backtop

Print E-mail del.icio.us 37 comment(s) - last by Silver2k7.. on Jan 2 at 5:13 AM
AMD processors made their way from Dubi to Iran, only to show up in Iran's most powerful supercomputer

Iran has a supercomputer constructed with 216 AMD processors, and it's got a lot of websites pretty mad. 

The whole thing reads like an advertisement really.  "An enormously powerful supercomputer" proclaims Slashdot. "such advanced U.S. computing technology is a real breach in U.S. sanctions," warns a UPI editor.

It's true that U.S. semiconductors are banned from entering Iranian borders, but I wonder if those editors know the computing power 860 gigaFLOPS really commands in the HPC world.

A total of four PlayStation 3s running Linux have just a hair less computing power than this computer.  A pair of IBM's newest and tiniest BladeCenter QS21 servers would out-compute Iran's new supercomputer without breaking a sweat.

InformationWeek was quick to get a quote on the story: "AMD fully complies with all United States export control laws, and all authorized distributors of AMD products have contractually committed to AMD that they will do the same with respect to their sales and shipments of AMD products ... Any shipment of AMD products to Iran by any authorized distributor of AMD would be a breach of the specific provisions of their contracts with AMD."

Just for comparison, IBM's BlueGene/L, the world's fastest publicly-known supercomputer, runs at a mere 478 teraFLOPs.

Amirkabir University of Technology, owner of the supercomputer, claims the system is for weather forecasting -- incidentally using the MM5 platform developed by the U.S. National Center for Atmospheric Research.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
Except...
By Zurtex on 12/11/2007 10:14:17 PM , Rating: 2
That's not quite true as it's both easier to write things for AMD x86 chips than it is for Cell CPUs and AMDs x86 chips do general things faster than Cells do.

Somehow I doubt down at the meteorological department they have dozens of PS3's running their system.




RE: Except...
By KristopherKubicki (blog) on 12/11/2007 10:44:07 PM , Rating: 4
MM5 runs just fine on Cell BE actually.


RE: Except...
By s12033722 on 12/12/2007 12:59:17 PM , Rating: 2
You assume that they really are running MM5, at least full time. I don't think that's a valid assumption. Also, while in terms of pure computational capability the Cell-based solutions may be able to compete, when you consider the extremely narrow field of computations Cell is capable of running efficiently and you also take into account the vastly superior cache and memory scheme that is a part of a system like the one Iran has built, comparisions between PS3s and that system become silly.

The argument this blog is pushing is similar to the following made-up example: Iran just purchased 16" artillery pieces, but my .22 has the same muzzle velocity as those artillery pieces, so who cares? They can do the same thing!

Obviously, if you choose a very narrow set of capabilities to analyze, comparisions between a 200+ node supercomputer (admittedly on the small end of supercomputers) and PS3s or other Cell-based systems can be made. When you look at them as total systems, however, it's really quite blatent that such comparisions are trite. Make the point that supercomputer is small, but don't bring the PS3 nonsense into it.


RE: Except...
By KristopherKubicki (blog) on 12/12/2007 1:32:31 PM , Rating: 5
Well, I thought that was a pretty safe assumption since it says all over the EDU website its going to be running MM5 full time :-P


RE: Except...
By s12033722 on 12/18/2007 4:28:46 PM , Rating: 2
What I was getting at is that I don't exactly trust the Iranians to be shining examples of truth and transparency. Just because they say it will be running MM5 doesn't mean it's so.... ;)


RE: Except...
By sinful on 12/14/2007 2:01:00 AM , Rating: 2
quote:
Make the point that supercomputer is small, but don't bring the PS3 nonsense into it.


Agreed, the PS3 comparison is totally nonsense.

The Cell is a special purpose chip that does well in a specific specialty subset and extremely poorly outside its niche.

The AMD based super computer is pretty much the swiss-army knife of supercomputing - it can do anything, and do it well.

Comparing the two is extremely misleading because of massive differences in architecture.

What is really amazing is how so many people hype the Cell.

The Cell's only real claim to fame is throwing lots and lots of Mhz at the problem, with a bunch of cores.
That's it. Each individual core is relatively primative.

In fact, a Cell is not much more than 8 486's @ 3.2Ghz glued together with better IO.
Each core in a Cell is ridiculously simple - but when you have 8 cores, and they're all running at 3.2Ghz, you can get good performance out of them in certain situations.


RE: Except...
By wordsworm on 12/12/2007 2:38:42 AM , Rating: 1
So, does that mean you think that Iran might have to buy Japanese CPUs to create supercomputers? How viable would it be to start making powerful servers from Sony's cells? I can't help but think that Sony would love to become a player in that market.


RE: Except...
By KristopherKubicki (blog) on 12/12/2007 3:12:01 AM , Rating: 2
They are already. That's why I mentioned the Cell BE servers from IBM.


You are telling me that 4 PS3s
By SilthDraeth on 12/12/2007 11:35:01 AM , Rating: 3
Are as powerful as 286 AMD processors.




RE: You are telling me that 4 PS3s
By KristopherKubicki (blog) on 12/12/2007 1:33:43 PM , Rating: 2
Pretty close.


RE: You are telling me that 4 PS3s
By soydios on 12/12/2007 1:37:26 PM , Rating: 2
but only with specially written and optimized code.


RE: You are telling me that 4 PS3s
By KristopherKubicki (blog) on 12/12/2007 1:49:14 PM , Rating: 3
If you're planning to run anything across 216 processors, it's already specially written and optimized code.


RE: You are telling me that 4 PS3s
By Zurtex on 12/13/2007 6:09:35 AM , Rating: 2
... well there's bandwidth, cache issues and general architecture of the CPUs.

I mean I could buy an 8800GTX and say it's better than 10 of any current x86 Quad CPUs in terms of sheer computer power (gflops and the like). But if I wanted to write a program to run a prime deterministic test or factor a number, I'd struggle like hell writing a code that worked in streams for the 8800GTX regardless of its gflops.

IBMs 2nd generation Cells are more suited to super computers than the ones that go in to the PS3s. But my understanding is even with those it's useful to hook them up with Opterons anyway because they can very easily negotiate where all the data needs to go which Cells have yet to to be able to do in large super computer environments, mainly because of compromises of number of cores over other parts of their architecture.


This Reminds Me...
By ziggo on 12/11/2007 11:04:46 PM , Rating: 2
Of the whole "use our radar in Kazakhstan" debacle with the Russians. We go in to inspect it only to find out the thing occupies a fairly large building and runs on vacuum tubes. (although more advanced ones than we had when we stopped pursuing them)

I guess ITAR might actually work, but it usually is just a major pain the ass from my perspective.




RE: This Reminds Me...
By maven81 on 12/12/2007 1:53:51 PM , Rating: 1
That installation is in Azerbaijan not Kazakhstan, which is important because it's right next to Iran. If the goal is really to detect Iranian launches, refusing to use the system is nothing but a game of politics. I also question your vacuum tube story.


RE: This Reminds Me...
By ttour on 12/12/2007 4:05:02 PM , Rating: 2
The whole point concerning the vacuum tubes is, they were/are the "Best of Breed" reliable technologies at the time the system was designed and implemented and can be utilized efficently, and cheaply in a (high) power generation section of radar systems, and I think it's really a standard way to design and implement radar systems. Most likely, radar is what we're talking about as one of the primary "detection" technologies there. Also, vacuum tubes were probably easily obtainable and the systems easily maintained. The vacuum tubes reported in the building I don't see it as unusual, especially since we are talking russian technology 40 years ago or longer when the site was implemented, (not sure of the history there for that site but I'm guessing it's been there in some form or another for a long time). Nothing wrong with cheap, reliable and easy maintenance. Also I don't think vacuum tubes are sensitive to Electro-Magnetic Pulses that can wipe out electronic systems. Important only if the other side gets a little giddy with war toys.


RE: This Reminds Me...
By maven81 on 12/12/2007 5:48:17 PM , Rating: 2
That's a good point... just because certain circuitry is old fasioned, it doesn't automatically mean it's crap. Witness the revival of analog, which no one wanted for around 2 decades!
I do remember reading somewhere that this installation was built in the 80s which is why I wonder how true that claim about tubes actually is. But I also recall that they were in use in the 70s. In fact I've actually come across russian schematics at one point, which used tubes, transistors AND chips all together! Which tells me the designers of that time were less worried about bleeding edge tech, and more concerned about getting something to work well.
I just don't buy into this myth that their tech somehow didn't work or was a joke. Witness nonsense like "space cowboys". As far as I know tubes were never used in ANY space missions! (unless you count vidicon tv cameras)


RE: This Reminds Me...
By ziggo on 12/12/2007 11:41:36 PM , Rating: 2
As requested a link to reference the vacuum tube claim:
http://www.nytimes.com/2007/11/05/world/europe/05m...

I don't think, at least I hope, that they are not using the tubes for processing, I think they are using magnetrons to generate the energy. As for the claim that the old technology is better:

From the "Radar Handbook" 2nd Edition (it is 1200 pages long; "handbook" indeed)

Solid-state devices have largely superseded vacuum tubes in logic and other low-power circuits and even in some very high power applications such as power supplies and power converters below 1 MHz. The only exception seems to be cathode-ray tubes (CRTs), which are less costly than large plasma displays. In radar transmitters, the transition from high-power klystrons, traveling-wave tubes (TWTs), crossed-field amplifiers (CFAs), and magnetrons to solid-state has been more gradual because the power output of individual solid-state devices is quite limited. However, compared with tubes, solid-state devices offer many advantages:

1.
No hot cathodes are required; therefore, there is no warmup delay, no wasted heater power, and virtually no limit on operating life.

2.
Device operation occurs at much lower voltages; therefore, power supply voltages are on the order of volts rather than kilovolts. This avoids the need for large spacings, oil filling, or encapsulation, thus saving size and weight and leading to higher reliability of the power supplies as well as of the microwave power amplifiers themselves.

3.
Transmitters designed with solid-state devices exhibit improved mean time between failures (MTBF) in comparison with tube-type transmitters. Module MTBFs greater than 100,000 h have been measured.

4.
No pulse modulator is required. Solid-state microwave devices for radar generally operate Class-C, which is self-pulsing as the RF drive is turned on and off.

5.
Graceful degradation of system performance occurs when modules fail. This results because a large number of solid-state devices must be combined to provide the power for a radar transmitter, and they are easily combined in ways that degrade gracefully when individual units fail. Overall power output, in decibels, degrades only as 20 log r, where r is the ratio of operating to total amplifiers.

6.
Extremely wide bandwidth can be realized. While high-power microwave radar tubes can achieve 10 to 20 percent bandwidth, solid-state transmitter modules can achieve up to 50 percent bandwidth or more with good efficiency.

7.
Flexibility can be realized for phased array applications. For phased array systems, an active transceiver module can be associated with every antenna element. RF distribution losses that normally occur in a tube-powered system between a point-source tube amplifier and the face of the array are thus eliminated. In addition, phase shifting for beam steering can be implemented at low power levels on the input feed side of an active array module; this avoids the high-power losses of the phase shifters at the radiating elements and raises overall efficiency. Also, peak RF power levels at any point are relatively low since the outputs are combined only in space. Furthermore, amplitude tapering can be accomplished by turning off or attenuating individual active array amplifiers.

Also, you will notice in the nytimes article above the Russian radar is functional as an early warning radar, but to precisely track a ballistic missile for a mid-course (exoatmospheric) intercept you need that radar to be more capable than its hardware permits.

Analog still has its uses, particularly in the signal processing world, where simple analog filters can offer more throughput than the best IC designs.

I did screw up the location of the Russian radar, but in my defense, the Russians do have 6 early warning radars in Kazakhstan


RE: This Reminds Me...
By ziggo on 12/12/2007 11:43:53 PM , Rating: 2
Edit:

FYI.. A magnetron in a type of vacuum tube. One of the oldest designs, but very power efficient. It is what is used in microwaves to generate the RF energy.


By s12033722 on 12/12/2007 1:18:01 PM , Rating: 1
It would have still been in the top 500 as of 2004. It's a powerful system, more powerful than the systems we were doing nuclear weapons simulations on only a decade ago, and quite capable of doing such simulations today, something the Cell systems are notably unable to do thanks to their extremely limited I/O capabilities. The comparision to Cell systems is extremely misleading.




By KristopherKubicki (blog) on 12/12/2007 1:34:42 PM , Rating: 2
quote:
something the Cell systems are notably unable to do thanks to their extremely limited I/O capabilities

Where did you read that?


By GeorgeOrwell on 12/12/2007 7:59:16 PM , Rating: 1
Are you referring to the systems that have been built using Cell processors or to the Cell processors themselves? For the Cell processor itself has very good memory and I/O capabilities:

"Cell’s On-Die Memory Controller

For years, we’ve known that Rambus’ memory and interface technology is well ahead of the competition. The problem is that it has never been implemented well on a PC before. The Rambus brand received a fairly negative connotation during the early days of RDRAM on the PC, and things worsened even more for the company’s brand with the Rambus vs. the DDR world lawsuits.

Rambus has had success in a lot of consumer electronics devices, such as HDTVs and the Playstation 2, so when Cell was announced to make heavy use of Rambus technologies, it wasn’t too surprising. As we’ve reported before, Rambus technology is used in about 90% of the signaling pins on Cell. The remaining 10% are mostly test pins, so basically, Rambus handles all data going in and out of the Cell processor. They do so in two ways:

First off, Cell includes an on-die dual channel XDR memory controller, each channel being 36-bits wide (32-bits with ECC). Cell’s XDR memory bus runs at 400MHz, but XDR memory transfers data at 8 times the memory bus clock - meaning that you get 3.2GHz data signaling rates. The end result is GPU-like memory bandwidth of 25.6GB/s. As we’ve mentioned in our coverage of this year’s Spring IDF, memory bandwidth requirements increase tremendously as you increase the number of processor cores - with 9 total in Cell, XDR is the perfect fit. Note that the GeForce 6800GT offers 32GB/s of memory bandwidth just to its GPU, so it would not be too surprising to see the Playstation 3’s GPU paired up with its own local memory as well as being able to share system memory and bandwidth.

The block labeled MIC is the XDR memory controller, and the XIO block is the physical layer - all of the input receivers and output drivers are in the XIO block. Data pipelines are also present in the XIO block.

As we’ve seen from AMD’s Athlon 64, having a memory controller on-die significantly reduces memory latencies, which applies to Cell as well.

Cell’s On-Die FlexIO Interface

The other important I/O aspect of Cell is also controlled by Rambus - the FlexIO interface. Cell features two configurable FlexIO interfaces, each being 48-bits wide with 6.4GHz data signaling rates.

The BEI block is effectively the North Bridge interface, while the FlexIO block is the physical FlexIO layer.
The word “configurable” is particularly important as it means that you don’t need to connect every wire. Taking this notion one step further, don’t look at the FlexIO interfaces as being able to connect to one chip, but rather multiple chips with different width FlexIO interfaces.

One potential implementation of Cell’s configurable FlexIO interface. While Cell’s XDR interface offers over 2x the memory bandwidth of any PC-based microprocsesor, Cell’s FlexIO interface weighs in at 76.8GB/s - almost 10x the chip-to-chip bandwidth of AMD’s Athlon 64.

In Playstation 3, you can pretty much expect a good hunk of this bandwidth to be between NVIDIA’s GPU and the Cell processor, but it also can be used for some pretty heavy I/O interfaces.

One of the major requirements in any high performance game console is bandwidth, and thanks to Rambus, Cell has plenty of it.

Quoted from here:
http://www.anandtech.com/printarticle.aspx?i=2379

Much more information available here:
http://www-01.ibm.com/chips/techlib/techlib.nsf/pr...


what about
By baseball43v3r on 12/12/2007 1:26:29 AM , Rating: 1
quote:
Just for comparison, IBM's BlueGene/L, the world's fastest publicly-known supercomputer, runs at a mere 478 teraFLOPs.


anybody know what the worlds privately-owned, or secret government project runs at? or is projected to run at? anybody got an insider into the government? although i would assume the government would just outsource this kind of thing.




RE: what about
By killerroach on 12/12/2007 9:59:55 AM , Rating: 2
It's mainly due to the fact that supercomputers presently in the planning or building phases don't have good information on their final capabilities yet. That being said, I think I recall reading something recently about IBM wanting to make an expanded BlueGene supercomputer capable of cracking the petaFLOP barrier, although I don't know where in the pipeline that is yet.


By greylica on 12/12/2007 4:02:16 PM , Rating: 1
United states could think twice before entering another war, all over the world people are talking about the recent economical crash over home financing.
Course, Iran have to think about energy too, but Nuclear energy is dangerous to the world, they have to stop it too and try to avoid that energy, but they have a bunch of sand in their territory, sand haves radioative fuel for them , making it very cheaper.
Wind and sun energy can help them too, but without computer help, I doubt they could do efficient systems.
But, for the U.S. economical ecosystem, this position over Iran could lead to a second disaster.
You have a big territory for your nation, ocupying yourselfs with external problems is not as good than to build wind generators to use over petroleum.
This could lead to avoid a new disaster like Irak is.

There is a bunch of energy clean and useful, for you and for them, both of you are doing the wrong path.

It´s just a shared opinion outside the U.S.




By Christopher1 on 12/14/2007 12:21:02 AM , Rating: 1
The US economy is not fake, and neither is the global economy.

I do not know where you got that idea, but there has always been an economy like the one that we have now (though I don't like this economy and would prefer something more communistically-based).

You are right however that inflation has NOT been controlled well at all. If it had been, we would still be paying the prices for goods and services that we paid in the late 1950's now.
We are not doing that, and that shows that inflation, definitely was not controlled well in the slightest.

As to the energy problems and the problems with oil prices.... those are coming from the rampant oil market speculation. If we would do away with that speculation on oil markets and make them take into account things RIGHT HERE AND RIGHT NOW..... the price for gasoline, even taking into account the war in Iraq, would be about $1 dollar a gallon.


By GeorgeOrwell on 12/17/2007 2:43:45 AM , Rating: 1
The global and US economies are based on currency values that are managed by a Central Bank. These currency values are not based on any tangible/real/physical value store.

As the currency of these economies is fake, the economy itself is fake.

It is not to say cars don't get made and sold, but that the currency value of all economic transactions has little, if any, relation to an authentic value model. Just look at the price of cars over time.

Inflation exists as a way of compensating for the Central Bank making more money out of nothing. This is the primary reason that a US dollar buys next to nothing compared to the 1950s.

Incredibly poor decisions are made because money is fake. The banking game will likely result in the death of humankind, all for the sake of some sick people who cannot kick their "love of money" habit. It will not be foreign invaders that destroy this world, but the pathetic creature, full of hubris, that we can "Man".


there's a way around all this
By serajadeyn on 12/13/2007 11:20:42 AM , Rating: 2
Iran could just buy up all the N64s from game stores and cluster them. I mean, they could buy several thousand of them for what? $50? The electricity to run them would cost more and they could exceed their current computing power. Then again, who wants to build a warehouse the size of the Chrysler building to house them? :P




By Christopher1 on 12/14/2007 12:27:04 AM , Rating: 1
Does anyone really think that if these sanctions were on the United States instead of Iran, that we would REALLY kowtow to them?
No, and this is the reason why these limits on technology really do not make sense to me. I am sure that our military has things that are..... thousands of times faster than these that have not been publicly acknowledged..... so why is the military and government so worried about them having older computer hardware or even current generation public computer hardware?

It just does not make sense in the slightest, and makes me think that this is a 'straw man' to beat the drums of war on.

If anyone can explain this, and the reason why this is a danger to us (and don't start with the 'they can put it in missiles!' argument, that is a fallacious one), I am willing to listen.




By ZJammon on 12/12/2007 9:44:49 AM , Rating: 1
Sounds almost like a lame tv series. Wait a second, aren't you suppose to be on strike?


By Terberculosis on 12/12/2007 11:19:42 AM , Rating: 2
Is it just me, or are there a lot of people here who don't apreciate sarcasm?


By GeorgeOrwell on 12/12/2007 7:45:20 PM , Rating: 1
As the masses slip further into ignorance and dullness, the wit and gameplay of sarcasm shall be lost, replaced by the base humor and crude violence that puts on a fancy cape and calls itself "television" or "hollywood". Maybe for all intents and purposes, it is already so.

Out of the caves man came. But back into caves, darker than any cave ever was, are what man lives in today. Man laughs away the pain, drinks to dull the dullness, and tells himself that happiness is only a channel change away, that his life matters because he watches the "right" channels.

So much for "in his image". So much for the power and glory.

Yes, my man. The number is large, countless. THEY ARE LEGEND.


By ZJammon on 12/12/2007 11:11:40 PM , Rating: 2
One has to wonder what kind of dribble is being passed off as sarcasm these days. As for your rant about the ignorant and dull masses of uneducated cavemen… Do I detect a hint of elitism? You seem enamored with the idea that you’ve somehow grasped a truth that none have accessed. Such intellectual snobbery sickens me as I run into it often in the world of academia.

On a side note, you make many references to “Hollywood”. Scary the day one falls into darkness they charge others have…


By OxBow on 12/12/2007 11:44:50 AM , Rating: 1
Excellent!

In response to the question of how fast the governments secret supercomputer runs, I don't think they've been able to get that memo out of their committee to review legacy boondogles.


By Silver2k7 on 1/2/2008 5:13:45 AM , Rating: 2
It took what 20 years for the government to aknowledge that they had a faster chip than intel did back in the 8086 dats.. so do they still make chips and how fast are they now.. would be interessting to know this..

If the government/military do have a secret fab somewhere, I guess someone should be able to follow the money trail.. since a fab costs what 30-40 billions ?


"Young lady, in this house we obey the laws of thermodynamics!" -- Homer Simpson











botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki