. Peter Muller, Founder and CEO, and a small group of managing partners lead around 200 scientists, engineers, and professionals in New York and London.