Slashdot Mirror
NEC Strikes Back With SX-8 Supercomputer
News for nerds writes "It was just 3 weeks ago that we learned IBM's BlueGene/L with 36.01 TFlops edged out NEC's Earth Simulator, but today NEC announces a new SX-8 supercomputer with a peak processing performance of 65 TFlops (press release). It may be available in the U.S. as Cray's OEM like SX-6."
65 trillion calculations per second and all I ponder is if NEC would mind using my user id while running seti@home.
Tom's Hardware stated the the guts are Packard Bell, and it comes with a WinModem. That sucks.
If you think
...can it figure out the question to the answer 42?
This is like an arms race of yesteryear. The Germans and the Brits with their battleships, the Americans and Soviets with there nukes, the Yankees and Red Sox with payroll. Except this race is way cooler and will likely pay off in a much more productive way.
Note that 'SX-8' pronounced 'Sex-ay' :-)
It's very impressive and all, but how is this going to benefit me down the line? It's not like they're affordable to small/medium businesses like the Cray or HP's highly valued Alpha DEC workstations.
We are stiffling progress at the lower level by pricing these systems well beyond the reach of the average researcher or multi-national oil conglomerate.
Why is this?
Wearing pants should always be optional.
I was about to say "With a computer this powerful, the world probably only needs 4 of these", but history tells me otherwise :)
Rock that crushes, Paper & Scissors that don't matter.
It seems like it runs Linux as they are claiming that it will use the Global File System for clustered FS operations.. unless their Global File System is different.
Banu
The 65 TFlop for the SX-8 is only an estimate while the 36 TFlop for BlueGene/L was real performance. So it is not certain that SX-8 will be faster than Blue Gene/L
@ Virginia Tech it is rumored the US Gov't is gobbling(harhar) up Apple XServers in hopes to build the "Ultimate" Super computer...
But then again these are just rumors as Srinidhi Varadarajan is being *highly* recruited
Right now, this SX-8 announcement is just a publicity stunt to generate some "shock and awe" in the small supercomputing community of national and commercial research labs.
Perhaps, the management of NEC should consider generating some "shock and awe" among the greater engineering community. I suggest that NEC donate computing time on an SX-8 to all the startups designing spaceships (e.g. SpaceShipOne). These startups are short on cash and cannot afford the kind of supercomputer that is needed for modeling the spaceships. Free time on a supercomputer would greatly assist these startups and would generate considerable shock and awe among engineers who daydream about what the predecessor of the M-5 computer could have been.
Apparently, we are gradually building all the technologies needed to accomplish intergalactic space travel. The short list is matter-antimatter energy (which is undergoing top secret research in the American government) and high-performance computers (like the SX-8, which will model the spacecraft and possibly serve as the on-board computer).
"Space...the final frontier. These are the voyages of the startship ..."
In other news, scientists create a robot that could replace half of my friends.
No Grendel! Stop attacking my village! I think you forgot "cluster."
How's my typing? Call 1-800-eta-shut
1) I won't go out on a limb but I'll stick my NEC out for this supercomputer.
2) SX-8? Is that the next in line to SSX Tricky?
3) As you can see, I am resolute on this war on Tera...and my opponent is clearly a Tera-flopping machine.
4) Please don't write articles with "$COMPANY Strikes Back" as its title. It begs "Return of the $RIVAL_COMPANY" as a follow-up.
And why is this good news exactly?
I have found a truly wonderful proof of Fermat's Last Theorem, but unfortunately this sig is too small to contain it.
Remember that the 36.01 TF figure for BlueGene/L was only using 8 racks.
The final BG/L will use 64 Racks.
Also, the SX-8 figure is only an estimation.
BlueGene/L is only at 36TFLOPS today, but by next year the full-size version is supposed to clock in at 180TFLOPS. The SX-8 has no chance to survive.
Actually, the computer price is quite reasonable. It's the nuclear power plant to supply the 21 million jigawatts required to run it and its cooling system that will set you back big bucks.
Even with 65Tflops it still takes 3 min to apply my desktop settings.
John Anthony Hartman
NEC say in the press release that these super computers start at ~$10,000 a month rental. Maybe that's the base line model but isn't $10,000 a month peanuts for a serious spaceship business?
Do you really wish to move Doom from a game to Reality?
I prefer the "u" in honour as it seems to be missing these days.
Epson PX-8.
...phil
"For a list of the ways which technology has failed to improve our quality of life, press 3."
on a side note: Where does Lawrence Livermore National Laboratory get all that money to keep buying the latest and greatest super computer
LLNL does nuclear research (basically simulating nuclear weapon detonations). We spend $400 Billion on defense per year, what is $200 - $300 million for the latest and greatest super-computer?
It will most likely be very expensive. The Earth Simulator was nearly $300 million dollars and that's just for the machine! I don't understand why people keep spending that much money when they could build a G5 cluster with the same overall speed for easily 1/10th the cost.
I do understand that there are some things that the NEC computers are better at simulating and processing than a G5 cluster like the one at Virginia Tech, but for overall price/performance ratio these NEC computers cannot come close to matching Apple's G5 clusters.
LLNL does nuclear research (basically simulating nuclear weapon detonations). We spend $400 Billion on defense per year, what is $200 - $300 million for the latest and greatest super-computer?
Also, since the US agreed to stop testing nuclear weapons, they've moved on to totally simulating them. Most often the research is done under the guise of astrophysics because the physical processes are almost identical. The US has to stay on top of things, and simulation is the only option to do so.
Perhaps, the management of NEC should consider generating some "shock and awe" by having Oprah give away free SX-8.
Nice. ;)
If you have everything prepared, then you get the supercomputer at the end to do your rendering, and in that way, it is cheap.
Its just idle time.
Do you have your software written to actually run on it correctly?
Do the people who analyse the results work for peanuts?
Have you tested it?
"Just one more revision" is another 10k.
Imagine, you get some time on a supercomputer donated to you.
You sit yourself down at the console, what do you do?
Well, after writing the obvious pi to 1e999 dp, or for a=0 to 1000000000000000000000 do print "whoa!"
How would you use your time?
do they run a terminal environment - the worlds fastest computer interactively scrolling messages to 100s of users, and only throttling up when someone stops playing and gets on with some serious work?
I also wonder, do SuperComputers run screensavers?
liqbase
If you only care about price/performance, then go with a non-clustered PC. Nothing will beat that. (Ok for certain codes the non-clustered G5 could possibly beat it, but that's basically a PC anyway.)
Finally! A year of moderation! Ready for 2019?
Here's your M5 for you, right here.
Entertainment science at it's finest.
And why is this good news exactly?
Obviously you never played through the expansion pack for Battlezone. Let the Chinese get into space and the next thing you know they'll be building cloaked, super-fast hover tanks that kick the crap out of the American units you thought were so powerful in the original game.
"...always new atoms but always doing the same dance, remembering what the dance was yesterday." -Richard Feynman
Has it come down to this? A TFlops-size contest?
Slashdot: Failed Car Analogies. Amateur Lawyering. Anecdote Battles.
How well does it handle XP SP2?
Who would win this election: Andrew Weiner vs Andrew Weiner's weiner.
Comment removed based on user account deletion
-Those who would give up essential liberty to purchase temporary safety deserve neither. -Ben Franklin
Because the G5 cluster is a cluster. It is not useful in all applications. I believe that earth simluator was very special purpose, and I believe that Blue Gene/L is/will be multi-purpose, but not general purpose.
Sometimes you need a real supercomputer instead of a cluster.
WooHoo! Just in time for Halo2/Half Life 2!!!
...Had this been an actual emergency, we would have fled in terror, and you would not have been informed.
the West (which includes the USA and Japan)
Traditionally, Japan has been considered part of the orient, while the USA has been lumped in the occident.
I agree that dedicating supercomputer time to investigating new technologies is good.
Low impact alternative energy sources to supplant existing technologies would be a good start.
"Provided by the management for your protection."
So chinese and korean bigots consider themselves equal to americans, whose bigots, on the other hand, consider themselves superior to the rest of humanity.
Have you looked at the amount of taxes taken out of your paycheck recently?
But in the US you can get away with that. Even be praised for your courage. In Europe, and especially the UK, it's like you are a paedophile or psychopath... (The standard trick in the UK is to get your *wife* to be responsible for the next company, but most of us run out of wives pretty damn
quickly...).
Good luck. I just spent two years literally homeless here in Athens GR, and am scraping my life back together (looks good so far), so keep believing in yourself...
Hint: The doves or local friendly birds *always* believe in you if you feed them. That's great reinforcement. Not only that but if you work hard at it you might get enough material to write a book (which blows away the reason to muck around with computers to begin with...). (I did think about this and maybe I still will write something
perhaps for the infamous gentleman from the Northeast US with the fun book covers...)
Cheers,
Andy
And then comes Google with its massively distributed Array of Inexpensive Reduntant Computers (AIRC) and uses all those giga flops to fetch p0rn links and pictures. I'm coming!
How many flops is 100000 low cost PC:s?
Dyslexics have more fnu.
Actually, it's part of the Department of Energy budget, which is $24 Billion. From: http://www.whitehouse.gov/omb/budget/fy2005/energy .html
The 2005 Budget provides $9.0 billion for the national security activities of the National Nuclear Security Administration to include maintaining the safety, security, reliability and effectiveness of the Nation's nuclear weapons stockpile; preventing the spread of materials, information, and technology of weapons of mass destruction by eliminating or securing nuclear materials and related infrastructure and providing the U.S. Navy with safe, effective nuclear propulsion plants.
Two IBM's BlueGene/L.
The SX8 uses a 90nm CPU clocked two times as high as the one used in the SX6 respective the erarth simulator. Taking into account all the additional improvements it's save to say the SX-8 will be twice as fast as the old one - so 65 TFlops seem reachable.
/ pap247.pdf
For those interested in how vector processing compares with scalar processing in terms of absolute performance and computational efficience, this paper of Leonid Oliker et.al. is definetely worth reading:
http://www.sc-conference.org/sc2004/schedule/pdfs
I was picturing a massively parallel array of little PDIP Ubicom SX microcontrollers, all running at 50 MIPS. http://www.ubicom.com/processors/sx-family.htm
[begin silly] I don't know exactly what it'd take, but I'd think 50 8-bit operations might be able to handle most floating point operations.
That'd mean each SX could handle 1 MFLOP... so we only need a million of them. At 3$ each... with radio-shack proto-boards... it would only cost around $10 million including assembly. [end silly]
Note that they don't specify the bus width of the SX-8 system. Is it 32 or 64 bit? More?
Cool,
Now ID soft can release Doom3 written in VB! I bet this thing willl get close to 30 fps!
Artificial intelligence is no match for natural stupidity
... until they come out with something better than the SX-64.
We have more than sufficient computer power on our desktops to do the maths needed for a designing something like SpaceShip One. What's killing the startups isn't lack of cash, but lack of experience (both individually and across the industry) needed to make valid and rational engineering tradeoffs. (Not to mention that they aren't building for a market, but in hope of a market, thus making the design/tradeoff process even harder. Nobody knows what to design *to*.)
also, each cpu has 7 times higher io bandwith...
The 65Tflops estimate is for 512 8 cpu nodes, which less then earth simulators.
Plus each of the cpus now has 16GFlops (and can archive it with that much memory bandwith it has), but uses less than half of the power of the old sx6.
So a "earth simulator 2" with 65TFlops would have 20% less cpus and 60% less power consumption then the first.
HI O WISE PRINCE. WHT TOOK U SO DAM LONG?
...I'll wait for the DX version to come out, those SXs have the math coprocessor disabled.
For certain kind of calculations Virginia Tech cluster is about as useful as one X-server with two processors... Each node in Real Supercomputer can communicate about gazillion times faster with other nodes when compared to nodes in cluster based on slow 1gb/s ethernet. Clusters are great for Seti@Home, but they suck ass in another kind of number crunching. Btw, BlueGene/L actually is actually only 18TFlops when doing calculations that need lot of communication with other nodes. Every other processor is dedicated for communication only. BlueGene reaches its peak rate only in cluster favorable calculations. (It's still much better than VT's machine for non-cluster favorable stuff). This just proves that you really can't get a Real Supercomputer for free. But sometimes you almost can.
...you'd really want more than four if you were going to make a half-decent beowulf cluster of these things.
Karma: It's all a bunch of tree-huggin' hippy crap!
This is good news to all those who aspire to emigrate to the "West" from the third world via university admissions (you know, the F1, H1, GC route). Ah of course, because us third worlder's would rather travel half way across the world, forsaking family and friends, to obtain the education and professionaly opportunities from the west we would never have otherwise; rather than staying around to get it on our our own countries if that were possible. BTW, happy "Apping". Thanks but I've already got into one your "prestigous" universities, and with my other third world cohorts, we're eating away at it from the inside.
I have found a truly wonderful proof of Fermat's Last Theorem, but unfortunately this sig is too small to contain it.
when comparing 2 totally different architectures:
Blue Gene/L ( power pc-440 cpu vs Earth Simulator ( sx6 ) cpu.
Memory Latency: 7 microsec vs 2 microsec.
Memory Bandwidth: 1.4 GByte/sec vs 5 GByte/sec per node
Register size: 64Bit? vs 16Kbit ( 16 Thousand Bit! ).
-> There's certainly cases where one 1 BlueGene/L TFlop ~ 1 B Earth Simulator TFlop, but for plenty stuff Earth Simulator will always outperform BlueGene/L, even if it gets cranked up to 360 TFlop.
( Data taken from the latest CT journal from www.heise.de )
You can't compare those figures. The 65 Tflops is estimated peak speed, the 36 Tflops is measured sustained speed on LINPACK.
20 petaflops is due around 2009/10 under Moore's law. (And I for one offer an early welcome to our expected new AI overlords ...)
It doesn't matter which ape activates the Monolith
No, not really. ARM was "Acorn RISC machines" in the old days. Acorn was the UK company who did the "BBC Micro" (which was c.a. 1982) the most innovative machine around for the home market in the UK (and the basis for a pioneering effort by the BBC "Making the most of the Micro" to educate
people about the possibilities of "little machines" (phrase borrowed from Dr.Pournelle)).
(Ironically, it used a picture of an owl (Athena's
Owl) and I'm living here in Athens GR these days)
Later, they developed the ARM processor for the "Archimedes" range of machines, which had some nice
ideas, but sadly wasn't so successful (too expensive)).
(DEC had rights to the ARM, but didn't own the company). Sidepoint: ARM was RISC before most except the original IBM processor.
It was "droppable" into the corner of your design,
so it got designed in all over the place (many places in embedded processing you wouldn't be aware of). You point this out, and I think it was well liked because (cough) the bugs were well known (I haven't played with it, but *having* bugs
is much less important than having *unknown* bugs). I could mention a real *mother* of a
floppy disk controller which was deaf to commands.
You're right about the design of x86. It sort of
grew like a cancer. But it's well known that the 286 happened because Intel screwed up major league with their ambitious iAPX-432 (multi chip) project to replace the 8080. I've heard that
it (the '286) was designed in a couple of weeks. I believe that. It sure looks that way. Nothing since has changed that.
A lot of the "baggage" Intel processors carry around is stuff from the ill fated (but much
copied - even Unix copies some ideas) Multics project. Hence the rings and gates which are stuff
I never expected to see on my desktop after
banging on them at university. When Multics finally managed to deliver (and it started in the mid 60's and eventually delivered *something* usable in the late 70's!!!), the world had moved on... (For those who *like*
Multics, I understand, but computers as power stations is *so* 1960's). Looking ahead, it might
once again become an issue (Grid computing).
(Don't flame me if you're a Multics fan - I know
about B0 etc...)
I pretty much shudder at the thought of what rules the vector processor on the NEC uses. You'll need a custom compiler *and* you'll have to follow whatever black art coding rules they tell you to do. Scratch any idea of clean coding if you want fast...
Maybe I'm wrong.
Serious wish territory: is it too much to ask for
a processor like even ARM with hyperfast net interconnects a la Inmos Transputer, *and* fast cheap memory (the holy grail even in 386 days).
(Ever since then, memory hasn't kept up with Moore's law). Oh, and obviously I want a *really*
fast FPU.
Someone make me really *cheap* fast low latency memory. Pleeze. (Don't call me with DDR2 ok? or RAMBUS). Ok, so I want it all. But if you read
the "386" manual from Intel, you'll see that the
nightmare we are now stuck in is mostly the curse
of molasses slow memory. Static memory was expensive then, and is *still* expensive. Anyone
making *cheap* static memory could burn Intel...
(which would be deliciously ironic, since Intel's
first product was er.. (prob wrong) a 256 bit RAM chip)).
Personally, I kind of sympathize with a ghost voice from the past who coded for the wonderful
PDP-11. Now, even though I started during the '11's slow decline, I can *look* at assembly code
for that machine and understand it (mainly because
Motorola had the excellent taste to *steal* a lot of ideas when they made the 68000). (Except for
the 11's middle-endianness. Bloody 1(A0) problem.)
I keep promising myself I'll hack up an '11 emulator so I can look at the ancient root of
the OS I started my career porting (ok, you read slashdot, so you guess). Something about Pascal I'd guess. Poke me if you want an old-timer to *really* tell a story...
(yes I know there are a lot of them, but it's not the same...)
Time. She's a bitch.