Parts/Interconnect/IBM BlueGene 3D torus
Why this part
BlueGene ran five networks at once: a 3D torus for point-to-point, a separate tree for collectives, a global barrier network, a control network and Ethernet for I/O. Splitting them means an all-reduce does not compete with neighbour exchange for the same wires, and a barrier completes in microseconds across a hundred thousand nodes. All of it sat on the same die as the processor, so BlueGene has no interconnect vendor at all, which is why fabric-share charts consistently understate IBM's historical position.
Systems using this part
- Systems
- 13
- Rmax underneath
- 1.70 PFlop/s
- Units recorded
- -
0 of 13 edges disclose a quantity
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Shaheen 2009 | Interconnect | Saudi Arabia | - | 190.90 TFlop/s | IBM | #14 → |
| Intrepid 2008-01 | Interconnect | USA | - | 458.61 TFlop/s | IBM | #3 → |
| Frontier2 2008 | Interconnect | France | - | 95.45 TFlop/s | IBM | #14 → |
| JUGENE 2007-10 | Interconnect | Germany | - | 190.90 TFlop/s | IBM | #2 → |
| Genius (Garching) 2007-10 | Interconnect | Germany | - | 21.91 TFlop/s | IBM | #41 → |
| New York Blue 2007-06 | Interconnect | USA | - | 82.16 TFlop/s | IBM | #10 → |
| CyBlue 2006-01 | Interconnect | USA | - | 4.71 TFlop/s | IBM | #99 → |
| JUBL 2006 | Interconnect | Germany | - | 37.33 TFlop/s | IBM | #8 → |
| Stella 2005-04 | Interconnect | Netherlands | - | 27.45 TFlop/s | IBM | #6 → |
| BGW 2005 | Interconnect | USA | - | 91.29 TFlop/s | IBM | #2 → |
| Intimidata 2004-12 | Interconnect | USA | - | 13.78 TFlop/s | IBM | #42 → |
| BlueGene/L 2004-11 | Interconnect | USA | - | 478.20 TFlop/s | IBM | #1 → |
| Blue Fern | Interconnect | New Zealand | - | 9.43 TFlop/s | IBM | #99 → |
“Units recorded” sums only the edges where a public source states a quantity. It is a floor, not a total, and should never be read as installed base.