Aitken (pronounced AY-ken) is a modular computing system housed in NASA's Modular Supercomputing Facility, located less than a mile from the main NAS building. Aitken is configured as follows:
Intel Cascade Lake-Based Compute Resources
- 4 E-Cells (8 E-racks)
- 1,116 Cascade Lake nodes
- 44,640 cores
- 214 terabytes (TB) total memory
AMD Rome-Based Compute Resources
- 16 racks
- 2,028 Rome nodes
- 259,584 cores
- 1.038 petabyte (PB) total memory
AMD Milan-Based Compute Resources
- 4 racks
- 512 Milan nodes
- 65,536 cores
- 262 terabytes (TB) total memory
Aitken's total theoretical peak performance: 15.49 petaflops (PF)
Hostnames
For the Cascade Lake-based resources, there are four enclosures (individual rack units, or IRUs) in each E-rack, with 36 compute nodes per enclosure. For every two racks, there is a rack leader controlling them. The naming convention for the 288 nodes residing in every two racks (2 racks x 4 enclosures x 36 nodes) uses only odd rack numbers. Therefore, the hostnames of the Aitken compute nodes are r[x]i[0-7]n[0-35], where x are odd numbers between 901 and 907.
For the Rome-based resources, there are four enclosures in each rack, eight compute trays in each enclosure, and four nodes in each compute tray. The hostnames of the Aitken Rome nodes are r[x]c[1-4]t[1-8]n[1-4], where x are numbers between 201 – 216.
For the Milan-based resources, there are four enclosures in each rack, eight compute trays in each enclosure, and four nodes in each compute tray. The hostnames of the Aitken Milan nodes are r[x]c[1-4]t[1-8]n[1-4], where x are numbers between 705 and 708.
Processor, Memory, and Network Subsystems Statistics
The following table provides detailed configuration statistics for the processor, memory, and network subsystems for the Aitken compute nodes:
| Aitken Processor, Memory, and Network Subsystems Statistics |
| Architecture |
HPE SGI 8600-XA730i Gen10 |
HPE Apollo 9000 |
HPE Apollo 9000 |
| Processor |
| CPU |
Cascade Lake
20-Core
Xeon Gold 6248 |
Rome
64-Core
EPYC 7742 |
Milan
64-Core
EPYC 7763 |
| Newest Instruction Set |
AVX-512 |
AVX2 |
AVX2 |
| Hyper-Threading |
ON |
ON |
ON |
| TurboBoost |
ON |
ON |
ON |
| non-Turbo base CPU-Clock |
2.5 GHz |
2.25 GHz |
2.45 GHz |
| Maximum Double Precision Floating Point Operations per Cycle per Core |
32 |
16 |
16 |
| # of Cores/Node |
40 |
128 |
128 |
| Total # of Nodes |
1,116 |
2,028 |
512 |
| Total # of Cores |
44,640 |
259,584 |
65,536 |
| Total Double Precision TFlops |
3,686 |
9,437 |
2,569 |
| Memory |
| L1 Cache |
Local to each core;
Instruction cache: 32K
Data cache: 32K;
Associativity: 8;
Cache line size: 64B |
Local to each core;
Instruction cache: 32K
Data cache: 32K;
Associativity: 8;
Cache line size: 64B |
Local to each core;
Instruction cache: 32K
Data cache: 32K;
Associativity: 8;
Cache line size: 64B |
| L2 Cache |
1 MB per core;
Associativity: 16;
Cache line size: 64B |
512 KB per core;
Associativity: 8;
Cache line size: 64B |
512 KB per core;
Associativity: 8;
Cache line size: 64B |
| L3 Cache |
27.5 MB shared non-inclusive by the 20 cores;
Associativity: full;
Cache line size: 64B |
16 MB shared among 4 cores in a core complex;
256 MB per socket;
Associativity: 16;
Cache line size: 64B |
32 MB shared among 8 cores in a core complex;
256 MB per socket;
Associativity: 16;
Cache line size: 64B |
| TLB |
Local to each core |
Local to each core |
Local to each core |
| Default Page Size |
4 KB |
4 KB |
4 KB |
| Memory/Core |
4.8 GB; DDR4 |
4.0 GB; DDR4 |
4.0 GB; DDR4 |
| Total Memory/node |
192 GB |
512 GB |
512 GB |
| Memory Speed and Bandwidth |
2,933 MHz;
6 channels;
141 GB/sec read/write |
3,200 MHz;
8 channels;
204.8 GB/sec read/write |
3,200 MHz;
8 channels;
204.8 GB/sec read/write |
| Inter-socket Interconnect |
Ultra-Path Interconnect;
5.2 GHz,
10.4 GT/s,
or 62.4 GB/sec |
Global Memory Interconnect;
8.0 GHz,
16.0 GT/s,
or 96.0 GB/sec |
Global Memory Interconnect;
8.0 GHz,
16.0 GT/s,
or 96.0 GB/sec |
| Inter-node Network |
| IB Device on Node |
Dual single-port 4x EDR IB Mezzanine card (2 single-port HCAs); 100 Gbits/s |
Dual single-port 4x HDR IB Mezzanine card (2 single-port HCAs); 200 Gbits/s |
Dual single-port 4x HDR IB Mezzanine card (2 single-port HCAs); 200 Gbits/s |
| IB Switches Between Nodes |
4x HDR;
200 Gbits/s |
4x HDR;
200 Gbits/s |
4x HDR;
200 Gbits/s |
