Difference between revisions of "Bert and Ernie - An Overview"
Ibers-admin (talk | contribs) (→The IBERS HPC) |
|||
| (16 intermediate revisions by 2 users not shown) | |||
| Line 1: | Line 1: | ||
| − | + | == What is HPC? == | |
| − | We have | + | Scientific computing generally revolves around researchers performing complex calculations as fast as possible. The super computers of the 70s, 80s and 90s were custom built to provide researchers with access to a machine with more memory (RAM) and CPU cores than the small, slow and expensive desktop computers of the era. These large machines such as those build by Cray and SGI were expensive and took a huge amount of space, power and cooling. |
| + | |||
| + | In the early 2000s, the first multi core processors were released which dramatically changed the ability for users to being to take advantage of parallelisation when performing calculations. Today, the term High Performance Computing (HPC) is used to describe a form of supercomputing based upon off the shelf hardware. HPC clusters are often built from the same hardware as that which is found in high-end desktop workstations or web servers. | ||
| + | |||
| + | == The IBERS HPC == | ||
| + | |||
| + | Names Bert and Ernie, the IBERS HPC consists of a master node (ernie), a login node (bert), 12 compute nodes and five storage nodes. | ||
| + | |||
| + | The combined compute capacity is as follows; | ||
| + | |||
| + | We have 416 CPU cores, 4TB RAM and 43TB home directory storage capacity, 109TB of temporary scratch disk, 12TB of fast solid state disk scratch. This is backed by 250TB and 680TB storage arrays for storage of sequencing data. | ||
| + | |||
| + | Currently we are in the process of upgrading some of the older compute nodes and these numbers will be changing over late 2022 and early 2023. | ||
Specs Per node; | Specs Per node; | ||
| − | node001 - | + | node001 - currently broken and awaiting replacement |
| − | 4 x AMD Opteron(TM) Processor 6220 | + | |
| − | + | node002 (fat queue) - 16 Cores, Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.1GHz, 1024GB RAM | |
| + | |||
| + | node003 (intel queue) - 8 Cores, Intel(R) Xeon(R) CPU X5647 @ 2.93GHz, 192GB RAM | ||
| + | |||
| + | node004 (intel queue) - 8 Cores, Intel(R) Xeon(R) CPU X5647 @ 2.93GHz, 192GB RAM | ||
| + | |||
| + | node005 (amd queue) - not currently in service, 32 Core 4 x AMD Opteron(TM) Processor 6220, 98GB RAM | ||
| − | + | node006 (amd queue) - not currently in service, 32 Cores 4 x AMD Opteron(TM) Processor 6220, 98GB RAM | |
| − | |||
| − | |||
| − | + | node007 (amd queue) - not currently in service, 32 Cores 4 x AMD Opteron(TM) Processor 6220, 98GB RAM | |
| − | |||
| − | |||
| − | + | node008 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM | |
| − | |||
| − | |||
| − | + | node009 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM | |
| − | 4 x AMD Opteron( | ||
| − | |||
| − | + | node010 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM | |
| − | 4 x AMD Opteron( | ||
| − | |||
| − | + | node011 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM | |
| − | 4 x AMD Opteron( | ||
| − | |||
| − | + | node012 (highmem queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 512GB RAM | |
| − | 4 x AMD Opteron(tm) Processor 6376 | ||
| − | |||
| − | + | node013 (GPU queue) - 32 Cores, 1 x AMD EPYC 7452 Processor, 768GB RAM, NVIDIA A100 GPU | |
| − | |||
| − | |||
| − | |||
| − | |||
| − | |||
| − | + | [[File:Hpc.png]] | |
| − | |||
| − | |||
Latest revision as of 11:40, 27 October 2022
What is HPC?
Scientific computing generally revolves around researchers performing complex calculations as fast as possible. The super computers of the 70s, 80s and 90s were custom built to provide researchers with access to a machine with more memory (RAM) and CPU cores than the small, slow and expensive desktop computers of the era. These large machines such as those build by Cray and SGI were expensive and took a huge amount of space, power and cooling.
In the early 2000s, the first multi core processors were released which dramatically changed the ability for users to being to take advantage of parallelisation when performing calculations. Today, the term High Performance Computing (HPC) is used to describe a form of supercomputing based upon off the shelf hardware. HPC clusters are often built from the same hardware as that which is found in high-end desktop workstations or web servers.
The IBERS HPC
Names Bert and Ernie, the IBERS HPC consists of a master node (ernie), a login node (bert), 12 compute nodes and five storage nodes.
The combined compute capacity is as follows;
We have 416 CPU cores, 4TB RAM and 43TB home directory storage capacity, 109TB of temporary scratch disk, 12TB of fast solid state disk scratch. This is backed by 250TB and 680TB storage arrays for storage of sequencing data.
Currently we are in the process of upgrading some of the older compute nodes and these numbers will be changing over late 2022 and early 2023.
Specs Per node;
node001 - currently broken and awaiting replacement
node002 (fat queue) - 16 Cores, Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.1GHz, 1024GB RAM
node003 (intel queue) - 8 Cores, Intel(R) Xeon(R) CPU X5647 @ 2.93GHz, 192GB RAM
node004 (intel queue) - 8 Cores, Intel(R) Xeon(R) CPU X5647 @ 2.93GHz, 192GB RAM
node005 (amd queue) - not currently in service, 32 Core 4 x AMD Opteron(TM) Processor 6220, 98GB RAM
node006 (amd queue) - not currently in service, 32 Cores 4 x AMD Opteron(TM) Processor 6220, 98GB RAM
node007 (amd queue) - not currently in service, 32 Cores 4 x AMD Opteron(TM) Processor 6220, 98GB RAM
node008 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM
node009 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM
node010 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM
node011 (amd queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 256GB RAM
node012 (highmem queue) - 64 Cores 4 x AMD Opteron(tm) Processor 6376 512GB RAM
node013 (GPU queue) - 32 Cores, 1 x AMD EPYC 7452 Processor, 768GB RAM, NVIDIA A100 GPU
