Location: Bristol, UK
Tasks
Use your programming skills and curiosity to help AI models run efficiently on next-generation Graphcore hardware. In AI Compute Libraries & Performance, you'll contribute to C++ libraries and compute kernels for machine learning workloads. Your work will include:
•Developing compute kernels: Implement and improve operations such as matrix multiplication (GEMM), convolutions, reductions, elementwise operations and fused kernels that combine multiple operations.
•Understanding and improving performance: Use profiling and microbenchmarks to measure execution speed and investigate how threading, cache locality, memory layout and kernel launch efficiency affect performance.
•Checking correctness and reliability: Build regression tests and numerical checks, investigate issues and improve the quality of the libraries our customers use.
Through these projects, you'll develop practical experience connecting numerical computing with how code runs on hardware, helping deliver software that is fast, accurate and reliable. Over the 12 weeks, you'll build performance engineering skills through hands-on work on compute kernels, benchmarks and tests. Profiling and experiments will help you understand where time is spent and how changes to code and data layout affect performance. Numerical validation and regression testing will show you how to check that an optimisation preserves correct results and continues to work as software changes. Working with specialists and discussing technical trade-offs will help you develop your engineering judgement, explain your findings and connect mathematical operations to efficient implementations on hardware.
Requirements
•Currently studying towards a bachelor's or master's degree in Computer Science, Maths, Machine Learning, Data Science or a related field
•Experience with C or C++11, plus familiarity with Python or scripting tools for automation and testing
•An understanding of linear algebra, numerical methods or scientific computing
•Problem-solving skills and collaborative approach to unfamiliar technical challenges
•Previous industry experience is not required. Tell us what you built or investigated through coursework, personal or team projects or other relevant experience, how you approached it and what you learned.
•Coursework or experience with machine learning frameworks, parallel programming, code optimisation, maths libraries or performance analysis would be useful. These are additional strengths, rather than essential requirements.
Conditions
•Start date: June 2027
•Application deadline: Midnight, 29th November 2026 (we encourage you to apply early, as we may close applications before the advertised deadline)
•12-week summer internship with hands-on experience on real projects
•You'll join a team building high-performance compute libraries that help customers get more from Graphcore hardware, bringing together machine learning, numerical computing and accelerator performance
•You'll contribute to clearly scoped projects with guidance from engineers working in performance and low-level software
•A buddy will help you settle in and get familiar with tools and ways of working
•Your manager will meet with you regularly to agree goals, offer feedback and discuss your development, including time for learning
Benefits
•Flexible working: find a balance between your internship and life outside work
•5 days of annual leave during your 12-week internship
•Life in the office: get to know your colleagues in spaces designed for collaboration, with free food, a great selection of ice cream and an on-site barista
•Learning and development: access to O'Reilly, our online learning platform
Sponsorship
Applicants must already have the right to work in the UK for the full duration of the programme, without requiring visa sponsorship. Graphcore is unable to sponsor work visas for its Early Careers opportunities.
About Graphcore
At Graphcore, we're building next generation AI compute: from semiconductors and software to systems at datacenter scale. We are owned by SoftBank Group and work with other companies in the SoftBank family to deliver AI compute to customers working at the AI frontier.
How to apply
To apply, answer the application questions and upload your CV and cover letter. You'll hear the outcome of your application by early December at the latest. If successful, you'll be invited to complete a technical assessment. Candidates who reach the final stage will then be invited to attend in-person assessments in early spring.
