Benchmarks are $%#&@!!
At SC11 I ran into Henry Newman, CEO of HPC consulting firm Instrumental Inc. After exchanging the usual pleasantries and deeply offensive personal insults, we got to talking about some of the recently released benchmark results – and how irrelevant most of them are to the real world.
Meet the boffin who gave GPUs a bigger bang for the Buck
SC2011 One of the presentations I caught at SC11 was by GPU computing pioneer Ian Buck - which is a good name for a pioneer, I think.
Dell cooks up new HPC strategy
HPC blog As I trudged toward a swanky hotel for a meeting with Dell, the Seattle sky was spitting cold rain like an old man realising the soup in his mouth is way too hot. (Adding more drama to these intros, nice, right?)
Letting GPUs run free
Blog One of the most interesting things I saw at SC11 was a joint Mellanox and University of Valencia demonstration of rCUDA over Infiniband. With rCUDA, applications can access a GPU (or multiple GPUs) on any other node in the cluster. It makes GPUs a sharable resource and is a big step towards making them as virtualisable (I don’t think that’s a word, but going to go with it anyway) as any other compute resource.
HP confident on HPC future
Blog A quick meeting with HP at SC11 confirmed that the company is feeling good about their HPC achievements and prospects for the future. HP is the second biggest HPC vendor on the most recent Top 500 list with 141 systems (28 per cent). However, they’re still behind market leader IBM, which has a 44 per cent share with 223 total systems.
Free cluster for a good cause
Silicon Mechanics, a mid-sized manufacturer of rackmount servers, clusters, and storage arrays, is celebrating its 10th birthday by giving away a lot of stuff. First, the company did a great job sponsoring Boston University in this month's SC11 Student Cluster Competition in Seattle (SCC11).
Fueled by Silicon Mechanics gear, the team snared fourth place overall – quite an achievement given the caliber and experience of the competitors.
Team Boston, aided by Art Mann of Silicon Mechanics, had one of the largest and most powerful configurations - a hybrid system with 11 dual-socket nodes containing 336 AMD Interlagos cores, 352 GB of main memory, and four of the latest NVIDIA Tesla C2090 GPU accelerators.
But Silicon Mechanics isn’t done yet: the firm invites researchers to apply for a grant to receive the former ‘Team Boston’ cluster. The grant is open to any U.S. or Canadian post-secondary educational institution (or research affiliates); non-profit research labs; or researchers at government labs.
Three labs, one (big) cluster
We spent a few minutes at an Intel SC11 hospitality event in Seattle talking with Matt Leininger, Lawrence Livermore National Lab Deputy of Advanced Technology Projects, about the U.S. DOE National Nuclear Security Administration’s Tri-Lab procurement.
Gordon the supercomputer is intense about data
SC11 According to San Diego Supercomputing Center chief Mike Norman, his brainchild 'Gordon' is the world’s first data intensive supercomputer.
Titan = supersized science
SC11 We spent some time at the Oak Ridge booth at SC11 in Seattle talking with Jack Wells. He’s looking forward to the newest addition to the Oak Ridge supercomputer family, the Titan.
Cray takes Oak Ridge super to Titan
Peg Williams, Cray SVP of High Performance Computing, is working on the biggest computer upgrade in history – transforming the 200-cabinet, 42,000-processor Oak Ridge Jaguar supercomputer into its new Titan form.
When it’s finished, Titan will have 21,000 16-core AMD Interlagos processors and a roughly equal number of NVIDIA Kepler GPU accelerators. This will boost Jaguar’s 2.33 petaflop performance to an estimated Titan-scale 20 pflops. But a project this large with parts that are just coming onto the market can’t be done all at once.
The existing Jaguar can’t be down for long periods; that would bring a halt to a lot of science stuff. So the upgrade needs to be orchestrated in a way that minimizes both downtime and performance degradation.
However, this hybrid CPU-GPU system is a new animal for Oak Ridge and they’ll need some time to get to know it, experiment with it, and make sure they’re ready to make the shift. As you’ll see in the video, Peg has it covered with a plan that addresses these important concerns. She talks about the plan of attack for the upgrade and the timeline for turning Jaguar into Titan. ®
Jaguar to Titan? Not so bad…
At SC11 I had the opportunity to talk to some of the people responsible for the biggest computer upgrade known to man. Oak Ridge National Labs is upgrading its current Cray XT5 ‘Jaguar’ system to a Cray XT6 system that will be known as ‘Titan’.
Nvidia's Maximus: Hard-core 3D graphics on speed
Blog Part of Nvidia founder/CEO Jen-Hsun Huang's SC11 keynote session was a demo of something called Maximus that elicited oohs and ahhs from some of the folks around me.
Petaflop = Top500 table stakes by 2016
InsideHPC's Rich Brueckner and I sat down yesterday with Jack Dongarra and Hans Meuer from Top500.org, the guys who put together the twice yearly list of the 500 largest supercomputers.
ARM+GPU = dream super?
SC11 We've seen a lot of ARM server activity in recent weeks, with ARM chip upstart Calxeda announcing its 5-watt EnergyCore ARM chip (breakdown here) and a partnership with x86 giant HP (outlined here). Also the ARM architecture is getting a necessary but painfully slow 64-bit makeover, slated to roll out in 2014.
Would you spend $300m to save 6ms?
Blog How much is getting trades done 6 milliseconds quicker worth? About $300m, as it turns out. A new transatlantic fibre cable promises to reduce the languid 65-milliseconds it take to transmit a trade between London and New York time to a quicker 59ms. Traders are expected to line up to use the new service and pay extra for the privilege.
HPC 2.0: The Monster Mash-up
Pt 1. Blog IBM recently invited a handful of really smart HPC-centric industry analysts (and me too, for no apparent reason) to spend the day talking about where the market is going and how IBM intends to address it.
It was truly a conversation, rather than the typical vendor PowerPoint-palooza where they simply run through every product slide deck they can get their hands on.
Watching hurricanes
America’s National Oceanic and Atmospheric Administration (NOAA) put the cart before the horse to some degree when purchasing a new supercomputer to track hurricanes.
Titan Unveiled
The “Fastest Supercomputer” title may move 6,940 miles (11,167km) eastward in 2012 from Kobe, Japan to a small Tennessee town.
That’s if the folks at the Oak Ridge National Labs (ORNL), along with Cray and AMD, can pull off a massive upgrade of the existing Jaguar system. They’ll replace existing nodes with the new Cray XK6 nodes, which will run the new 16-core AMD Interlagos chips.
In the second half of 2012, these will be augmented by dual NVIDIA Kepler GPUs. Interlagos CPUs should be around 3x faster than their current hex core processors and the addition of Kepler GPUs (and lots of ‘em) will really crank up the performance potential.
Overall, they’re expecting a 9x increase in speed, which will put the system in the 10-20 PF range when it’s completed near the end of 2012.
This new system will carry the moniker Titan, which could allude to the largest moon of Saturn or to the Titans, who in Greek mythology were a race of gods who appear in a lot of old poems and probably dismembered and ate Dionysus. Either way, the system will be a screamer.
Hacking Baseball
Baseball is perhaps the most boring thing in the world to watch. The leisurely rate of play, the lack of constant action, and the pauses players take for impromptu meetings, spitting, and crotch-grabbing are torture for my ADD-riddled brain.
Reading about baseball is every bit as bad, and reading about baseball-stats geeks who painstakingly ‘score’ every move on the field makes me want to beat myself with a bat. On the other hand, I’m a big fan of money, and innovative ways to make more of it.
I found a very interesting article in my ever-growing pile of Businessweek magazines about how automation and deep analytics are playing an increasingly large role in the game.
The Baseball: Running the New Numbers story outlines, in highly readable form, how Major League Baseball, individual teams, and savvy techies are building out systems that log pretty much everything that happens on a baseball field.
I am a baseball hacker
The goal is to rate players on what they bring to the table. A guy who can move quickly in the right direction and make a difficult play is much more valuable than a slackjaw who happens to be standing in the right place at the right time. Traditional stats don’t pick up the difference between those guys, but the new systems will.
Firms and teams use humans to review each and every play and capture more of this differential data, but it’s much more subjective, and there are limits to how much detail they provide.
Feats of clay?
This is where new technology enters the picture. Get enough cameras looking at the field and you can log the speed and angle of every hit ball, and the position of the fielder in relation to it. That way you can judge whether a fielder is a gifted athlete or just lucky.
A typical game will generate around 2.5 million results or records, totaling 2TB – which seems a bit large until you factor in all the standing around, spitting, scratching, etc. The amount of data that teams will be interested in is probably closer to 750GB or so per game.
Businessweek also raises some interesting points about the use of this data. Right now, a lot of it is provided freely to teams and the general public. The league is mostly in favor of this, because it gets fans more involved with the sport.
But some teams aren’t so wild about this openness - they see this data as the raw material they use to build their competitive advantage. Still others figure that the data will get out there anyway, so why not make it available to all?
There are many more details that I’m glossing over – it’s baseball, for God’s sake, so I can’t be expected to pore over every word – so it’s definitely worth your time to read the article if you’re interested in how these systems work. ®
HPC storage purchasing exposed!
In a recent HPCwire report, Nicole Hemsoth discloses the back story behind a major storage purchase by Utah’s Center for HPC (CHPC).
HP opens Big Data front with Oracle
With all of the attention focused on the war raging between Oracle and Hewlett-Packard, a significant HP announcement in late June seemed to slip under the radar of the industry press.
King K super: does it refute hybrid HPC model?
ISC'11 It's been an eventful International Supercomputing Conference (ISC'11) in Hamburg. The Japanese sprang their K Computer on an unsuspecting HPC world, throwing down 8.126 Pflops on the table and raising the high-water performance mark by a factor of three.
Top500 founders talk big
As many of you know by now, there’s a new computer at the tippy top of the just-published Top500 list. I found out a bit earlier than most, via a webcast I was recording at 3:00 a.m. my local time on Sunday.
It was a bit of a struggle to record it; their net connection kept breaking, and the sound quality on my side was like a Dixie cup and string, but I heard enough to know it was big news.
Here’s the quick overview: The Japanese have taken over the top spot on the Top500 list with a SPARC-fueled ‘K Computer.’ It’s 8.16 Pflops – 3x faster than the former number one system (Tianhe-1A).
In fact, the K Computer is larger than the next five systems on the list combined. It sports 8-core SPARC processors – 68,544 of them, to be exact – for a total of 548,353 cores. Timothy Prickett Morgan has written an exhaustive (and excellent) account of the system here.
It’s also highly interesting that this is an old-school supercomputer. It has a sophisticated, proprietary 6D torus interconnect and doesn’t use GPUs or other accelerators to juice performance.
The lack of hardware accelerators is unusual these days and noteworthy. This is a very efficient system as well, delivering 93 per cent of its theoretical peak performance – a number that puts it at the head of the class. The K pulls a fair amount of power: 9.89 megawatts. (Damn!)
That’s a bit more than double the 4.04 megawatts that Tianhe draws, but K yields 3x more performance… so it’s kind of a bargain, right? This system was a surprise in a lot of ways. No one expected it this soon, or expected it to triple the performance record.
Oracle vs HP+Intel: Wassup?
There’s nothing like a vendor war to keep things interesting, and Oracle has launched a great one with its recent decision to stop development for future Itanium processors.
Nvidia flexes Tesla muscles
2010 was the breakout year for Nvidia’s Tesla division, according to Tesla VP Andy Keane, who spoke at the company’s Industry Analyst Day earlier this month. I think it’s pretty obvious that he’s right, and a quick review of the last year tells the story.
The curious incident of Oracle and HP-UX on Itanium
Comment When I saw the news on Wednesday morning, I thought I had picked the wrong week to give up sniffing glue. Or maybe Oracle did. Either way, Oracle’s announcement that it is halting development on HP-UX/Itanium versions of its products touched off a firestorm of phone calls, emails and tweets that just won’t let up.
Nvidia rushes to ARMs
Nvidia’s Analyst Day at its Santa Clara, CA headquarters on March 8 gave me new insight into the company and how it see the markets.
'R' is for Revolution Analytics
Blog Recently I met the analytics firebrands Revolution Analytics at their Bay Area offices.
Will world universities step to cluster challenge?
The April 15th team- application deadline for the SC11 Student Cluster Competition (SCC) is fast approaching. The SCC pits eight university teams from around the globe against one another in Seattle (site of Supercomputing 2011 - or SC11) to compete for clustering glory.
Honey I shrunk the chip ... now what?
Bigger is better in pastries, paychecks and bank accounts, but not in electronics. A recent story in HPCwire caught my interest and got me thinking about what the end of the shrink road might portend – and the potential alternatives.
Can HPC methods yield 92% help desk satisfaction?
You know how when you call a help desk or customer service number they have that short ‘this call may be recorded’ message? Have you ever wondered what they do with the recordings?
Easier to secure the cloud than your data center - IBMer
Security is probably the biggest factor keeping enterprises from moving more applications and data to public clouds. I argue that security is just one (albeit a hugely important one) of the reasons why public clouds will exist as a tool for data centers – rather than the default usage model – for the foreseeable future.
SeaMicro: Intel proxy shows server moxie
SeaMicro last week announced its newest ultra-dense, ultra-low power Atom-based server solution. Our pal TPM gives it full coverage in this Reg article, but here’s the basic story.
DARPA to pay hackers and hobbyists for help
Internet grandfather DARPA (US-based Defense Advanced Research Project Agency) is throwing open its doors and wallet in an attempt to solicit unconventional security solutions.
It is reaching out past the usual huge, multinational technology consulting firms and egghead think tanks and trying to engage with small start-ups – and even hackers – to get some new ideas into the mix. Looks like they’re willing to throw some cash at them, too.
So far, so good – and a reasonable strategy to pursue. What brought me up short was a brief mention of their asking ‘hobbyists’ for input … do they have any idea what they’re getting themselves into?
Flickr flap illuminates cloud concerns
Briefly, a photographer used Flickr to store his pictures online – amassing more than 4,000 photos over five years. Being the helpful sort, he alerted Flickr to another user who was stealing content. Flickr deleted his account, by mistake and those pictures looked to be gone forever. Following a media firestorm, Flickr had a quick rummage and recovered his images.
X-ISS builds turnkey HPC for the masses
X-ISS is an interesting company. Based out of Houston, it's been around for almost 20 years, offering a full slate of HPC systems and services.
Supercomputer performance leaps off the charts
Webcast In our first HPC channel webcast of 2011, we talk to Rich Brueckner of InsideHPC to take the measure of current HPC performance and talk about the factors that got us to this point.
Overclocking for servers
SC10 Bring up the topic of overclocking to the major system manufacturers and you hear a variety of responses ranging from ‘we’re plenty fast enough now – take a look at our _____ (insert benchmark here) that’s enabled by our use of ___________(insert techno feature here)” to “it’s just not safe, son, and our customers need to be protected from this kind of thing.”
Appro talks super-dense servers
At this point, every server vendor+dog has some sort of integrated CPU-GPU system that they’re offering to their high performance computing (HPC) and other number-crunching customers.
Student Cluster Competition: And they're off!
SC10 Here I am in rainy New Orleans covering the Student Cluster Competition at SC10.
It's Tier 1 vs Tier 2 in the HPC slugfest
SC10 When I first started attending industry trade shows, I was always surprised by the range and number of smallish players competing head-on against much larger industry leading companies. (Incidentally, at the first trade show I attended, the wheel edged out fire as the most disruptive technology on the floor.)
HPC clustering: A new spectator sport in Lone Star state?
SC10 Texas is at SCC with the most elaborate booth of the competition - certainly in terms of artwork and theme.
Blogs and Opinion
ARM+GPU = dream super?
SC11 NVIDIA: 'Yup, we're on it'
Nvidia's Maximus: Hard-core 3D graphics on speed
Blog Quadro-Tesla mashup for animation acceleration
