Cray Supercomputers to be Based on AMD Opterons

PsychicX writes "AMD and Cray have announced an agreement to base Cray supercomputers on AMD's Opteron line until the end of the decade, and to collaborate on Cray's 2006 proposal for Phase 3 of the federal government's DARPA HPCS (High Productivity Computing Systems) program. Cray already offers the XT3 and XD1 supercomputers based on Opteron."

Comments

excellent

[harryoyster]

That is excellent news for AMD even though there wont be massive volumes compared to home markets it will still be some heavy industry weight backing the AMD opteron processor. Hopefully AMD will adopt some additional features that could make the Opteron even better suited to the super computer market.

Re:excellent

[14erCleaner]

Except that this "industry heavyweight" is actually Tera, the little company that bought out the Cray name as the supercomputer industry was dying. Only the name Cray remains, not the old-time reputation.

Re:excellent

[Rude Turnip]

This is really just a marketing play on AMD's part. Now they can sell you:

"The SAME CPU used in CRAY ***SUPERCOMPUTERS***, now available for your desktop!"

And some rube will buy one based on that statement.

Re:excellent

[fgodfrey]

Tera bought far more than a name when they bought us. They also bought a bunch of software and hardware people, many of whom (myself not included) have been with Cray Research (the original Cray) for many years. So, while it's certainly not the Cray of the mid-1980's, the tradition still goes back there, especially with the vector machines like the Cray X1/X1E and its impending follow-on.

Re:excellent

[Aardpig]

Only the name Cray remains, not the old-time reputation.

That's not quite true; they still sell Cray-specific technology. One of my colleagues has just bought a small 24-core Opteron system. Each node contains two dual-core processors, and the 6 nodes are linked together by 80 Gb/s Craylink cables. I think this interlink technology is also licensed to SGI for use in their Origin computers.

Re:excellent

[evilviper]

The firm already has manufacturing bottlenecks; its production fabs are cramped as is. There are also problems with inventory.

What bullshit. Provide some sources for this info, or shut up. AMD is opening a new fab, and has a contract with a 3rd party to produce cores if AMD can't keep up... They're doing fine producing Opeterons.

Incidentally, I can provide real, actual sources that show Intel is the one who is having problems producing enough chips to meet demand.

Re:excellent

[flaming-opus]

Which is true only insomuch as the old-time reputation could not possibly exist today. That was the cold war, this is the post-coldwar era. The old cray was a mammoth beast with its own share of myopia, but a lot of technical tallent. This allowed a few really brilliant concepts, and a lot of clever implementation to power two decades of brilliant computers. That said, they were brilliant solutions for the era. Old-school cray systems were built in an era when doing fundamental pieces of math was still pretty difficult, and the government was willing to pay ten million dollars for a machine that was proficient at doing math, and many tens of millions for a machine that was really good at it.

The difficult problems in building computers has changed, and the financial climate around supercomputers has changed quite a lot. Among other things, CMOS finally became fast enough to put bipolar in its grave, single microprocessor workstations became powerful enough to do all but the hardest of scientific tasks, and the average price of high performance (not top 10 on the list, but still fast) computers has plummeted. To ask the new cray to be like the old cray would be foolish.

That said, New Cray is still offers impressive products. All of Cray's 3 product lines have much lower entry-prices than similar crays of the 90's. They all have more managable power/cooling/physical size characteristics. They make much greater use of industry standard Disks and networks, and also can be administered and programmed much more like any other unix computer. You program a New Cray more or less the same as other contemporary HPC systems.

When cdc introduced the 6600, the president of IBM complained to his staff aking (paraphrase) 'how has cdc managed to best IBM's fastest computer with a staff of just 14 engineers and 4 programmers?' Seymor Cray responded "It seems like Mr. Watson has answered his own question." Because new Cray is tiny does not mean that it is not capable of making impressive innovations. Old Cray's Gorilla days were very wasteful, and not necessarily full of the best moments of innovation.

Now, if only they could put four X1e CPUs into an air-cooled, rack-mount server and charge a reasonable amount for it. I'd much rather have a handful of vector processors than a few dozen opterons, anyday.

Re:excellent

[I Like Pudding]

One of my colleagues has just bought a small 24-core Opteron system

Small? Small? You've lost your sense of scale with all the weather modeling and nuclear RSA NP-hard protein folding simulations you've been running.

Re:excellent

[tzot]

...by 80 Gb/s Craylink cables. I think this interlink technology is also licensed to SGI for use in their Origin computers.

Which they call NUMALink (for Non-Uniform Memory Access --a node obviously accesses its memory faster than through the interconnection), and they're used in their Altix (Linux on Itanium2) computers too; I don't know how much they have evolved the technology, I just know that it's 3.2GiB/s each direction.

NUMALink

Have you got any links to some page describing some configuration similar perhaps to your colleague's system? I haven't heard about 10GiB/s CrayLink and find it very intriguing.

Re:excellent

[flaming-opus]

Close.
Craylink was designed at SGI, and renamed to craylink after they bought Cray. They introduced craylink in the origin2000, which they started selling half a year after buying cray, so I'm sure they couldn't have integrated any cray-designs into their product in that span.

After they sold Cray to Tera, SGI started calling the technology Numalink, and currently use it in their origin3, altix3, and altix4 product lines. They are on the 4th generation of the technology, which is 3.2GB/s per direction. The cray that was sold to Tera included the half-finished X1 system, which also uses numalink. It uses the older 1.6GBps/dir links, but uses 32 networks in parallel for a total of ~50GB/s/dir per node.

The Cray XT3 uses a newer network interconnect called seastar, which offers 3.8GBps/direction. This is probably what will be used in the X1's successor.

The Cray XD1, which your colleague bought, is a product cray acquired when they bought OctigaBay. They use an interconnect called the RappidArray switch, which provides 4GBps/direction of interconnect.

All of these interconnects are high-bandwidth and low latency. The XD1, is also very inexpensive for a cray, which is always nice.

Re:excellent

[flaming-opus]

No they won't! They have no reason to. The vector units that a cray uses aren't like altivec, sse, or other "bolt-on" vector units. The vector unit on a cray (or NEC) is a latency hiding mechanism. It's a method for forcing the programmer/compiler to structure the code such that the data loaded from memory is used a significant period of time after the load is initiated. This works pretty well on the HPC code that is used on crays, but not at all for the everyday server/workstation code that opterons run. Furthermore, to support that sort of vector unit, you need to have about eight times as much memory bandwidth as an opteron, which means many more pins on the socket, which are very expensive.

I think you're much more likely to see the cray vector processor retooled with lots of hypertransport connections, so it can use an opteron as its scalar unit, and use the same seastar routers that the xt3 uses. On the X1, the scalar unit already runs ahead of the vector unit, so I bet it's not all that important for the scalar unit to be on-die.

Re:excellent

[Frumious Wombat]

So, just curious, but why don't you put your efforts into crunching your traditional vector processors onto some more affordable piece of silicon, rather than trying to recreate the T3E out of Opterons?

I know chemists who claim there there are still algorithms than don't run as well on modern MPars as they did on mid-90s vector Crays. I know we're not a huge market, but I bet there are some other fields that would rather have a deskside T90, rather than a multi-proc Opteron box.

Re:excellent

[joib]

No they won't! They have no reason to.


Yes, you're probably right that it doesn't make sense for AMD economically. But I want to run numerical codes at more than 5 % peak performance on my cheap Opterons, so I want to believe. ;-)


The vector units that a cray uses aren't like altivec, sse, or other "bolt-on" vector units. The vector unit on a cray (or NEC) is a latency hiding mechanism. It's a method for forcing the programmer/compiler to structure the code such that the data loaded from memory is used a significant period of time after the load is initiated.


Yes, I know. And that's precisely the reason why I'd like to see real vectors instead of the sse/altivec toy ones. Main memory latency is hundreds of cycles, and it's getting worse all the time.

Additionally, from a microarchitecture perspective, vectors have quite a few advantages there too.


This works pretty well on the HPC code that is used on crays, but not at all for the everyday server/workstation code that opterons run.


I'm not sure about that. I guess technical apps vectorize just as well as HPC codes (well perhaps not the UI, but the code that runs the actual simulation or whatever). Heck, even some database code vectorizes nicely (sorting and hash joins).


Furthermore, to support that sort of vector unit, you need to have about eight times as much memory bandwidth as an opteron, which means many more pins on the socket, which are very expensive.


Yes, as I said some Alpha Tarantula like design is probably overkill for the vast majority of the market. My point was that a vector ISA extension with modest execution resources wouldn't need that much die area, and could help make better use of the available bandwidth, whatever that bandwidth is. As you said yourself, the expensive thing is IO. Transistors are cheap by comparison. So not having instructions that allow one to effectively use the available IO resources is a real shame.


I think you're much more likely to see the cray vector processor retooled with lots of hypertransport connections, so it can use an opteron as its scalar unit, and use the same seastar routers that the xt3 uses. On the X1, the scalar unit already runs ahead of the vector unit, so I bet it's not all that important for the scalar unit to be on-die.


Yes, that sounds feasible. IIRC it is something like this that Cray has cooked up for the Cascade project; I.e. a node consists of 8 (or was it 4) scalar processors connected to memory (I guess these could be Opterons or further in the future some kind of Processor-in-memory (PIM) stuff), and a vector unit with its own cache and fast access to the main memory via the scalar cpu:s.

As for the seastar thing, I think you're right that that's what they'll use for inter-node communication. Currently X1(E) uses Numalink licenced from SGI, so they're certainly looking at replacing that with existing in-house tech. BTW, 2H2006 will see the XT4, with the new Opteron sockets with DDR2 memory and the Seastar2 router that provides twice the BW compared to the existing Seastar.

It only makes sense

[heatdeath]

Specialized computing hardware for supercomputers has always seemed like a fiscally bad choice. It'll be good to see what kinds of improvements we can see in research possibilities as supercomputing costs come down from using mass-marketed parts.

Re:It only makes sense

[fgodfrey]

I've never completely understood this argument (yes, I admit, I'm heavily biased). If I want to build a skyscraper, I'm not going to use the "mass market" crane that puts up the roof of a residential house. I'm going to use a specialized crane that's meant for building skyscrapers.

That doesn't mean that there isn't a place for commodity hardware in supercomputing, but to say that there's no room for custom hardware either misses the point. The only thing "off the shelf" about an AMD based Cray is the AMD. The logic board, and, most importantly, the network that interconnects the processors is entirely custom. Not to mention the fact that Cray will still build some entirely custom processors...

By the way - this is hardly the first Cray based on a commodity processor. The T3E and T3D were both Alpha processors, yet nobody calls those machines "commodity".

Re:It only makes sense

[tzot]

Slashdot readership is not an elite, it's just an assembly of geeks. The up-modding of your article just means that some random readers found it insightful; nothing more, nothing less. It's not a medal or a decoration; judging just by the difference in our /. IDs I would assume you would know by now ;-)

And moderating is voting.

Re:Had to be done

[nnnneedles]

Had to be done

No, it didn't.

Can we have a "-1, Catch phrase" option, please? The old jokes are not even remotely funny anymore..srsly.

Re:But

[fgodfrey]

Yes, actually.

Sign You Invested In The Wrong Supercomputer, #342

[John_Booty]

#34. Your "supercomputing" vendor has an AOL email address.

From the press release...

Contact:
AMD
Teresa Osborne, 512-602-0040
teresa.osborne@amd.com
www.amd.com
or
Cray Inc.
Steve Conway, 651-592-7441
sttico@aol.com
www.cray.com

Sooooo... if I scrape together a few million bucks and buy a computer from these guys, will I still be able to contact my Cray rep once his 500 FREE TRY AOL NOW HOURS have expired?

Re:Put on your tinfoil hats...

[B3ryllium]

DARPA created the Internet.

You are using the Internet.

You are part of the conspiracy.

Re:...and now some real numbers

[Vengeance]

Hrmmm. In six months AMD went from 25 systems on the list to 55 systems on the list, and you think Intel is doing well?

Let's extrapolate for a moment, shall we? I'll even do Intel a favor and clamp down on the AMD increases each time. Basically, AMD more than doubled their share of this elite group in six months' time.

Six months from now, they've almost doubled to 100 systems.

Twelve months from now, slowing down and growing only 75%, they've got 175 systems.

Two years in the future, with even more slowing down of their growth, 300 systems on the list are AMD. I wonder whether the preponderance of that growth comes from the current 400-odd Intel machines or from the 73 IBM setups...

Likely? Maybe not. Possible? Yeah, it just might be.

nVidia motherboards

[CCFreak2K]

Hey, maybe the motherboards are nForces too. I bet all the new Crays will have digital 5.1 sound, an important feature for today's supercomputers.

Re:...and now some real numbers

[hattig]

On November 2004 the list had 84 computers with Itanium 2 processors. In June 2005, the number shrunk to 79.

Now only 46 computers contain Itanium 2 chips according to the latest list, released Monday.

That's quite a collapse. Intel is propping up their high-end systems with volcano-simulator Xeons?

Meanwhile, the number of supercomputers using Advanced Micro Devices' Opteron chips has increased. A total of 55 Opteron-based computers made the list, up from 25 in June. (Opterons were found in just 29 computers on the November 2004 list.)

A near doubling in a year. And that's with AMD's first real server standard processor. HORUS comes out today, that'll put AMD into the 32 and 64 core marketplace. Not bad for a company with 0 server marketshare, nevermind Top500 systems two years ago.

As for the rest of your troll, I think most of the people here are clever enough to see it for what it is.

The irony of it all.

[Anonymous Coward]

Crays supercomputers were known for their high performance vector operations. These operation have very little to do with PC world except its close cousin - SIMD operations (gaming, graphics). Now the fact that AMD tops cray (at least on commercial merits) is like having an AMD instruction set adopted by Intel (oh, wait).
More ironic is the fact that the compiler that will be used for those supercomputers is probably the PathScale variant of Open64 - SGI's compiler that was released as open source after it was retargetted to the Itanic architecture.
I might have some misconceptions, careful readers, please fill-in the blanks.

Re:The irony of it all.

[Shoeler]

Aren't you missing the point though? It's not that the AMD systems are prevailing on the merits of their amazing vector math (they aren't) - it's that they do a PRETTY good job of both vector and scalar math, but at the prices you can get them, your cost per computation is SIGNIFICANTLY lower then it would be with one of the massive vector systems.

The research / development arm of the organization I work for just got a 4000+ CPU XT3. Last I checked, they planned on using the PGI compiler for most stuff.

Re:The irony of it all.

[convolvatron]

disclaimer: i worked at cray on the xt3

its not ironic at all. its a question of resources and volume.
cray has a few very bright people (still, sort of). they are
essentially a us government lab. they do a bad job, but its insane
to think that 100 people can build and maintain several different
supercomputer architectures.

a $300 opteron is almost always more effective than a $60000 X1
processor. they have alot of bright people too, and alot more of
them.

the only reason that cray still exists is support for parallelism
and the provision of high memory bandwidth systems. but even that
niche is being eroded pretty severely. the xt3 communications chip
runs at 3.5GB/s in each direction. it costs about $250 for cray to
have each of them made. for the same $250 i can buy a mellanox nic
that runs at half the speed

its no suprise that cray is using opterons. they actually got lucky
by committing to amd early and having it turn out so well.

the real question is whether there is any more room for a cray at
all. the commodity world moves so quickly. the xd machines (which
they purchased) are really their best asset, but it still hard to
justify that kind of margin for what is essentially a well
constructed cluster.

Opteron is not NexGen's tech

[tomcres]

K6 technology was acquired and modified by AMD. The K7 and K8 were designed by AMD. True, many of the engineers on the K7 and K8 teams were probably ex-NexGen since AMD acquired that company, but so what? They are truly AMD innovations. At least they didn't sink all of their research into the Itanic!

Re:Opteron is not NexGen's tech

[sgt scrub]

Actually the K7 was designed with most of the DEC technology. I'm not knocking AMD though. I've used them since the 386 days. I think they have surpassed the Alpha --me ducks-- in speed and efficiency, though I still wouldn't give up my 433a withought a fist fight.

Their bus arch and chipset tech is the most interesting. (if someone has proof that AMD didn't design this it better be solid). This attention from Cray, and the super computer people in general, is due more to this success. AMD has the best design and it shows. It is one thing to buy schetches of a something and another to make it fly this good.

More to the point regarding Cray is their XD1. THAT is a cool machine! I was looking around at different FPGA stuff and almost shorted my keyboard with drool. Damn, I wish I was rich. -sniffle-

Finally....

[Prince Vegeta SSJ4]

A Computer capable of running Duke Nukem Forever....oh wait...

The Chipset is the key.

[mosel-saar-ruwer]

Specialized computing hardware for supercomputers has always seemed like a fiscally bad choice. It'll be good to see what kinds of improvements we can see in research possibilities as supercomputing costs come down from using mass-marketed parts.

Cray likes to build classical vector-driven machines. In that space, you can't rely on some external kludge like Myrinet for your communications; instead, your value-add is in the chipsets that get all those CPUs talking to one another [and to the memory subsystem].

In one of Cray's previous incarnations, they once possessed a chipset/backplane tech for the Sparc processor that Sun purchased off of Silicon Graphics for a song and a dance, and immediately turned into the insanely profitable Sunfire series. The big question here is whether this new agreement requires Cray to share their chipset/backplane tech with AMD [in which case some of it might filter its way back down to the level where mere plebians like us would be able to afford it].

It had to be said...

[rubberbando]

Finally! Something that can run Windows Vista at a descent speed... :-P

Re:...and now some real numbers

[Anonymous Coward]

Wow, seems Intel is still 6x AMD market share.

Wow, seems AMD *doubled* it's share of spots in the Top 500 list in *six months*. I bet Intel is ticked, and worried...this is very good PR for AMD.

Go AMD! Milk that NexGen core for all its worth, too bad you didn't invent it, you just bought it.

LOL! Intel fanboys don't have anything real to say these days, they have to resort to cheap ad-hominems. Don't worry, I'm sure someday Intel will come out with competitive chips again. Pretty sure, anyhow.

And as to AMD "just buying it", how would that relate to Intel getting so much Alpha technology and talent from it's deals with HP/Compaq/DEC?

It would be nice if you would start innovating one of these days.

Yeah, if AMD can produce better processors than Intel without innovating, just imagine what'll happen when it does innovate...! =)

Behold!

[katana]

I give you...the Crapteron!

Re:Behold!

[sharkey]

It comes with Vista pre-installed then?

Does this mean...

[__aaclcg7560]

That prices on the older Clay computers will drop? Holy flipflops! Now I'll have something to put into my empty warehouse building. :P

Not to be a jerk...but this is old news....

[jamesgomez]

dated from June 16, 2005

Check out the article here...
http://www.hypertransport.org/consortium/cons_pres srelease.cfm?RecordID=79/

Sun..

[Bulmakau]

I was just about to buy 40 sun machines, based on AMD. Maybe I should wait for Cray to come out with their product? Anyone knows the estimated retail price that machine is expected to hit the market with? ;)

Re:Sun..

[iamlucky13]

I know you're trying hard to make it look this is work-related, but for goodness sake, don't make major purchasing decisions based on what you read on Slashdot!

Re:Hmmm

[turgid]

I wonder what the governmnet will do with these cheaper, powerful supercomputers?

Why, decrypt your 66,000ft high stack of SSH traffic in case you're a would-be terrorist, of course! :-)

The answer is quite simple

[drgonzo59]

If they think they'll make more money in the long run, all things considered, they'll do it.

Re:Had to be done

[P3NIS_CLEAVER]

In Korea old jokes aren't funny anymore.

Re:Put on your tinfoil hats...

[Neoprofin]

To be fair, AMD is proud of helping spy on it's citizens more efficiently and we can all be proud of increased productivity. If that wasn't funny maybe it was insightful?

Re:Compilers !! ???

[cimetmc]

Have you tried Pathscale compilers http://www.pathscale.com/ekopath.html? They seem to give quite decent performance on AMD64 chips.

Re:why not use the ultrasparc T1?

[NerveGas]

They have 8 cores!

So? How much memory bandwidth do they have? Not I/O bandwidth, but memory bandwidth. I highly doubt that they have as much bandwidth PER CORE as the Opterons do, and in big applications, memory bandwidth can be a very important factor.

You cant build an enterprise machine without Ultrasparc (or Power4 or PA_RISC) CPUs.

    I guess that Cray thinks differently.

Makes me wonder ...

[cpu_fusion]

The continued big-name backing of AMD (e.g. Sun, Cray) makes me wonder how sweet a deal Apple must have gotten to go with Intel over AMD. :)

Re:Makes me wonder ...

[Watts Martin]

My impression is that Intel has been irritated that they come up with new technologies that the PC world isn't particularly interested in. Despite the "Wintel" moniker, Intel has had a love-hate relationship with Microsoft; back in the BeOS days, Intel offered engineering assistance (and I suspect financial support) directly to Be to get the OS ported. And today, Apple doesn't have 20 years of backward-compatibility hardware baggage to deal with, so they have the potential to be a showcase partner in a way that the Dells of the world just aren't.

As for why not AMD, though, Intel has placed a much higher focus than AMD has on very low-voltage chips, and from what I've heard, that's what ultimately gave them the Apple nod. Arguments about production capacity aside, AMD doesn't have the R&D resources of Intel, and they have to pick their battlegrounds carefully. They've picked them wisely, but as of right now, they don't have anything competing with chips like Intel's low-power, dual-core lineup for 2006. If that's remedied in 2007, I'm sure Apple would have no problem revisiting it, but right now, AMD just doesn't have the chips they want.

(As for the production capacity arguments, I've seen people here point out that Intel has had problems meeting their demand recently and AMD hasn't. While this is true, it's important to keep in mind that Intel's overall demand is still over five times that of AMD's, and the gap in notebooks -- the segment where Intel's production capacity fell behind demand temporarily -- favors Intel by an even greater margin.)

AMD makes more than chips!

[scoobrs]

The parent poster needs to be reminded that a large chip manufacturer like Intel, IBM, and AMD makes much more than CPUs! They play a fundamental role in the design and system architecture of the machines built out of their chips. Interfaces like Hypertransport, PCI Express, and DDR are the work of these chip giants. To claim that changing the fundamental design of the CPUs has anything to do with the interaction of a supercomputer company and AMD is naive. Far more likely are changes in Hypertransport, interfaces to memory, or other bus-level projects that are more useful to a supercomputer vendor looking for the best possible overall system bandwidth anyway.

Re:When does this translate to bank?

[jimicus]

Stock prices seldom reflect what the company is actually doing, and are more often driven by either the market in general or mindless "analysts" who wouldn't know what AMD does if a chip were to bite them in the bum.

Re:why not use the ultrasparc T1?

[turgid]

Because they don't do floating-point in hardware, or at least not to any useful level of performance.

The 8-core Niagara (T1) has 1 floating-point execution unit on only 1 of the 8 cores. Buy a 6- or 4-core Niagara, and do you get a floating-point execution unit at all?

On Niagara (aka UltraSPARC T1) floating-point will mostly be accomplished with software emulation of the SPARC V9 FP instructions.

That's why you wouldn't use Niagara for supercomputing. Web serving, yes, computational fluid dynamics or numerical general relativity, no.

The Irony of Marketing

[scoobrs]

The parent poster posts well for one ignorant of the simplest precepts of marketing. The first things a marketer learns is he must segment a market and only compete in the segments or niches in which competition is profitable. Cray isn't competing directly against clusters because clusters don't have the bandwidth necessary for the sorts of problems Crays are aimed at and Crays tend to be overkill for the problems clusters are aimed at. Cray doesn't seek out customers $.5M for that reason. Anyone who actually uses the supercomputers to solve problems knows that a 50% difference in interconnect speed per single link could mean a 90+% slowdown on a large system using a large program with high overhead. Plain old clusters aren't targetted against Crays, except by some communities that don't buy supercomputers for supercomputer problems anyway, like most Slashdot users. In the supercomputer world, MTTI is everything! That means mean time to interrupt. A bad memory module or a CPU fan blowing out on your single CPU might happen every 3 years on average, but multiply these sorts of problems by 10,000 CPUs on a supercomputer and your cluster will never get any useful work done before something goes out and it crashes. Disclaimer: I worked on the X1/X1e, which is still faster than any other chip on select problems which vectorize well. I agree that the AMD partnership was and continues to be an excellent decision, but it only says that AMD does SCALAR performance better, not everything!