Parts/Interconnect/Mellanox InfiniBand DDR
Interconnect · DDR · 2006
Mellanox InfiniBand DDR
20 Gb/s InfiniBand, the fabric under Roadrunner. Supplied by Mellanox as an independent company.
Why this part
Roadrunner is where an open-standard commodity fabric replaces a proprietary one in a number-one machine. InfiniBand was designed by a committee for enterprise storage and general server clustering, so it arrived cheap because a market larger than HPC was paying for it, which is the same logic that put commodity CPUs in ASCI Red. Everything about the current position of NVIDIA's networking business follows from InfiniBand having won that argument.
Systems using this part
- Systems
- 22
- Rmax underneath
- 2.28 PFlop/s
- Units recorded
- -
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Millipede 2010-05-17 | Interconnect | Netherlands | - | 27.25 TFlop/s | Mellanox now NVIDIA | #414 → |
| Stokes 2010 | Interconnect | Ireland | - | 36.56 TFlop/s | NVIDIA | #332 → |
| Hrothgar 2010 | Interconnect | USA | - | 68.27 TFlop/s | Mellanox now NVIDIA | #111 → |
| RICC 2009-08 | Interconnect | Japan | - | 97.94 TFlop/s | NVIDIA | #47 → |
| JSS FX1 2009-04 | Interconnect | Japan | - | 110.60 TFlop/s | NVIDIA | #60 → |
| Chinook 2009-03 | Interconnect | USA | - | 97.07 TFlop/s | NVIDIA | #21 → |
| Jeraisy BL460c Cluster 2009 | Interconnect | Saudi Arabia | - | 39.70 TFlop/s | NVIDIA | #99 → |
| Nagoya FX1 2009 | Interconnect | Japan | - | 28.51 TFlop/s | NVIDIA | #243 → |
| Diamond (ERDC) 2009 | Interconnect | USA | - | 160.20 TFlop/s | NVIDIA | #24 → |
| Hera (LLNL) 2009 | Interconnect | USA | - | 102.20 TFlop/s | Mellanox now NVIDIA | #52 → |
| Aitzaloa 2008-11 | Interconnect | Mexico | - | 18.48 TFlop/s | NVIDIA | #226 → |
| Roadrunner 2008-06 | Interconnect | USA | - | 1.03 PFlop/s | Mellanox now NVIDIA | #1 → |
| T2K Tsukuba 2008-06 | Interconnect | Japan | - | 77.28 TFlop/s | Mellanox now NVIDIA | #21 → |
| T2K Kyodai 2008-06 | Interconnect | Japan | - | 50.51 TFlop/s | Mellanox now NVIDIA | #35 → |
| Netuno 2008-05 | Interconnect | Brazil | - | 16.24 TFlop/s | NVIDIA | #139 → |
| Magic Cube 2008 | Interconnect | China | - | 180.60 TFlop/s | NVIDIA | #11 → |
| Chebyshev 2008 | Interconnect | Russia | - | 47.17 TFlop/s | NVIDIA | #37 → |
| Tachyon 2008 | Interconnect | South Korea | - | 16.99 TFlop/s | Mellanox now NVIDIA | #278 → |
| Queen Bee 2007 | Interconnect | USA | - | 34.78 TFlop/s | NVIDIA | #23 → |
| Calhoun (MSI) 2007 | Interconnect | USA | - | 17.31 TFlop/s | NVIDIA | #45 → |
| Bio-X2 2007 | Interconnect | USA | - | 15.57 TFlop/s | NVIDIA | #54 → |
| Kaon 2006 | Interconnect | USA | - | 6.89 TFlop/s | Mellanox now NVIDIA | #95 → |
“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.