Series – ada:100 tranche 2
21 brand new DELL C6620 servers, ready for power and networking. Each server has two 24 core Xeon Gold 6442 processors with 384GB of RAM. And to top it all off, a DELL R760XA with four L40S GPU cards. …
UCT HPC cluster updates
IMPORTANT – SOFTWARE INSTALLS We have noticed an increasing number of people installing software\libraries directly via the head node. Please do not do this. Many of these installs require a large amount of RAM or induce heavy load on the…
The return of Hex
Our new cluster is now live. The head node is hex.uct.ac.za, for those that remember our rollout of Dell C6145’s in 2013. At present the cluster consists of 16 C6620 nodes with 48 cores and 384GB of RAM each. …
New HPC racks
Over the past few days the HPC and networks team relocated 3 racks worth of servers to a new location in the upper campus data center. This will free up space for the continuation of the refurbishment project. The end…
HPC moving to new racks in UCDC
The UCT HPC cluster is being migrated to its final location in the newly refurbished ICTS data center. The move will take place in the first week of June during which time the cluster will be unavailable. The cluster will…
HPC unavailable 20-21 April
The HPC cluster will be unavailable during the weekend of the 20th of April due to a network upgrade. We will be placing the cluster into draining mode at 17:00 on Friday the 19th at which point no new jobs…
HPC degraded
On Monday 5 Jun at 13:45 an environmental event in the Upper Campus Data Center caused damage to the HPC rack. Currently 5 worker nodes are offline. We have ordered replacement parts from our suppliers, however the implication is that…
New GPU server
We have added another GPU server to our a100 partition. This server was purchased with funding from several groups as well as ICTS and additional resources will be dedicated to the shared a100free account. The server contains four a100-80GB cards…
Time based analysis of core\energy usage
Being able to analyze the energy usage of every core in every CPU of the cluster enables us to detect jobs that are not making good use of allocated cores over time. Here is a node that is using 1…
Performance graphs
We have moved away from Cacti\Nagios for graphing and now make use of Grafana. Unfortunately there is no public facing portal for Grafana, however there is a way to export graphs as static png files, so we have set up…