Project Description:
The ROCm Communication Collectives Library (RCCL) is a stand-alone library that provides multi-GPU and multi-node collective communication primitives optimized for AMD GPUs. It uses PCIe and xGMI high-speed interconnects.
Responsibilities:
* Provide deep technical leadership and guidance for GPU communication technologies, define the technical vision and direction for the GPU communication software stack.
* Engage with executives and key stakeholders to provide insight into industry trends and recommend strategic initiatives. Influence the future direction of the company's technical portfolio.
* Represent AMD in leadership positions at industry organizations and standards bodies.
* Engage with clients and industry partners to deeply understand technical needs, ensuring their satisfaction with tailored solutions that leverage your experience in strategic customer engagements and architectural wins.
* Collaborate with hardware and software architects, system engineers and business teams in identifying requirements and building roadmaps for future products.
* Mentor engineers and technical leaders, fostering a culture of innovation and excellence. Help develop the next generation of leaders through coaching, training, and feedback.
Mandatory Skills Description:
* Experience architecting and developing communication software solutions for accelerators using RDMA and accelerator-to-accelerator fabrics (eg. Infinity Fabric, UALink), from low-level device drivers and OS internals up through applications and AI/ML frameworks
* Experience with communication stacks and low-level GPU drivers.
* Experience developing or modifying compiler toolchains or debugging tools.
* Deep understanding of graphics pipeline internals and GPU architecture
* Deep expertise with distributed programming models (MPI, SHMEM), and the implementation and optimization of collective communication algorithms
* Deep expertise with RoCE, RDMA, and network topologies
* Experience with system software development in C/C++, and GPU software development and parallel programing
* Analytical and performance analysis skills
* Effective communication and problem-solving skills
* Proven history of communication software thought leadership, backed with patents, publications, and participation in industry standards bodies
Nice-to-Have Skills Description:
* Advanced degrees, such as Master's or Ph. D. are preferred
Languages:
* English: C1 Advanced