The ExCALIBUR H&ES programme has supported the Durham HPC hardware lab which contains a range of latest generation compute architectures and is hosted by COSMA at Durham University.
A new AMD MI300X GPU system has been made available for code porting, testing and benchmarking. The AMD Instinct™ MI300X is AMD’s latest GPU and a powerhouse for AI and HPC workloads. Each GPU contains 192GB of HBM3 and is paired with large caches and a high bandwidth to keep the compute units fed with data. The Durham testbed contains eight MI300X GPUs and these are housed in a single node and have been provided courtesy of Dell and AMD. In addition to the hardware, the team have installed the AMD ROCm software suite and CUDA code can be ported using the hipify tool, although further performance tuning is then likely to be necessary in order to gain the maximum performance from the GPU.
To access this system please sign up to project do018 on the DiRAC SAFE https://safe.epcc.ed.ac.uk/dirac and let Alastair () know what you intend to use this system for. The team in Durham will be requesting feedback from people who use it, so please feel free to send them any useful information and comments.