GPU Performance Engineer

Two Sigma; New York (NY); On-site; USD 120,000 - 190,000; Full time

Job summary

Two Sigma seeks a GPU-focused engineer in New York to design and optimize CUDA kernels for financial workloads and to advance precision management across next-generation hardware. You will profile with Nsight and build reusable GPU libraries for modeling teams.

The role requires strong C++ and Python, expert CUDA, and a track record of delivering speedups on real workloads, with experience in multi-GPU and HPC environments.

Qualifications

• BS or MS in Science, Technology, Engineering or Math.
• Minimum 1 year of experience required; 4–10 years preferred.
• Expert-level CUDA programming: kernel development, memory management, stream and graph optimization.
• Deep understanding of GPU architecture: SM structure, warp scheduling, memory hierarchy (registers, shared memory, L1/L2, HBM).
• Experience with performance profiling and optimization of GPU workloads.
• Strong C++ and Python skills, as well as familiarity with mixed-precision computation and numerical stability.
• Track record of delivering meaningful speedups on real workloads (not just benchmarks).

Responsibilities

• Design and implement GPU-accelerated kernels for financial computation workloads.
• Optimize GPU code for throughput, latency, and memory efficiency across current and next-generation hardware (Blackwell, Rubin).
• Develop procedures for precision management (FP8/FP4 training and inference) in financial applications.
• Profile and optimize GPU workloads using NVIDIA tooling (Nsight Systems, Nsight Compute).
• Build reusable GPU libraries and abstractions that modeling teams can use without requiring deep CUDA expertise.
• Evaluate and integrate GPU-accelerated libraries (RAPIDS, CUTLASS, cuBLAS, TensorRT) for financial use cases.

Skills

CUDA programming

Kernel development

Memory management

Performance profiling

Python

C++

Numerical stability

Multi-GPU programming

Education

Bachelor's or Master's in STEM

Tools

Nsight Systems

Nsight Compute

RAPIDS

CUTLASS

cuBLAS

TensorRT

NCCL

MPI

RAPIDS cuDF

Job description

You will take on the following responsibilities:

• Design and implement GPU-accelerated kernels for financial computation workloads
• Optimize GPU code for throughput, latency, and memory efficiency across current and next-generation hardware (Blackwell, Rubin)
• Develop procedures for precision management (FP8/FP4 training and inference) in financial applications
• Profile and optimize GPU workloads using NVIDIA tooling (Nsight Systems, Nsight Compute)
• Build reusable GPU libraries and abstractions that modeling teams can use without requiring deep CUDA expertise
• Evaluate and integrate GPU-accelerated libraries (RAPIDS, CUTLASS, cuBLAS, TensorRT) for financial use cases

You should possess the following qualifications:

• BS or MS in Science, Technology, Engineering or Math
• Minimum 1 year of experience required; 4-10 years of experience preferred
• Expert-level CUDA programming: kernel development, memory management, stream and graph optimization
• Deep understanding of GPU architecture: SM structure, warp scheduling, memory hierarchy (registers, shared memory, L1/L2, HBM)
• Experience with performance profiling and optimization of GPU workloads
• Strong C++ and Python skills, as well as familiarity with mixed-precision computation and numerical stability
• Track record of delivering meaningful speedups on real workloads (not just benchmarks)

Preferred experience:

• Background in HPC, scientific computing, or computational finance
• Experience with multi-GPU and multi-node GPU programming (NCCL, MPI)
• Familiarity with GPU-accelerated data processing frameworks (RAPIDS, cuDF)