Parts/Accelerators/NVIDIA Tesla K20X
Accelerators · Kepler · 2012
NVIDIA Tesla K20X
6 GB GDDR5. One per node in the Cray XK7 design that made Titan the first GPU-led No. 1 in the US.
Why this part
Titan was an upgrade rather than a new machine: Oak Ridge kept the Jaguar cabinets and the Gemini fabric and added 18,688 K20X boards, which is why it appeared so quickly and so cheaply. The choice was contested at the time, because a machine that only performs when code has been rewritten in CUDA is a bet on the users doing the rewriting. Some did and some never did, and Blue Waters at NCSA was bought the other way round for exactly that reason.
Systems using this part
- Systems
- 11
- Rmax underneath
- 24.90 PFlop/s
- Units recorded
- 19,962
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| ILGTI Cray XC30 2015 | Accelerator | India | - | 558.80 TFlop/s | NVIDIA | #114 → |
| Mesabi 2015 | Accelerator | USA | 80 | 474.07 TFlop/s | NVIDIA | #139 → |
| HPC2 2014 | Accelerator | Italy | - | 3.19 PFlop/s | NVIDIA | #12 → |
| QB-2 2014 | Accelerator | USA | - | 1.05 PFlop/s | NVIDIA | #47 → |
| TSUBAME-KFC 2013-10 | Accelerator | Japan | 160 | 125.20 TFlop/s | NVIDIA | #1 → |
| Hydra (MPCDF) 2013-10 | Accelerator | Germany | 676 | 1.28 PFlop/s | NVIDIA | #28 → |
| Lichtenberg I 2013-06 | Accelerator | Germany | 86 | 213.78 TFlop/s | NVIDIA | #148 → |
| Blue Waters 2013-03 | Accelerator | USA | - | - | NVIDIA | - |
| Fionn 2013 | Accelerator | Ireland | - | 140.45 TFlop/s | NVIDIA | #359 → |
| Titan 2012-11 | Accelerator | USA | 18,688 | 17.59 PFlop/s | NVIDIA | #1 → |
| Todi 2011 | Accelerator | Switzerland | 272 | 273.70 TFlop/s | NVIDIA | #91 → |
“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.