Cray, Intel, and Microsoft birth baby supercomputer Supercomputer maker Cray today announced a new desk-side, low-end, bladed office supercomputer in conjunction with chip partner Intel and software partner Microsoft. The new CX1 supercomputer is the first product to come to market after Cray tapped Intel as its future strategic chip supplier, dissing long-time chip partner …
...lots of Vista jokes. Or not.
Anyway, I thought users of supercomputing apps usually had a clue about computers and technical stuff like that. In other words, not your typical Windows users. Using graphical stuff just to "make it easier" sounds kind of paradoxical here... Sounds like a cool piece of kit (fans whirring like crazy and all that) though.
:: Anyway, I thought users of supercomputing apps usually had a clue about computers ::
I think that's the whole point - Cray et al are aiming the "personal super-computer" at people who *don't* necessarily have a strong technical background.
Like CG artists, climate scientists, mechanical engineers or whoever has big computing jobs, and usually relies on someone else to feed these jobs to the central supercomputer. These people potentially have a need for something like a CX1, but are not computer geeks like you and I. This is a largely untapped market.
I've spent a bit of time at ORNL, and their HPC facilities, and I can tell you that the biggest complaint among users/would-be-users is that HPC applications all use some weird screwed up system that takes too long to learn or requires hiring some high-strung super geek to deal with. HPC users will be able to outsource non-sensitive projects and have an almost unlimited pool of trained users.
If Windows can get into that market they'll displace the weird stuff nearly overnight.
I would have thought that the effort of tuning hpc machines and libraries to run most efficiently thereon would just dwarf the effort of learning linux, and could probably be hidden quite well by a few tools and scripts. People want familiarity in their software, and for such people that software would probably comprise the language (fortran, sigh...), the compiler and the libraries (linpack/lapack??) and specialised apps.
I can't make sense of this new tack of MS'.
... [click here]
No pain, no gain, no brain...
Windows for W*nk*Rs
MCSEies to the fore!
If only Switzerland would sue the M$ for infringing on the "How many holes can a cheese possibly have?" concept.
- Paris, who naturally trumps Swiss Cheese.
We all giggled a bit at the supercomputing conference I recently attended when Microsoft got up and talked about their grand supercomputing plans. The silliest part was that they still expected to charge a couple hundred dollars a processor, in this day of multi-million core computers. Their idea of a big cluster was 200 processors... I think they should stick to fail-over clusters of 3-5 servers and leave HPC to the rest of us.
To me, easy is a box with a ton of cores in it so I can just fork or thread. If I've got to go out across an interconnect (10GigE, InfiniBand, whatever), or in other words go to the trouble to code for MPI or something similar, I don't know that this is really gaining me that much...
Cray should speak to SGI about how successful an anti-Linux/Unix strategy in partnership with Microsoft can be! Note to cray: an inability to compete with the likes of HP and IBM in the HPC space will NOT be resolved by offering sub-par products with Microsoft. You need to get your asses in gear and start offering Linux based products that people want to buy for more than just your brand name.
Yeh, right, that many high performance cores in the office behind the desk. How many amps is your local office wall socket rated for again?? I can see health and safety guys getting all over this as the boffins in the corner crank up the AC and start fires. The IT security guys will have a fit too.
We've spent too long getting IT away from the office floor and into the Data Center where it belongs. If this catches on a await the inevitable headlines once one is melted/hacked/stolen/flooded with a cuppa!
Leaving aside the contradiction in terms of a "desk-side supercomputer", Cray is trying to do the right thing for the right reasons with the Cray CX1. The world can only be a better place if more people are conversant with what HPC can help them do. Making it easier for scientists and professionals to use HPC is the real HPC challenge of this decade - that's why Microsoft's initiatives are so important. Cray's challenge is making a business model that works for it as well as for its new friends. Cray has a lot of experience in what not to do in entry-level HPC - it's been totally unsuccessful in the shark pool with Dell, HP, etc with entry-level products, most recently the XD1. And whilst struggling in the shark pool Cray needs to make sure that the CX1doesn't become another distraction from its core mission which is of course making real supercomputers.
Windows "mini-super" computers on the tarwub soon? I see a whole lot of petaflops busy sending spam...
Scientists are not gonna buy into that (for the ones who need that kind of computer power also know how to get it -and know better than wasting all theses precious CPU cycles on Windows). They might buy the barebone version though. From what I read, it looks like pretty good value for money. For the price of a workstation, you get a bit less, but a bit more (a bit less in the software department, but a bit more in the hardware department... sounds like a very, very good deal when you just want to crunch numbers).
My opinion, though, is that the real core target for the fully loaded thing is CGI fiddlers, designers, and, of course, yoof-targetting startups with more money than brain. In short, people who will think they can now get a (semi) super-computer all by themselves, no scary IT geek needed (lab geeks -sorry, scientists- are not afraid of IT geeks. They've even been able to breed in some occasions. Reportedly). And my bet is that they're in for a nasty surprise (and, incidentally, so are we, I for one welcome our super-botnet overlords).
I can rant all I want, I know that I'll happily get me a fully functional one. 5 years from now. In a dumpster. Happy in advance.
Supercomputers were in the 80's, the 90's saw it become High Performance Computing, the 00's and it became Advanced Computing. One might suggest that this change came roughly with the change from single processor huge vector machines (proper supercomputers, immersed in liquid Freon to keep them cool) to big parallel machines that included massively parallel designs and machines built from commodity processors, but custom everything else (now HPC), and thence to Beowulf clusters, and slowly to machines that are almost totally COTS components (Merley Advanced).
OK, Windows on a a supercomputer.
So where are the apps? It is all well and good to suggest the technically illiterate users will prefer Windows, but high performance applications are a totally different animal to programming for the desktop. Trivial task farming applications will be fine - renderers, sequencing (i.e. BLAST), and the like. After that it gets grim quickly. Software vendors are starting to come to grips with SMP and multi-threaded programming. But distributed memory over MPI is a very different breast. It is hard, and takes a lot of deep knowledge of the possible algorithms, and usually needs tuning to the specifics of the hardware. Speed in MPI systems often requires careful overlap of communication with computation, and this requires knowledge of bandwidth, latency, computation speed, cache effects and so on. Whist there are useful libraries for well understood numeric functions, the idea that you can just link against these and have it all work wonderfully is naive. Any fool can write a parallel version of a program that runs slower than a mono-processor version. Getting even a reasonable fraction of the potential speedup on a distributed system takes real hard slog. We will probably see some of the mechanical CAD apps slowly appear - turnkey FEA, CFD, and the like. But to imagine that any real HPC codes will appear is a different matter.
These little Cray boxes are not supercomputers. Not any more. Supercomputers sit in the Top500. The smallest machine that would have got you into the latest Top 500 has 1500 cores. You will need well over 2000 cores in six months. A noddy little box with a few blades on Infiniband wouldn't get you into the top 10,000. And has been noted above - if MS want to charge a couple of hundred per blade, real supercomputer owners will continue to just say no. The value added by Windows over Linux on a real machine is close to (and mostly less than) zero, most certainly not justifying the hundreds of thousands of dollars MS would want.
Finally, it isn't as if there is anything new in these Cray boxes. They are competing in a very crowded space. About the only slight innovation is the idea that it is a deskside machine and is quiet. They are not the first to try here either. All the others failed. Personally I would not be interested in a machine for compute intensive loads that was not on a proper UPS and was not protected from all the usual reliability issues that plague desktop/deskside machines.
A couple of comments here.
1) 486 using 240V (AC). Back home in Australia 240V is the standard.I still have a 386 & a 486 (in storage) that run on 240V. I moved to North America and was stunned to see special 240V runs were put in for "big iron". hahaha :)
2) There are reasons servers are put in server rooms: power consumption, temperature, security, etc.
Leaving a $25-80,000 server outside a server room is a pretty bad plan. It is at far too much risk of theft, damage, etc.
Just because it runs on 110V or whatever doesn't mean the desk has enough power points or that the local circuit can take the current demand. Did these guys even think this through?
> Cool! I guess it's the extra 10 volts that make those data centres so very expensive.
Assuming you're in the UK ...
Have you actually measured your mains supply recently? Do you really believe that the grid operators went out and replaced all their expensive step-down transformers overnight when the spec for the mains voltage was changed from 240V +/-5% to 230V +/-10% to "harmonise" with most of the rest of Europe, who also changed from their former 220V +/-5% to 230V +/-10%. No, I thought not.
What actually happened was ... nothing.
240V +5% is 252V, which is less than 230V +10% (253V). Similarly, 220V -5% is 209V, which is greater than 230V -10% (207V). So the whole of the ranges permitted under the two previous standards fits neatly into the single range of the shiny new EU harmonised standard and nobody has to do anything. Magic, innit?
Of course, now somebody will prove me wrong and show that new substations actually do generate a nominal 230V, but I doubt it. The power distribution companies probably use the extra range to permit a substation to supply a wider area, since the power drop between substation and consumer is proportional to the length of the cables.
Interesting that here in Germany lightbulbs seem to last a lot longer than they did back in Blighty. http://members.misty.com/don/bulb1.html#mll
"Technically, this desk-side mega-server is certified to run Linux distros from RedHat and Novell. But they weren't invited to launch party." .... http://www.theregister.co.uk/
Is that El Reg putting ITs desk-side mega-server, the Register, into the IntelAIgents Circle for some MaJIC Siamese Surprise in Shanghai Rain and Novel WAIs? What an Absolutely Fab Idea/Step/Quantum Leap.
"In short, people who will think they can now get a (semi) super-computer all by themselves, no scary IT geek needed (lab geeks -sorry, scientists- are not afraid of IT geeks. They've even been able to breed in some occasions. Reportedly). And my bet is that they're in for a nasty surprise (and, incidentally, so are we, I for one welcome our super-botnet overlords)."
What Sort of Beings bet on Nasty Surprises? Beings with Zero Positive Input ? Share a Few Lines of Encouragement 42 Change the World is Pretty Simple for CompleXXXX Situations/Environments.
... You get a comment to the effect "Windows has downloaded a critical upgrade and needs to reboot now".
That's what happened to me, I lost a huge simulation run, and the next day I rceompiled the whole lot under Linux and have never looked back.
Microsoft? What a charade they are.
... there is also a contradiction. For one, Windows kernel is not efficient enough for HPC - there will be too much locking and synchronization cost (in API calls, especially IO) to allow user threads to do any real work on dozens of cores available, concurrently . Of course, they MIGHT have fixed that in Windows 2008 HPC edition (or what Microsoft calls it). Even if they did, there is a question of security - one does not buy fastest machine possible, paying 6-figure bill, just to cripple it with antivirus. On the other hand, one does not run Windows WITHOUT antivirus. Unless it is on private and well-secured network. Which of course makes sense in the datacenter, but under the desk ... ?
While cute, it is just a bit overlyoptimistic. It should be:
10 Goo
20 Agiii
30 Whaaaaaaaaaa
40 Goto 10 : REM Colics, wet nappy and so on.
Actually, I forgot, it is a Microsoft HPC so there is one extra line:
25 Barf : REM "Projectile Vomit" again, system rebooted
Try walking into the boardroom and telling the suits you want to buy a supercomputer. To them, seeing as they lag about 20 years in IT knowledge, a supercomputer conjures up an image of a warehouse-sized datacenter packed with rows and rows of cabinets all running in one massive and very expensive proprietary system, with very little flexibility and poor re-use. Now, try going to the same suits and saying you want to save on the new project budget by taking advantage of "free software" (Linux) and x86 servers in an HPC cluster - same thing, different description, but even the ones that spend most of their lives "dining at the club" have read nice things about Linux and HPC and associate the whole shebang with the words "small" and "cheap", which really appeal to them as they think they're winners with the shareholders. Then say you would also like to look at a Windows option using the same kit. The same suits also see Windows as meaning "safe at only slightly more cost", so Windows and HPC to them will sound just fine.
I think that the description of this machine is what referred to as a Minicomputer... more powerful than a Microcomputer (PC) but not a "Mainframe" as super computers of old were thought of... The name they have come up with is a marketing term.
Microsoft's problem is not that nobody in the realm of supercomputing won't talk to them. It is that Microsoft's best efforts aren't good enough. Perhaps this is what is needed to run Vista... joking.
Really, Given the product as hardware, Linux will run quite well on it. It may well fit in a niche market... and once Cray figures out that Microsoft is just using them and that the real future is Linux, well, there it is...
I suspect that the BSOD on this must look like a train wreck... joking, but if you had 80 K to purchase a Minicomputer, would you risk it running "Windows".
Cray must be desperate.
Were did the dead bird go, that fits perfectly for this story.
I stare in disbelief. Why would one want to waste half of the systems resources to run some Windows? *g* Nonono, if I wanted a supercomputer with clickety interface, I'd rather have a cluster made of Xserves. Should be cheaper too, and as folks mentioned, in real life you would not want to operate this under your desk anyway.
You can actually run a PC just fine from 240 volts DC. If you look inside the tin box which keeps all the nasty voltages safely out of harm's way and lets out only the nice safe voltages, you'll find the first thing that the mains encounters is a rectifier.
And yes, you could run your PC from 200 D cells in series (200 * 1.2 = 240). Some fool probably has already tried this.
As pointed out earlier, this is a load of bollocks aimed at gobshites who want a supercomputer so they can namedrop 'cray' while talking shite. Windows for supercomputers is a joke, it isn't worthy to sniff the steam of the penguin's piss as (again, as pointed out earlier) the apps have to be optimised for the hardware to get real performance, one of these and a copy of 'stargazer extreme professional edition' isn't going to do much in the way of plotting the expansion of the solar system.
As to sheer number crunching, I very much doubt this thing could match a cluster of $25000 worth of ps3's, so whats the point of it? Lots and lots of off-the-shelf single threaded apps. The vista comments weren't to far off, its probably the only off the shelf system that can serve vista to all the desktops in a small to medium business.
Windows is pretty ridiculous for a processing cluster, the whole premise of windows was that it's a gui (hence the name)... It requires a videocard to boot, and will not support a serial console...
Why would you want all that extra hardware in your cluster nodes? Boot everything headless, redirect the console over a serial connection for debugging...
Also being closed source, you can't tune the software to your individual needs.
Also the costs, once you have a large cluster you want each node to be as cheap and power efficient as possible - so you can afford more nodes... You don't want to waste money on software for each node, and you don't want surplus hardware that you had to pay for, and which will consume power which you also have to pay for, and produce heat which you have to pay to get rid of...
You want each node to be diskless and network boot, have as much ram and cpu cores as is optimal for your workload, have whatever interconnect is necessary and nothing else. No video hardware, no onboard sound, no usb, no ps/2, nothing that will waste power.
Just look at the current market leader in the top500 supercomputers list - IBM.
Guys, there are too many unknowns, and there ARE people out there who look for power without the upfront pain (there will be pain afterwards IMHO, but then they have spent the money).
This could be a good move for MS (and Cray) - it's IMHO *way* too early to tell. Distributed computing has at lest the ability to make Windows a bit stable, and ought to be enough to keep up with animated cursors. Having said that, if history tells us anything it's that MS engineering will soon get rid of those benefits..
Wait and see. Meanwhile, I'll stick with Linux. I go for what works for *me*..
"The personal supercomputer is not a new idea, but this is the first time that Cray is trying it out in the market."
Wrong. Search google for "Cray YMP-EL92".
Another very sad story illustrating how Cray has gone downhill in recent years. Wintel box shifing!
Mine's the one with "I coded for Crays, and all I got was this stinking lab coat" on the back.
I mean what do they try to accomplish with that?
Traditional nubmercrunching doesn't care about GUIs, all users there are perfectly able to use a shell to complile their fortran programms and run them.
CGI? Well sure, paint your pictures with MSPaint. It's not like multiple processors would help you with that.
The main problems with Windows is just that there are virtually no applications availiable for it. And those few applications usually cost quite a lot of money and rarely are of acceptable quality. Just imagine having to download software from a wiered unsecure source onto your 80kEuro box. Imagine getting a virus or worm onto your expensive device during a multi-week computing run?
Microsoft sees that it cannot grow on the PC market anymore and is therefore trying to get into any market it can get. And they want to get away from the "we are total idiots" image.
Just look at Sony. They gained quite some reputation for having professional products.
They will all want one to run their ridiculously complex excel spreadsheets on. That way they don't get laughed at when they ask the Geeks if they can make it better only to be told their accounts database app should do that for them (If they hadn't bought it on purchase price and give away shiny things).
Of course they could get an app written but the tied in Accounts app vendor wants a million Pound for ten minutes work and anyway they would have to specify what it the spreadsheet does and that would mean admitting to all the fudges and bodges they use to the get the bottom line to say what they want.
SO... The crap that you have written yourself in excel visual basic might be able to return an answer in less than 5 days just by spending $25,000 or so. No Geeks. No questions about the crappy accounts system you recommended. No questions about the various 'suspense account journals' around 'training weekend' expenses.
At least that's what Microsoft and Cray hope.
So, 8 slots...
Smallest storage is 2 slots, smallest graphics is 1 slot, leaving maximum 5 slots for "compute" nodes (assuming you want to see the windows), in which you can get at most 2 processors.
So max of 10 processors, I suppose multicore makes that worthwhile?
It's not much more than one of those servers aimed at SME's without the added bonus of rack mounting.
I used to love Cray and SGI when they really did cool stuff. (They used to be siamese twins for a while until some surgery. )
I mean, they used to make computers with octagonal bench setas around the outside. They looked like something off Space 1999 and that's what you wanted if you'd spent a bundle on kit.
This sort of stuff is OK if you want to crunch some numbers/render some graphics *relatively* quickly and without requiring proper expertise, but Win HPC and a fairly small number of cores means it's pretty limited. And to get some grunt you have to put 3 chassis together. You're going to stick that behind your desk??
I yearn for the days when the engineering in these things was first rate and you got what you paid for. Even after my SGI Challenge boxes were overtaken in speed by laptops, you could turn it into a cool-arse fridge. http://home.planet.nl/~mourits/koelkast/
For a bit of grunt in a box, I'd rather go for the IBM blade chassis S. Then I can choose what to do by way of OS. It's got wheels too.
Fetch me coat, I'm orf.
"The specs say each CX1 box has up to two (or four for redundancy) 1.6kW PSUs. Yes that's going to be one toasty office!"
Not for long, the average US office outlet is on a 15A, 120V circuit, which can only be loaded to 80% -- 1440W.
It'll be way cheaper to stack a few short rackmount dual quadcore servers and stick a "Vista Ready" logo on it, then place a script on a cheap desktop to copy the job parameters to the "HPC cluster". The user will never know/care what OS the cluster is using.
Oh--if it's not on a raised floor and occupying an entire wing of a building, like a 1108 3x2, it's not "big iron".
netgeek
This product is too noisy, and produces too much heat for anything but a machine room. Note the suspicious lack of environmental specifications. It is more expensive and less capable than similarly configured Nexxus boxes.
It will go the way of Rocketcalc (doing this kind of thing almost a decade ago), Orion Multisystems, and SciCortex. (OK, I guess SciCortex's little machine is still around.)
Stick with two-socket SMP for desktop use. Bigger machines belong in the machine room.
This kind of box is great for monte-carlo simulation of derivative financial instruments.. used by Traders and Quantitative analysts who are more likely to run Excel on their grid than BLAST.
They can run Mathamatica or MATLAB or event Excel Services all out-of-the-box without a geek in sight.. but the big benefit is that you don’t need an army of IT support costing $100,000+ to support it.
Face facts winServ’08 is as good a kernel as Linux and MSMPI is an efficient implementation
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
