Research Computing Platforms
We offer a range of advanced platforms for computationally intensive research. These services are free for UCL researchers to use on a fair share basis.
We currently provide three platforms: Legion, Grace and Emerald. Together, these platforms enable our users to perform all but the most demanding analyses and simulations. Our team of Research Computing Analysts provide support for users of these platforms as well as general advice and consultancy for researchers dealing with tricky research computing problems.
Legion is a centrally funded computing facility aimed at supporting all types of research at UCL. The heterogeneous architecture means that many different types of computational tasks are supported by a common infrastructure and software stack that can adapt quickly in response to the demands of the research community. Flexibility is key to Legion's design, allowing for continual evolution of the system to keep pace with advances in technology.
Grace was launched at the end of 2015 and is designed to
take care of large parallel computing tasks. It is designed specifically for this purpose with a high speed Infiniband network
linking together its compute nodes, allowing them to coordinate effort on a
single task very effectively. By sharing a software stack with Legion,
it is relatively simple for researchers to scale up their work to run on
Emerald is a GPGPU cluster: a specialist platform for performing computing
on Graphics Processing Units (GPUs). By using the hundreds of
computational cores found in GPU accelerators it is capable of performing some operations more than an
order of magnitude faster than traditional CPUs.
Popular uses include ray tracing, machine learning, image processing, molecular dynamics and protein folding calculations.
Emerald is a shared regional facility owned by the Science and Engineering South Consortium, of which UCL is a founding member. It is a unique facility within the UK and among the largest GPGPU systems in Europe, with a sustained capability of 114 teraflops (trillion floating-point operations per second).