Staff AI Performance Engineer, Graphcore
2+ days a week in office
Advanced AI systems and services
Posted today
Advanced AI systems and services
501-1000 employees
Posted today
2+ days a week in office
501-1000 employees
Company mission
To make machine learning faster, easier and more intelligent.
Role
Who you are
- BS/MS in Computer Science, Electrical Engineering, or related field
- Experience with distributed systems and communication libraries (MPI, NCCL, UCX, libfabric)
- Strong programming skills in C++ and Python
- Experience profiling and optimizing HPC or AI/ML workloads
- Familiarity with ML benchmarks such as MLPerf
Desirable
- Experience with GPUs or accelerated computing architectures
- Knowledge of HPC networking and interconnect technologies (InfiniBand, RoCE)
- Familiarity with ML frameworks such as PyTorch or TensorFlow
- Understanding of ARM architectures and toolchains
- Strong debugging, profiling, and performance optimization skills
What the job involves
- Graphcore’s AI/ML training and inference infrastructure is rapidly scaling to meet the growing demands of AI workloads across mobile, edge, and datacenter environments
- This role focuses on optimizing performance across ARM-based architectures and large-scale distributed systems, ensuring efficiency, scalability, and reliability across the full hardware-software stack
- The System Engineering Performance team architects and optimizes high-performance infrastructure for large-scale datacenter deployments. The team works across hardware, software, networking, and system architecture to deliver cutting-edge AI solutions and ensure optimal system performance at scale
- Analyze ML models’ compute and memory requirements using roofline analysis and simulations
- Collaborate across hardware and software teams to optimize large-scale AI workloads
- Benchmark, monitor, and troubleshoot system performance across distributed systems
- Optimize communication stacks including MPI, NCCL, UCX, RDMA, and networking fabrics
- Profile and optimize AI workloads, focusing on performance bottlenecks
- Develop high-quality, ARM-compatible code and documentation
Our take
The hardware required to train AI models comes at a high cost, both financially and environmentally, producing millions of dollars in expenses and hundreds of tons of greenhouse gases. In response, Graphcore develops accelerator chips designed to expedite AI system training while significantly reducing power consumption.
AI accelerators like those crafted by Graphcore specialise in enhancing specific AI applications, such as artificial neural networks, deep learning, and machine learning. Utilising multicore designs and prioritising low-precision arithmetic or in-memory computing, these accelerators elevate the performance of large AI algorithms, achieving breakthrough results in natural language processing and computer vision.
Graphcore has ambitious plans to create an AI supercomputer, aiming to be one of the most powerful in its class. However, progress on this has extended beyond its planned 2024 debut, and the company was purchased by Softbank later that year. Though the acquisition by the Japanese multinational lowered Graphcore's valuation and highlighted structural problems in the local financing of UK tech pioneers, the buyout will undoubtedly aid the company in its ambitions to shape the future of AI computing.
Freddie
Company Specialist at Welcome to the Jungle
Insights
Company
Funding (last 2 of 6 rounds)
Dec 2020
$222m
SERIES E
Feb 2020
$150m
SERIES D
Total funding: $682m
Company benefits
- Work-Life Balance: We believe you should have a good work-life balance. Everyone's situation is different, so we offer flexible working arrangements and start and finish times, allowing you to choose when to work from home or come into the office
- Wellbeing: We want to ensure that our employees have the resources and support they need to prioritise their health and wellbeing
- Healthcare: We offer Private Medical Insurance (PMI) to our employees to help them take care of their health and that of their loved ones
- Progression: We prioritise learning and development as an integral part of our culture at Graphcore. Our Talent Development team is dedicated to equipping all our employees with the necessary tools, resources, and support to enhance their skills and reach their full potential in their careers
- Pension: We believe that our employees deserve the best when it comes to planning for their future
- Income Protection: Graphcore understands that unforeseen circumstances can happen, and we want to ensure our employees have the necessary financial security in place
Company HQ
Old City, Bristol, UK
Leadership
Nigel Toon
(Co-Founder & Exec Chair)Previously the President and CEO at picoChip Designs (processors for telecoms). As well as the VP and GM of Europe at Altera (semiconductors, merged with Intel).
Simon Knowles
(Co-Founder & CTO)Founded two successful semiconductor companies, Element14 and Icera, for a combined value of over $1 billion.
Salary benchmarks
We don't have enough data yet to provide salary benchmarks for this role.
Submit your salary to help other candidates with crowdsourced salary estimates.
Share this job
View 96 more jobs at Graphcore