Scientific Computation group at the University of Leeds is advertising a 3-year
Research Fellow position to work on floating-point error analysis and computer
arithmetic with a closing date of 14th of January, 2025.
Please apply at https://jobs.leeds.ac.uk/Vacancy.aspx?id=47395
An EPSRC-funded project “Informing Future Numerical Standards by
Determining Features of Non-Standard Mathematical Hardware” is a 3-year
project that aims to develop tools to improve our understanding of several
numerical features of floating-point hardware present in datacentre GPUs and
domain-specific machine learning accelerators. The main goal is to improve the
existent almost entirely manual methods by focusing on three core aspects:
automating, removing the need for specialized in-depth knowledge of
mathematics of floating-point arithmetic, and informing the public about the
features of the current hardware.
The work is spread across four work packages (WPs). Each WP contains a wide
variety of tasks that will provide you with a unique experience of working with
the low-level numerical software and mathematical hardware, as well as
understand mathematics that underpin that software and hardware. As
examples, it will involve creating models of mathematical hardware available in
literature by utilizing custom-precision simulators, as well as designing high-
level software wrappers that can manage low-level code targeted at an array of
diverse hardware devices: from NVIDIA and AMD datacentre GPUs to Intel
Ponte Vecchio accelerators.
You will become a member of the Scientific Computation research group
working on a wide array of topics, including partial-differential equations for
fluid dynamics, numerical linear algebra, scientific machine learning, computer
arithmetic, numerical analysis, algorithm design, software and hardware
benchmarking and performance optimization, and hardware design and
analysis. You will become a member of a new informal group Leeds
Mathematical Hardware and Software Lab.