In this podcast, the Radio Free HPC team wraps up the SC14 conference. It was a big week for all of us, so in the interest of time we each picked one big thing to talk about:
- Dan was excited to announce that the University of Texas at Austin won the overall Student Cluster Competition for the third time. They are the first three-peat team in the history of the competition. Being, Dan, he couldn’t stop at one thing, so he babbled on about how synchronicity of showfloor bumpings-into might someday result in a Petaflop supercomputer the size of a dishwasher.
- Henry’s big thing was the Micron’s Automata processor for Big Data. We all agreed on that one. Here is Rich’s interview with Paul Dlugosch from Micron, who does a great job of explaining why this thing is so revolutionary.
- Finally, Rich is still recovering from all the video interviews he did on the show floor. His one big thing turned out to be about next year at SC15 in Austin, which has less exhibitor space than this year in New Orleans. Faced with this dilemma, the SC conference committee announced that vendors would not have the opportunity to choose their own booth space for next year. Instead, the committee will do it for them. Rich is crying foul and he is not the only one.
Rich Brueckner: Hello again, and welcome to Radio Free HPC. This is where we talk about super computing, high performance computing, and other technology topics. I’m your toastmaster, Rich Brueckner, from InsideHPC, with my co-host, Dan Olds from Gabriel Consulting, and Henry Newman from Instrumental. Now, let’s get to the show.
All right, we’re back fresh from SC14 in New Orleans. And this is Radio Free HPC, and I’m Rich Brueckner, your toastmaster. And with me as always, is Dan Olds and Henry Newman. Gentleman, how are you today?
Henry Newman: I am wonderful, Rich. All recovered from all the good food in New Orleans. I’m still recovering. I’m glad I didn’t get my physical this morning. Think my cholesterol would still be up [chuckles].
Dan Olds: Yeah, right. I like to keep my cholesterol fully optimized, right in that line at the top. I’m okay, too. Just an exploding head from trying to figure out everything that I saw and go back and start writing about it.
Rich Brueckner: Okay, okay. Well, what I thought we’d do for this show guys, because reminiscing about a week of major activity is tough. But pick one thing that sticks out, that you’d like to share with the audience. And Dan, shall we start with you?
Dan Olds: Let’s see. There is so much and I’m going to be covering an awful lot of it. First of all, the student cluster stuff. We did have, for the first time ever, in student cluster competition history a three-peat where the University of Texas, Team Longhorn, took a traditional cluster, with a whole bunch of just CPUs and cores, and won the overall championship, unprecedented.
Rich Brueckner: You’re talking about the Austin team, right?
Dan Olds: The Austin team, three times in a row. They are three-peat winners .
Rich Brueckner: If I recall, you predicted that would be the outcome on our earlier show that weekend. You said they had a good shot at it.
Dan Olds: Yeah, yeah, they were the first three-peat ever, in any competition. Pretty cool.
Rich Brueckner: Now, what would you attribute that to, besides, “Keep Austin weird”? They have good coaches, I would assume.
Dan Olds: Well, they’ve got the TACC boys behind them, who are taking a great interest in it, and that’s great for the kids.
Rich Brueckner: Well, it’s got to be something. It’s not just an anomaly when it’s a three…
Dan Olds: No, and I’ve got to go through and analyze all the detailed results. I think what Texas did right is that, they took a very careful look at those applications and didn’t see much in the way of benefit from using GPUs. So, they went whole hog with CPUs. And it looks like the dice rolled right with them. Although, again, I haven’t had a chance to really analyze the results, which I will do, but I think it was pretty close.
Henry Newman: Well, and Dan, someone mentioned something about coaching. And they do have a good coach. Sean Keel is one of the smart guys in this industry, and he should be given some credit here.
Dan Olds: No, absolutely it’s– yeah. Keel really does a great job with that team, and the rest of the guys that he has working with them. Another team – and I can let this out, since I did tweet it publicly – that went from worst to second is, the University of Tennessee Knoxville kids, who had some pretty good coaching on their own. And they have pretty good relationship with Oakridge. And they made a big move in the rankings. So, it was a great competition.
But one of the things that I’d like to talk about is, sort of the thing I love about SC, is you get a chance for these sort of random encounters that might turn into something pretty cool. And the one that’s top of mind with me, was visiting with– excuse me, not Fujitsu, with 3M, and taking a look at their liquid cooled stuff, which I’ve blogged about and done some videos on before. But I wanted to check them out again this year and see what was new. And I’m over there talking with them about just the incredible density you can get out of using their particular fluid. It has an extremely low boiling point, it’s not mineral oil, it doesn’t have any sort of harmful property. There’s no smell to it or anything like that.
So, I said, “Have you ever talked to the guys at a company called One Stop? Because they make these peripheral PCI connect things that can put a whole lot of GPUs in a very small space.” They said, “Well, no.” And I happened to look to my left, and noticed that One Stop was in the booth right next door. Went, “Oh, okay.” Drag them over, and get to talking with the guys there, and take a look at a particular enclosure they have that’s designed for SSDs, to be able to put I believe at 52 SSDs in essentially one 2U rack them out enclosure.
Saying, “If we took the heat sink fans off of a GPU, you could probably slide 52 GPUs into this thing.” And we were thinking about that and doing some math and all of that kind of cool stuff. Then Ian Buck from Nvidia wanders by. And I reach out and snag him and say, “Hey, just exactly the right guy to bring [chuckles] in on this.”
And we start talking, and coming up with the thought that you might be able to put a petabyte worth of compute in something the size of a dishwasher – utilizing the enclosures from One Stop, GPUs from Nvidia, and the miracle cooling fluid control of 3M. Might be something we’ll see on a show floor at some point.
Rich Brueckner: Okay, okay. Well, good, Dan. I know that competition was a big deal, and it’s not even fair to ask you for one thing, but we got to give Henry a shot now. Henry? I mean, you were like a rock star. You’re on stage stirring up all kinds of stuff. I just put that video up, by the way.
Dan Olds: I watched that. The point was that tape is dead, right?.
Henry Newman: Tape is dead, yeah. But I want to talk about something different. I want to talk about Micron’s Automata.
And what I think is the most interesting thing of it is – this was actually an idea that Seymour Cray had back in the ’80s. Actually while Dan was talking, because it’s oftentimes it’s boring, I just did some searching on it. There were a couple ACM articles about processors and memory. But I think the time– I think Seymour was way ahead–
Rich Brueckner: Hang on, hang on. Let’s describe what this thing is, for the people who don’t have the benefit.
Henry Newman: Well, Dan? It’s basically ability to do processor in memory and manipulate data in the memory before it hits the processor.
Rich Brueckner: Well, the form factor is it’s a PCIe card. It does not boot Linux. It does basically a memory array from what I can tell, right?
Henry Newman: It has processing in it. Has millions of elements in there, and it doesn’t run in OS, is the point. Doesn’t run programs. It’s not a substitute for memory.
But the cool thing about it, and I have more material. I shot a video; I got some video notes. I’m going to talk with a guy from Georgia Tech, who did some really good testing on this. He was running – I’ll have more details later, so this is going to sound a little vague. But he was doing genome sequencing, and ran a particular problem through this thing, that it finished in like 51 seconds. And to do the whole problem, they couldn’t accomplish it at all with a 48 node cluster.
Well, here’s the thing. If you’re doing bit manipulation, this thing will smoke anything else that’s out there.
Rich Brueckner: Okay so, Henry, what kind of applications are we talking about here?
Henry Newman: Well, one of the things I use bit manipulation is for searching large things for – in this case, I was doing file system search algorithms. And those kinds of genome sequencing, anything with repetitive data, visualization…
Dan Olds: A matching problem?
Henry Newman: A matching problem. Like a search algorithm for maybe Hadoop. I mean, there’s all kinds of things you could hash and turn into bits or…
Dan Olds: Traveling salesman-type problems, routing, that sort of thing, or no? Not a good fit?
Henry Newman: I don’t think router programming would fit, but maybe, I have thought about that.
Dan Olds: Okay, but there’s a niche; there’s a big niche.
Henry Newman: I think the difference when Seymour did the machine and today is we didn’t have the requirement and the knowledge base for genomic work, for other kinds of searching that we do today in the general-purpose world. And I think that this is a technology whose time has definitely come, and I hope it really takes off, because I think it is really, really interesting stuff.
And so, a potential game-changer when it comes to some of these big data things. And I can see a lot of potential uses. The cool thing about this is, it’s not like what Seymour Cray did in building an entire machine like it. This is an add-in card. This is a peripheral, to be exact.
Well, it’s not only a card. Well I mean, it’s eventually going to be on a DIMM, potentially.
Potentially, sure, but it’s not like you have to hand-craft an entire machine to do this.
Like back in the day. But I think, going back to what I’ve said occasionally is there’s no new engineering problems, just new engineers solving old problems.
The difference at this point is, that there is a– the market has caught up with the technology.
Well, the demand, the need; the market needs have caught up with the technology. Well, it’s more than just the market needs. There were no algorithms that did those kinds of things, that could utilize it that were going to be financially profitable, in Seymour’s day. And now there are things that are potentially financially profitable. That’s what drives the industry is the ability to make money, and I encourage people to start thinking about these problems in different ways.
Dan Olds: Okay, okay. Well, what about yours, Rich?
Rich Brueckner: That’s right, I was supposed to come up with one, right? I ran around all that show floor, and I think I shot 50 interviews or some ridiculous thing.
Dan Olds: I ended up transferring over 100 gig of [laughter] data.
Rich Brueckner: Yeah, my laptop it too full to even hold it all.
Dan Olds: Well, actually one thing that bothered me, there’s a vendor there that had a guy doing a– one of these talking heads, trying to gather, like the carnival barker. And I’m not going to mention the vendor, but he’s saying, “And we’re going to give you a 400 gig USB stick.” Which, I’m like, “Well, wait a minute.” I did not think that existed, but the vendor that this guy was speaking for was one that could have crafted something like that as a giveaway.
But wait a second, right now I think the only vendor doing 3D NAND is Samsung. Is that correct?
Henry Newman: Potentially, you’re right.
Dan Olds: But they’re talk about it, it’s the USB thing. He might have said– he might have backed it down, said, “Oh, I think it’s 40 gig or something like that.” I’m going, “Oh, well wait a minute. I need a big thumb drive because of all this data.” So, I hung around, and I made eye contact and I talked to the women that were doing the badge scanning, and I got my hands on one of those thumb drives.
Rich Brueckner: Yeah, man?
Dan Olds: Four gig.
Henry Newman: Not worth the time. So, supposedly, the marketing people missed a couple orders of magnitude when they were trained at their training session?
Dan Olds: Looks that way, or else this guy was freelancing. He didn’t work for them. He was a hired gun. He was a glib talking head. So, anyway, that’s my bit of rage.
Yeah, whatever. All right. Richie!
Rich Brueckner: Okay. I forgot what I was going to even talk about, other than I ran around and shot videos and it was amazing. But one of the things was the Automata processor. So, when we do publish this show, I’ll attach that video. I just got to dig it up out of my many, many gigabytes of stuff.
Anyway, I wanted to talk about next year actually, a little bit.
Dan Olds: What about next year? SC15 is in Austin, yeah?
Rich Brueckner: Yeah. And there’s a bit of a controversy.
Dan Olds: Uh oh.
Rich Brueckner: And I just want to bring this up and get your thoughts here guys, because there’s less floor space for the exhibits next year than we had in New Orleans.
Dan Olds: I’ve heard that from many people, Rich, and there’re a lot of people unhappy about that.
Rich Brueckner: Well, they knew that going in. There’s no mystery here, so I don’t get it. Anyway, this conference has outgrown Austin and it will not return after next year, which is kind of sad. We’ve just gotten too big for our britches.
But here’s the deal, here’s the controversy part. Because there was less space, the conference committee at the Booth Selection meeting informed everyone that you vendors will not be picking where your booth is located. Instead, “We will do it for you. And you will take what you get, and shut the bleep up.” Yeah. Yeah. It was not a discussion. It was a declaration.
And I think it’s abhorrent. I’m sorry.
You know, I ran Sun’s booth for ten years, and that was my job to do that. That is very critical to my marketing efforts. And if you’re going to tell me you know what’s right for me, why my company is spending at least $200,000 with you people, that’s adds up in the millions over time, and they’re not treating them like customers. I think they’re being complete bleepers. Yeah. Just let the beep go off.
Dan Olds: How is that?
Rich Brueckner: That’s where I’m at. Right? It ain’t my problem, but I feel their pain.
Dan Olds: How did that go over with the crowd?
Rich Brueckner: You know what, I think Jill King from Adaptive is the only one that raised her hand and said, “Are you kidding me?” The rest of them, I think were in shock.
All I’m saying is, ACM and IEEE are the sponsors, right? And for all practical purposes, they’re the producers of the movie. Okay? They write the checks. So, I should really be putting this on them. But I wasn’t part of the meeting, so I just think it’s really a tragedy. And whatever.
Dan Olds: The thing is though, is it is somewhat self-correcting, because when these people, when these companies see their booth that they’re allotted, they have the choice either we’re going to be there or we’re not, and if enough of them– or we don’t want to pay this amount for that much. And if enough of them make that choice and decide to go with some of the conference facilities at the hotels nearby, or something like that, that can make a change.
Editor’s note: we have since learned that vendors have to sign a contract to hold the space, not knowing where their booth will be located.
Rich Brueckner: Well, the problem is that if you are in this industry, you must be at this show, or no one’s going to take you seriously. I don’t care if you are SGI, IBM, or a start-up like Avago, whatever, right? You got to be there, or you’re not in the game. And they know that, and I think they’re taking advantage. So, that’s my opinion.
Dan Olds: Well, what bothers me about this is that my master plan had finally come together where I was going to get one of the ten-by-ten media booths that they give out for free and then put a little– I found the perfect ten-by-ten travel trailer. It’s actually nine and a-half by nine and a-half, that I could move into that booth and live in it over the several days of SC and be able to do everything from there. But it sounds like that’s not going to happen.
Rich Brueckner: Doesn’t sound like it. So, we’ll see, maybe there will be a collective of people who are willing to rise up.
Henry Newman: I don’t– Rich, Rich. I don’t think it’s possible. It’s a done deal.
Rich Brueckner: Get over it, I guess, is the message.
Henry Newman: And it’s too late to disagree. If you people had known this beforehand, then maybe they would. But right now, it’s just too late.
Rich Brueckner: Henry, you’re an exhibitor. You’re part of this.
Henry Newman: “It’s too late, baby. It’s too late.” Remember the Carole King song from the ’70s?
Dan Olds: Nope. Things change. These guys will change.
Rich Brueckner: The rebel alliance will rise.
Dan Olds: It’s happened before. There were several times when like the cluster competition was cancelled, but then it’s come back. Things can change.
Rich Brueckner: Right. That is true. There is precedent there, because they did try to kill that, because of political nonsense. But you guys stood your ground and you made it stay.
Dan Olds: Well that, and the community rose up.
Rich Brueckner: Right. Well, I’ll tell you what though, when they killed the Storage Challenge, nobody raised enough of a stink and it died. It’s still dead, as they used to say on Monty Python.
Dan Olds: Because no one cares about storage, it’s boring. Sorry, Henry. There we go. You care.
Rich Brueckner: Well, guys, we’ve got to wrap this. I just have to end the show now.
Dan Olds: We’re going to be talking about other topics as we go on. There’s a lot more to chew over on this. But let’s take off now, and I got a lot of crap to– well, no, I’ve got a lot of high quality very good content to generate. And Rich has got crap, and Henry’s got crap that they’re going to generate.
Rich Brueckner: Yeah. It’s going to be fun to watch as this stuff trickles out over– and most of my stuff should be out after Thanksgiving that week. All those videos and stuff, and I’m looking forward to seeing what you did, Dan, as well.
Dan Olds: Okay.
Rich Brueckner: And Henry’s always full of surprises, so yeah.
Dan Olds: Full of something. Usually, mercury and cholesterol (from sushi) [laughter].
Henry Newman: Well that, Dan, actually, I’ll find my mercury tomorrow [chuckles].
Dan Olds: Excellent. On that note, good bye and we will talk to you again on the next Radio Free HPC.
Rich Brueckner: We’ll see you.