Postdoc: Open-Source Accelerator Communication for AI

ETH SINGAPORE SEC LTD.; Singapore; On-site; SGD 70,000 - 110,000; Full time

Benefits offered by this job

Flexible hybrid work

25 days annual leave

Birthday leave

Dental benefits

Healthcare insurance

NS mark certification

Networking opportunities

Job summary

Singapore-ETH Centre in Singapore invites a Postdoctoral Researcher to co-lead the development of OCCL, an open-source accelerator communication library for distributed AI workloads, supporting host and kernel initiated communication across NVLINK and InfiniBand networks.

You will design primitives for ML training, optimize patterns, ensure portability across HPC substrates, collaborate with researchers in Singapore and Zurich, publish results, and supervise MSc/BSc students.

Qualifications

• PhD in Computer Science or related field.
• Strong background in parallel computing, compilers, performance optimisation and HPC.
• Excellent knowledge of C and C++.
• Knowledge of MPI and NCCL design principles.
• Experience with UCX, NVLINK, InfiniBand, UltraEthernet or related frameworks.
• Experience or publications in distributed ML systems or HPC.
• Experience managing large software engineering projects.
• Ability to work independently and communicate with remote colleagues.
• Proficiency in written and spoken English.

Responsibilities

• Co-lead design and development of OCCL for distributed AI workloads.
• Implement communication primitives optimised for ML training and inference.
• Investigate and optimise communication patterns for large-scale DL.
• Ensure portability across diverse hardware and networks (UCX, NVLINK, InfiniBand, UltraEthernet).
• Evaluate and optimise OCCL on GPU clusters and large systems.
• Collaborate with researchers in Singapore and Zurich to integrate OCCL with FastTrackAI.
• Maintain communication with supervisors and partners to align objectives.
• Publish key scientific outputs and papers.
• Co-supervise MSc and BSc students on related topics.

Skills

Parallel computing

Compilers

Performance optimisation

High-performance computing

English proficiency

Education

PhD in Computer Science or related field

Tools

MPI

NCCL

UCX

NVLINK

InfiniBand

UltraEthernet

Job description

Singapore-ETH Centre in Singapore invites a Postdoctoral Researcher to co-lead the development of OCCL, an open-source accelerator communication library for distributed AI workloads, supporting host and kernel initiated communication across NVLINK and InfiniBand networks.

You will design primitives for ML training, optimize patterns, ensure portability across HPC substrates, collaborate with researchers in Singapore and Zurich, publish results, and supervise MSc/BSc students.