There was some confusion initially about yesterday’s SGI Octane III announcement, and some news has come out that might help to answer some of the questions.
One big question was how you got 10 “trays” but 19 “nodes” when using the Xeon 3400 or Atom processors. Apparently the system uses only a single set of disk drives for booting, so most of the nodes operate “diskless”. In a single “tray” you can fit two separate nodes when using these smaller processors, which means 20 “nodes”. However, a node must be sacrificed in order to make space for the drives, which yields the final number of 19 nodes, 38 Atom cores. Also, the Xeon 3400 configuration is not currently available. Right now the Atom build is available, and the Xeon 3400 will be available “soon” although no firm deadline is given.
News about the SGI Octane III has gotten around:
- BoingBoing
- EnGadget
- Gizmodo
- and many others.
While those of us that work with such systems regularly already know, one common misconception going around is that the system is a massive 80-core 1TB monster computer, rather than the 10-node system it actually is. That means:
- It won’t run Photoshop any faster. As a single computer, it’s just a regular dual-processor/quad-core 96GB Ram machine.
- It won’t set any Crysis benchmark records. As a single computer with a video card, it’s a dual-processor/quad-core 144GB Machine with a regular NVidia card.
- You won’t be loading up any gigantic datasets on it with your usual stuff. You’ll need some that can properly utilize the distributed architecture like the new Fieldview, Paraview, or Visit.
- For 3D rendering, it might actually help. It won’t make a single render go much faster, but think of it as a “Render Farm at your Desk”, where each of the 10 nodes is a render node and you can see how it might help in certain circumstances.
So there you go.. Hopefully this will help answer a few of the questions I’ve seen floating around the internet. Have anything to add?
