SGI ICE XA is SGI’s flagship scale-out platform and when combining speed, scale, and efficiency, provides the most powerful distributed-memory supercomputers in the world.
insideHPC: Well, yeah. We’re almost done with this thing already, but I thought I’d stop by and ask you what’s going on with the SGI ICE X machines.
Mark Fernandez: Oh, very good question. The ICE X is in production now. It is a compute-focused platform. At this show, we’re introducing the ICE XA. It adds several degrees of flexibility. With the ICE XA, we’ll still maintain the compute-only-density that we have in our ICE. But we have another blade available for that, which will enable accelerators, both GPU and Xeon Phi cards. We have an even third blade, will allow people to add hard drives and other IO devices. So, the flexibility of the ICE XA is what we’re talking about today at the show.
insideHPC: Okay, great. Now the fact that you can integrate that into the same blade patches, is that all about density and all the other goodness that you were telling of ICE X in the first place?
Mark Fernandez: Well, yeah you got one of the three. So, with ICE, it’s an integrated cluster environment. We focus on density, power cooling, and performance. We’re adding flexibility to that list with these three types of blades we have now available.
insideHPC: Okay, so what’s the interconnect on the ICE X?
Mark Fernandez: Oh, very good. One of the knocks on the ICE X was, we required you to have SGI’s hypercube. With the ICE XA, we’re adding flexibility. You’ll be able to have all to all, the traditional factory, as well as our classic hybercube. Alright?
insideHPC: Okay, so a choice of topologies?
Mark Fernandez: A choice of topologies. A choice of I/O. A choice of blades.
insideHPC: All right. And then, what are we talking in terms of scale? I mean, does it start at one rack and go up to petascale or what is the goal?
Mark Fernandez: Well, ICE XA, as you can see from our booth is on our path to X’s -scale. We sort of couldn’t get there with ICE XA, but we think we can do that with the next generation – ICE EXA – and it’ll be out in about two years.
insideHPC: About two years? I’m trying to do the math.
Mark Fernandez: 2017 time frame.
insideHPC: Okay. Any big wins of note that you want to tell me about with ICE X, or is XA too new for that?
Mark Fernandez: Yeah. We’ve got a couple of new wins with ICE X. One’s that NASA got her and it’ll be our fifth petascale machine on the list. That’s new. It’s being installed this week. So I don’t know if it made the list or not, but it’s certainly one of ours. And we also have the Idaho National Labs which is on the top-500 and is the most power-efficient X86 box on the list.
insideHPC: Holy cow. Is that from the Green500 that you do that cross number? Is that how they figure that out?
Mark Fernandez: No, it’s one of the TOP500 numbers. They have a power listed there now, so it’s real simple. This is the power we use. This is the number we’ve got. There’s no trickery there
insideHPC: Okay, that’s right. No stupid benchmarking tricks. This is straight-up right?
Mark Fernandez: This is straight-up. [laughter].
insideHPC: All right. Well, this is exciting. The XA, you’re saying, is it shipping soon, or now, or when?
Mark Fernandez: We announced it this week. It will ship in the March time frame.
See our Full Coverage of SC14.