Hybrid ML Training Performance Engineer — GPU/HPC

Tower Research Capital; New York (NY); Hybrid; USD 180,000 - 220,000; Full time

Benefits offered by this job

Generous PTO

Financial wellness plans

Hybrid work opportunities

Free meals daily

Wellness reimbursements

Volunteer opportunities

Social events and learning programs

Continuous learning opportunities

Job summary

Tower Research Capital is a leading quantitative trading firm that builds high-performance training systems for machine learning models. In New York, you will bridge research and engineering to accelerate end-to-end training from data ingestion to hardware utilization.

You will work on distributed training optimization, GPU kernel development, and performance benchmarking to enable researchers to iterate across complex models. A hybrid work setup is available in NYC.

Qualifications

• 3+ years of experience optimizing machine learning training workloads in high-performance, distributed or large-scale computing environments.
• Deep knowledge of machine learning frameworks such as PyTorch or JAX, including their execution models and distributed-training capabilities.
• Strong programming skills in Python and C++, with experience developing or optimizing performance-critical systems.
• Proven experience with GPU kernel development and optimization using CUDA, Triton, CUTLASS, cuBLAS, cuDNN or related libraries.
• Strong understanding of GPU architecture and memory hierarchy.
• Experience with distributed-training technologies and communication libraries such as NCCL, DeepSpeed, Megatron-LM, XLA or equivalents.
• Proficiency with performance-analysis tools such as Nsight Systems, Nsight Compute, PyTorch Profiler or comparable tools.
• Understanding of high-performance networking, storage, and accelerator interconnects (InfiniBand, RDMA, NVLink).
• Demonstrated ability to benchmark heterogeneous compute platforms and make data-driven performance/cost recommendations.

Responsibilities

• Train and optimize scalable ML models across CPUs/GPUs and accelerator platforms.
• Design and optimize distributed training strategies (data, tensor, pipeline, model parallelism).
• Improve communication efficiency across multi-GPU and multi-node environments.
• Analyze and optimize the full training pipeline from data loading to checkpointing.
• Develop GPU kernels and framework components for ML workloads.
• Collaborate with ML researchers, HPC engineers, and systems teams to deliver efficient training systems.

Skills

ML training optimization

PyTorch/JAX

Python

C++

GPU kernel development

NCCL/DeepSpeed

Profiling tools

Distributed training

Tools

CUDA

Triton

CUTLASS

cuBLAS

cuDNN

Nsight

Kubernetes

Slurm

Ray

Megatron-LM

Job description

Tower Research Capital is a leading quantitative trading firm that builds high-performance training systems for machine learning models. In New York, you will bridge research and engineering to accelerate end-to-end training from data ingestion to hardware utilization.

You will work on distributed training optimization, GPU kernel development, and performance benchmarking to enable researchers to iterate across complex models. A hybrid work setup is available in NYC.