October 19th: all active Maths research IT services available as usual
nvidia7 GPU server: excessive loading of this server by a user initially caused networking service to fail shortly before midnight last night followed by disconnection of the PCIe bus, which contains the GPUs. Control of this server has been regained but we are now having to strictly enforce GPU usage limits, retaining the option of terminating running user programs that exceed this limit.
NextGen HPC: this is now back up and running with the operating system on all nodes upgraded and many software packages have been updated to the latest available versions. The management node was damaged during May's power incident and has been replaced by a newer server while the main submission node, macomp001, has been replaced with a new server identical to macomp002, fitted with newer CPUs that support the Intel AVX and AVX2 instruction sets, matching all of the other Intel-powered compute nodes.
Failed memory DIMMs have been replaced in three of the compute nodes and these are now back online but there are possible issues with the latest version of Matlab 2026a which is now being actively investigated.
Relocation of GPU network: with all the main GPU servers and the clustor4 GPU storage server now moved over to the CAGB data centre, the associated 10Gbit/s local network has since moved here as well although data stored on clustor4 is already accessible to the GPUs as well as via the College gigabit (1 Gbit/sec) network. clustor4's mirror server, clustor4-mirror, remains in the Huxley server room for better data storage resilience through using a separate building.