SuperMUC Petascale System
SuperMUC is the name of the new supercomputer at Leibniz-Rechenzentrum (Leibniz Supercomputing Centre) in Garching near Munich (the MUC suffix is borrowed from the Munich airport code). With more than 155.000 cores and a peak performance of 3 Petaflop/s (=10^15 Floating Point Operations per second) in June 2012 SuperMUC is one of the fastest supercomputers in the world.
System purpose and target users
SuperMUC strengthens the position of Germany's Gauss Centre for Supercomputing in Europe by delivering outstanding compute power and integrating it into the European High Performance Computing ecosystem. With the operation of SuperMUC, LRZ will act as an European Centre for Supercomputing and will be Tier-0 centre of PRACE, the Partnership for Advanced Computing in Europe.
SuperMUC is available to all European researchers to expand the frontiers of science and engineering.
- 155,656 processor cores in 9400 compute nodes
- >300 TB RAM
- Infiniband FDR10 interconnect
- 4 PB of NAS-based permanent disk storage
- 10 PB of GPFS-based temporary disk storage
- >30 PB of tape archive capacity
- Powerful visualization systems
- Highest energy-efficiency
SuperMUC uses a new, revolutionary form of warm water cooling developed by IBM. Active components like processors and memory are directly cooled with water that can have an inlet temperature of up to 40 degrees Celsius. The "High Temperature Liquid Cooling" together with very innovative system software promises to cut the energy consumption of the system. In addition, all LRZ buildings will be heated re-using this energy.
Why "warm" water cooling?
Typically water used in data centers has an inlet temperature of approx 16 degrees Celsius and, after leaving the system, an outlet temperature of approx. 20 degrees Celsius. To make water with 16 degrees Celsius requires complex and energy-hungry cooling equipment. At the same time there is hardly any use for the warmed-up water as it is too cold to be uses in any technical processes.
SuperMUC allows an increased inlet temperature. It is easily possible to provide water having up to 40 degrees Celsius using simple "free-cooling" equipment as outside temperatures in Germany hardly ever exceed 35 degrees Celsius. At the same time the outlet water can be made quite hot (up to 70 degrees Celsius) and re-used in other technical processes - for example to heat buildings or in other technical processes.
By reducing the number of cooling components and using free cooling LRZ expects to save several millions of Euros in cooling costs over the 5-year lifetime of the system.
Figure: SuperMUC in the computer room
System Configuration Details
LRZ's target for the architecture is a combination of a large number of moderately powerful compute nodes, with a peak performance of several hundred GFlop/s each, and a small number of fat compute nodes with a large shared memory. The network interconnect between the nodes allows for perfectly linear scaling of parallel applications up to the level of more than 10,000 tasks.
SuperMUC consists of 18 Thin Node Islands and one Fat Node Island which is at first also used as the Migration System SuperMIG. Each Island contains more than 8,192 cores. All compute nodes within an individual Island are connected via a fully non-blocking Infiniband network (FDR10 for the Thin nodes / QDR for the Fat Nodes). Above the Island level, the high speed interconnect enables a bi-directional bi-section bandwidth ratio of 4:1 (intra-Island / inter-Island).
In June 2014 the Intel Xeon Phi based cluster SuperMIC has been integrated into SuperMUC. Details on SuperMIC can be found under http://www.lrz.de/services/compute/supermuc/supermic/.
The SuperMUC system will be expanded in 2015 by doubling the performance.
Figure: Schematic view of SuperMUC
|Installation Phase||Phase 1||Phase 2|
|Thin Node Islands||Fat Node Island||Many Core Island||Thin Node Islands|
|System||IBM System x iDataPlex||BladeCenter HX5|
|Processor Types||Sandy Bridge-EP
Intel Xeon E5-2680
Intel Xeon E7-4870 10C
|Ivy-Bridge (IB) and Intel Xeon Phi 5110P||Intel Haswell|
|Number of Islands||18||1||1|
|Nodes per Island||512||205||32|
|Processors per Node||2||4||
2 (IB) 2.6 GHz + 2 Phi 5110P
|Cores per Processor||8||10||8 (IB) + 60 (Phi)|
|Cores per Node||16||40||16 (host) + 120 (Phi)|
|Logical CPUs per Node (Hyperthreading)||32||80||32 (host) + 480 (Phi)|
|Total Number of nodes||9216||205||32|
|Total Number of cores||147,456||8200||4352|
|Peak Performance [PFlop/s]||3.185||0.078||tbd||3.2|
|Linpack Performance [PFlop/s]||2.897||0.065|
|Total size of memory [TByte]||288||52||2.56|
|Memory per Core [GByte]
(typically available for applications)
|4 (host) + 2 x 0.13 (Phi)|
|Size of shared Memory per node [GByte]||32||256||64 (host) + 2 x 8 (Phi)|
|Bandwidth to Memory per node [Gbyte/s]||102.4||136.4||Phi: 384|
|Latency to local memory [ns (cylces)]||~ 50 (~135)||~70 (~170)|
|Latency to remote memory [ns (cylces)]||~ 90 (~240)||~120 (~200)|
|Level 3 Cache Size (shared) [Mbyte]||20||24|
|Level 2 Cache Size [kByte]||256||256||60 x 512|
|Level 1 Cache Size [kByte], Associativity||32@ 8 way||32||32|
|Level 3 Cache Bandwidth and Latency (shared) [byte/cycle]||1 x 32 @ 31 cycles||1 x 32|
|Level 2 Cache Bandwidth and Latency [byte/cycle]
Latency is much longer, if data are also in L1 or L2 of other core.
|1 x 32 @ 12 cycles||1 x 32|
|Level 1 Cache Bandwidth and Latency [byte/cycle]
Latency is much longer, if data are also in L1 or L2 of other core.
|2 x 16 @ 4 cycles||2 x 16|
|Level 3 Cache line Size [Byte]||64||64|
|Electrical power consumption of total system [MW]||< 2.3|
|Network Technology||Infiniband FDR10||Infiniband FDR14|
|Intra-Island Topology||non-blocking Tree|
|Inter-Island Topology||Pruned Tree 4:1|
|Bisection bandwidth of Interconnect [TByte/s]||35.6|
|Filesystem for SCRATCH and WORK||IBM GPFS|
|File System for HOME||NetApp NAS|
|Size of parallel storage [Pbyte]||10||5|
|Size of NAS user storage [PByte]||1.5 (+ 1.5 for replication)|
|Aggregated bandwidth to/from GPFS [GByte/s]||200||100|
|Aggregated bandwidth to/from NAS storage [GByte/s]||10||5|
|Login Servers for users||5|
|Service and management Servers||12|
|Archive and Backup Software||IBM TSM|
|Capacity of Archive and Backup Storage [PByte]||> 30|
Details on processors
SuperMUC uses following software components:
- Suse Linux Enterprise Server (SLES)
- System management: xCat from IBM
- Batch processing: Loadleveler from IBM
From the user side a wide range of compilers, tools and commercial and free applications is provided. Many scientists also build and run their own software.
SuperMUC has a powerful I/O-Subsystem which helps to process large amounts of data generated by simulations.
Home file systems
Permanent storage for data and programs is provided by a 16-node NAS cluster from Netapp. This primary cluster has a capacity of 2 Petabytes and has demonstrated an aggregated throughput of more than 10 GB/s using NFSv3. Netapp's Ontap 8 "Cluster-mode" provides a single namespace for several hundred project volumes on the system. Users can access multiple snapshots of data in their home directories.
Data is regularly replicated to a separate 4-node Netapp cluster with another 2 PB of storage for recovery purposes. Replication uses Snapmirror-technology and runs with up to 2 GB/s in this setup.
Storage hardware consists of >3400 SATA-Disks with 2 TB each protected by double-parity RAID and integrated checksums.
Work and Scratch areas
For highest-performance checkpoint I/O IBM's General Parallel File System (GPFS) with 10 PB of capacity and an aggregated throughput of 200 GB/s is available. Disk storage subsystems were built by DDN.
Tape backup and archives
LRZ's tape backup and archive systems based on TSM (Tivoli Storage Manager) from IBM are used for or archiving and backup. The have been extended to provide more than 30 Petabytes of capacity to the users of SuperMUC. Digital long-term archives help to preserve results of scientific work on SuperMUC. User archives are also transferred to a disaster recovery site.
Visualization and Support systems
SuperMUC will be connected to powerful visualization systems: the new LRZ office building houses a large 4K stereoscopic powerwall as well as a 5-sided CAVE artificial virtual reality environment.