First 96-Node Desktop Cluster Ships

Panaphonix writes "The Register reports that Orion Multisystems is shipping the first 96-node desktop cluster. 'With the new, larger system, customers get pretty much the most powerful computer around that can plug into a standard electrical socket.' According to the spec sheet, the DS-96 runs Fedora Core 2 and gets 110 GFlops sustained, 230 GFlops peak."

Comments

Imagine a Beowulf...

[Anonymous Coward]

Oh, never mind.

I FAIL IT!

$100,000

[J_Omega]

Subject says it all.

Wait, one of the partners is The MathWorks? Creators of MATLAB?!! I'm in!

Re:$100,000

[themoodykid]

If you've ever had to work with MATLAB, you'd know that it needs as much horsepower as it can get!

Re:$100,000

[_merlin]

MATLAB is also memory-intensive, too. I've seen it die from lack of RAM too many times.

Re:$100,000

[raehl]

$100,000/96 = ~$1,000.

Not a bad deal.

Re:$100,000

[should_be_linear]

Actually, too expensive. When building cluster of stock PCs it should not be more then $500/PC. And you get AMD64, not Transmeta CPUs (double performance). That being said, kudos to Orion for shrinking whole cluster to big-tower case. Drawback is that it's not so easy to replace dead CPUs (yes, I know it *is* possible).

Re:$100,000

[tph]

If you've ever had to work with MATLAB, you'd know that it depends heavily on HOW you write your code. Sure, for-loops and such are slow, but vector and matrix operations are fast!

Re:$100,000

[dark-br]

100k for 96 processors? Figure you can get a barebones system with 256 MB ram for around $250. That's $24k for the boxes, a 96-port switch, and some good clustering software.

Where's the rest of the cost coming from?

Re:$100,000

[BlueHands]

Your totally right that you can get more, but whats the power/wattage difference? I can not envision 96 AMD64s being so easily powered as this system is. It seems to me that they are not competing on price, but size and power consumption.

Re:$100,000

[Donny Smith]

>When building cluster of stock PCs it should not be more then $500/PC.

What are you talking about?

The now (in)famous Apple cluster cost them about 5 million for 1,100 nodes or $5K/node.
http://radio.weblogs.com/0112083/G5cluster.html
And that was supposed to be a good deal.

Re:$100,000

[should_be_linear]

you need to add $3000 for Air Conditioning and $5000 for furniture for compueters and switches.

Re:$100,000

[Boogiesbunny]

Generic stock PCs are a piecemeal of various secondrate vendors, that crap needs to stay cheap. Comparing the value of Cluster nodes to PCs is like comparing homecooked dining to fastfood, one is healthier and other sends you running to the tidy bowl man. Remember you get what you pay for.

Re:$100,000

[Boogiesbunny]

That is a 150lb paperweight. It reminds me of the T.V. VCR combo. Once the novelty wears out and a couple of processors die, you will be out of luck.

Re:$100,000

[Mistah Blue]

Oh I don't know, how about: design, development, support, etc. Obviously you aren't the target market if you are quibbling over the price per processor. They charge $100K for it. Take it or leave it.

Re:$100,000

[Grayputer]

How about "plugs into a single 110 volt US outlet". This thing draws 1500 watts PEAK. That's about 14 amps for the math challenged or well under the recommended max for a 20-amp circuit from your household panel (think coffee maker on steroids). Let's try your barebones system approach and let's say you tweak it to use 50 watts per system (good luck). That's 50 watts per system multiplied by roughly 100 systems => 5Kwatts or 3+ times the power consumption. PLUS as an added bonus you get 96 cases, external cabling, at least 96 fans, a dozen power strips, and assorted other toys to trip over. Oh yeah and a few weeks of setup, integration, burn in, and testing.

Re:$100,000

[trentblase]

And watch the power grid fail when all the drives spin up at once.

Re:$100,000

[PingPongBoy]

1500 W is about the same power as my hair dryer. That's quite a heater on a desk.

This reminds me, two years ago I met an insurance broker who ran his cube server beside his desk. It sounded like a hair dryer on full blast but it never fazed him to work all day like that.

mmm clusters

[Worminater]

My uni recently got some a 12 system dell cluster that came loaded with redhat. mmm; paralizing is fun:)

dup

[pg133]

96 Processors Under Your Desktop

Re:dup

[fenodyree]

Not a dup, the story from August reported the intent to ship a then non-existent product and the original ship date was October as TFA reads. This article is about the actual shipment of the product.

Re:dup

[strider44]

an article saying "this new technology is here" isn't a dupe of an article saying "this new technology is scheduled to be coming out in a few months".

This is a notification that they finally released it. By your post I know it's 7 months later than scheduled.

Re:dup

[indifferent children]

If /. carried a story saying that Duke Nuk'em Forever had been released, somebody would bitch that it was a dupe.

Will it make or break orion?

[nate85]

Do you think it will go over well?

Give it to me in terms I can understand.

[spineboy]

Like how many SETI units it can do in an hour, or like how fast it can spell check my Microsoft word document.

Re:Give it to me in terms I can understand.

[spectrokid]

You want to be impressed heh? Well, it is powerfull enough to play solitaire on Longhorn!!!!

Re:Give it to me in terms I can understand.

[Anonymous Coward]

i doubt it, you microsoft astroturfer.

Re:Give it to me in terms I can understand.

[William Robinson]

You might want go through this

Re:Give it to me in terms I can understand.

[jaavaaguru]

or...
How quickly can it start OpenOffice.org?

Re:Give it to me in terms I can understand.

[Anonymous Coward]

I didn't think that SETI@Home units were all the same size?

Anyway, the best metric is how fast it can count up to infinity,

Bonus points if it deduces the existance of rice pudding and income tax.

Re:Give it to me in terms I can understand.

[nametaken]

Or powerful enough to build Gentoo in under 48 hours? :)

Now THAT would be useful.

Re:yeah

[Anonymous Coward]

No, it runs an inferior OS. I can't remember what it's called.

Question

[ErichTheWebGuy]

I am not familiar with the architecture of clusters, so I am a little surprised by the more than 100% difference between sustained and peak GFlops. I know what a GFlop is and all that, I just don't immediately see why there is such a huge difference.

Can someone summarize why there is such a huge difference?

Re:Question

[lachlan76]

I don't know much about it, but my guess is that it would need time to allow some of the processors to cool down.

Re:Question

[bobbozzo]

finite bandwidth between processors makes it impossible to sustain anywhere near peak performance for most real-world applications.

Linpack is what is usually used to measure sustained performance on HPC systems.

Re:Question

[katana]

Sustained: Riding down the street.
Peak: Taking it off some sweet jumps.

Re:Question

[Ruie]

You are right - on pure number crunching, with little I/O the cluster peak and sustained performance for the same program should be identical.

It might be that their peak number is derived assuming code particularly favorable to the processor architecture in use - say using SSE to do the floating point math. This can easily produce the factor of 2 difference.

Re:Question

[brsmith4]

The theoretical max gives a rough estimate of the raw Floating-point power for all of the processors on the system. You pretty much add up the GFlops potental for each node (not exactly, but pretty much). The sustained and demonstrated GFlops of the cluster is based on the Linpack benchmark. The reason there is such a huge difference between the two numbers can be a result of a few factors. 1) The interconnect is GigE and Linpack tends to make use of Message passing comms which are affected adversely by the latencies produced by the GigE connections (myrinet would have been a good choice, but I suppose it was probably impossible to squeeze that into that case) 2) Memory speeds also are a factor as pushing floating point numbers around involves memory. This cluster isn't using anything fancy when it comes to the memory and I suspect this may be another cause for this.

When they say that this line of clusters can "make or break" Orion, I am right now, leaning for broke. For the cost of this machine, one can get a real cluster with a lot more performance. I know this thing is nice because of the power requirements and the fact that you don't need a dedicated server room to store it, but for $100,000, you can get Microway to build you a pimptacular cluster with Dual-Opteron nodes, high-speed memory and a phat interconnect with either myrinet or infiniband. You will get a lot more work done for the same price.

Re:Question

[Daniel Phillips]

I know this thing is nice because of the power requirements and the fact that you don't need a dedicated server room to store it, but for $100,000, you can get Microway to build you a pimptacular cluster with Dual-Opteron nodes, high-speed memory and a phat interconnect with either myrinet or infiniband. You will get a lot more work done for the same price.

You forgot a couple of things:

* HVAC costs
* Realestate costs

Remember, this is a deskside cluster. Try that with your dual-opteron cluster. And try adding up all the costs.

Re:Question

[Emil Brink]

Not to shoot you down or anything, but it makes me giggle a bit that you say "I know what a GFlop is and all that", heh. As you then know, the basic abbreviation means "floating point operations per second". So, the last 's' is not to make it plural, it's part of the acronym. You can't say "a GFLOP", it's "a GFLOPS". The difference between 1 GFLOPS and 2 GFLOPS is 1 GFLOPS, not 1 GFLOP. And so on. This is one of my big zoo of pet peeves, not meant as a personal attack at all. :)

Re:Question

[pixel.jonah]

You're right, but the post you're commenting on is also technically correct.

A GFlop would be one billion floating-point operations. No telling how long they would take though. (i.e. a GFloph machine might take one hour to do a billion floating-point operations.) Much less meaningful, but still valid.

It's like horsepower. "One horsepower is the amount of work necessary to lift 33,000 lbs. one foot in one minute."

My pet mouse could lift 33,000 lbs. one foot with enough gearing on his exercize wheel. Now, it might take him years to do it. OTOH, my horse can lift that same weight the same distance, but will need much less gearing to do so.

Re:Question

[glesga_kiss]

He's also forgotten about hiring an extra 2 guys at least to maintain the extra machines. A selling point of this box will be that "it just works". Pay for a support contract and wammo, you've got a cheap low maintainence cluster. For people working on top-secret stuff (who else needs clusters? ;-), hiring people is a risk and the vetting process is expensive.

Re:Question

[Hannes Eriksson]

Give or take a few megabyte of cache per node, it wouldn't be too hard for atlas to do something about the weird choice of interconnect. For some problem sizes. But I guesstimate that these nodes don't have 4MB level 2 cache.
I for one wouldn't buy one of Orions desktop clusters. I'd say most all research on latency effects of running GbE (not really that interresting) could be done on regular workstations running LAM (the desktop cluster in question also uses LAM).
I'm not against Linux clusters. I'm using those for my studies. I'm just sceptical of using GbE as interconnect in anything that costs more than $10,000. I mean, it's not impossible to get to the Top100 with $1,000,000 (it's been done), and that doesn't allow for GbE interconnect...

Re:Question

[Daniel Phillips]

I'm not against Linux clusters. I'm using those for my studies. I'm just sceptical of using GbE as interconnect in anything that costs more than $10,000.

It uses 10GigE on the backplane. For good measure it has something like 80 GB of disk on each node.

Re:Question

[Vo0k]

I'm surprised with more than 200% difference between the price of the cluster and equivalent in separate desktop computers.

Re:Question

[Big Mark]

There's more to making a cluster computer than wiring a load of Dells together with Cat-5, you know.

Re:Question

[coopex]

A post on slashdot that isn't about how he could do a much better/faster/cheaper job that what's in the article?!? My god, it must be the end of the world!

That said, to the GP, if you've ever taken a computer architecture course, you'd know that supercomputer makers bend over backwords, jump through hoops, and thread camels through needles to get half the theoretical max performance.

Re:Question

[Hannes Eriksson]

I'm not against Linux clusters. I'm using those for my studies. I'm just sceptical of using GbE as interconnect in anything that costs more than $10,000.

It uses 10GigE on the backplane. For good measure it has something like 80 GB of disk on each node.

Even with "10GigE" I wouldn't expect an MPI barrier to take less than 10 us. Ethernet just isn't designed for low-latency applications. Most parallel applications send lots of messages (Like Cannon's algorithm for parallel matrix multiplication sending 2+2*sqrt(p) blocks per node for p nodes, plus overhead for data distribution and collection). If they don't send a lot of messages, you could probably do the computations cheaper and faster with grid computing or similar techniques like distributed.net, *@home etc.

Oh, and about 80GB of storage per node: how long would it take to do a simple checksum on that? Memory bandwith is another key issue in parallel computing that I doubt Transmeta is really, really, really good at.

Sorry if I ruin someones dream. It might be that this desktop cluster really has a valid application, but instead of seeing it, I see someone trying to ship oil inter-continent by transporting oil-filled balloons in VW Beetles...

It does have a flashy case, though. Somehow it reminds me of the Toaster SGI (O2, that is).

Re:Question

[Tim C]

The vetting process isn't just expensive, it's time consuming too. I don't know about the States, but here in the UK getting security cleared (to "normal" level, I forget the correct term) takes about 3 months. That's not 3 months of filling in forms, of course, but it is three months during which the person can't work on the project while you wait.

Re:Question

[Anonymous Coward]

Space (e.g.real estate) and energy(electricity) is a premium in places such as Japan. Could be interesting to see how sales develop in such places.

Re:Question

[Daniel Phillips]

Oh, and about 80GB of storage per node: how long would it take to do a simple checksum on that?

Huh?

Re:Question

[Daniel Phillips]

Sorry if I ruin someones dream. It might be that this desktop cluster really has a valid application, but instead of seeing it, I see someone trying to ship oil inter-continent by transporting oil-filled balloons in VW Beetles...

Don't worry, you did not ruin anybody's dream. You bleated about 10 usec network latency and otherwise don't seem to have a clue.

Re:Question

[brsmith4]

You don't need two extra guys. I manage 5 clusters myself on a regular basis. Granted, it would be a hell of a lot of work if I couldn't enlist the help of a few people every once in a while, but for the most part, its a one man job. As well, to the poster claiming that I didn't take into account the cost of storage, I know this thing is nice because of the power requirements and the fact that you don't need a dedicated server room to store it, I'm fairly sure that I roughly addressed that. BTW, we need clusters and none of our researchers has top secret clearance ;-P

Re:Question

[Mad Hughagi]

Bonus points for the Napolean Dynamite reference!!
You gonna eat those tots?

Peak is determined by having every floating point unit active during every clock. It is the theoretical maximum number of operations that you could ever possibly do. No program (other than making heat in a tight loop?) ever comes close to this.

Sustained is arbitrary. Basically you run a program that is optimized to avoid un-necessary cache misses and that produces some meaningful result (like the matrix multiply in linpack). Real world apps don't even come close to this level of effeciency, often having to wait for data to be sent from RAM or disk, which slows down the system even further (stalls the pipeline).

Re:Question

[fshalor]

When the acryonm is used as a unit of measure, it's a single quantity. The parent it correct in it's "standard and accepted use" as a unit of measure. You're correct in that if the acronym is parsed fully, it's 1 gfpops and not 1 gflop.

Either way, this reminds me of the "sgi" days.

Howver, it is a lot of power for the money and it's a clean solution to a very unclean problem.

I applaud orion for it. This would be the ideal solution for mobile cluster needs.

I can immediatly see applicatins in the filed, and on oil rigs and such where they need to do rappid model scaling of polutant distributions. Or processing lots of flow data against a model.

I've got tons of rainfall and flow models here which can't run on a single PC in less than several hours which would be eaten up in a few minuts on this thing.

Re:Question

[femtoguy]

This point is a biggy in the scientific computing world. It is easier to get capitol equipment money than it is to get salary money. (This is because equipment money is overhead free, while salary money incurrs a 50% overhead rate). We just had a donor give us many millions to guy a cluster, but he would commit to long term money for sys-admin support. We ended up including a lot of vendor support into the bid for the contract in order to turn support money into capitol equipment money. Considering that a sys-admin can easily cost $100K per year, this isn't such a bad deal.

Re:Question

[dillon_rinker]

From where I sit, it takes 6-24 months to clear. Until you clear you're practically useless.

Re:Question

[cybernautix]

who else needs clusters?

• Biostatistical Researchers (here)
• Genomic Researchers (down the hall)
• Atmospheric Researchers (across the street)
• Seismic Researchers (upper campus)
• Fluid Dynamicists (at Boeing)
• ...etc.

And I spend an average of four hours a week maintaining two clusters...but not with those GFlops.

Wrong term

[AndrewStephens]

Its not duped, this article has been clustered.

Re:Wrong term

[bobcat7677]

OMG! I think the parent is the funniest thing I've seen on Slashdot in months. Rock on! Mod on!

Re:Just imagine...

[Anonymous Coward]

At 150lbs or 68KG you need a Beowulf cluster of people to move it!

Still The Wrong term

[Johan Veenstra]

It's called a follow-up:

"In October, you'll be able to choose between"

"is shipping the first 96-node desktop cluster. '"

whoa!

[sonictheboom]

Tuxracer on one of those !

Re:whoa!

[suitepotato]

Oh good, I thought it was just the cold medication making it seem that frigging slow on my brand new machine. Though the real question is, will it be on the supported list for Duke Nukem?

Re:Fedora Core 2?

[lachlan76]

Because they started development before FC3 came out maybe?

Great for researchers

[_merlin]

This will be great for researchers with CPU-hungry simulations to run. A small box with a lot of grunt is exactly what you want when you're simulating the PHY layer for your 802.11n proposal.

Re:Great for researchers

[DrSkwid]

state
the
bleedin'
obvious

Practical measures...

[crottsma]

With power requirements quintupling that of a standard desktop computer, I'd probably have to use it at my local coffee shop, or only turn it on briefly to scare away song birds.

Re:Practical measures...

[Daniel Phillips]

With power requirements quintupling that of a standard desktop computer, I'd probably have to use it at my local coffee shop, or only turn it on briefly to scare away song birds.

The 12 node version uses a 250 watt power supply, which runs 12 disks and 24 GB of memory too.

Cool . . .

[ElectusUnum]

Can it play tetris?

Re:Cool . . .

[mrjb]

Actually with this kind of power, what really matters is if it plays tic-tac-toe. Or a nice game of chess.

More Importantly...

[mtrisk]

Will this make or break Transmeta? It uses their processors (Transmeta Tinside as the Register calls it). Slashdot already pronounced the death of Transmeta though (it has no more niches!), maybe this could revive interest?

Re:More Importantly...

[nate85]

It would have been much better using PowerPC chips...

Re:More Importantly...

[jericho4.0]

Then you wouldn't be able to plug it into a wall socket.

Re:More Importantly...

[grendel_x86]

The problem w/ the PowerPC procs is the energy requirement. You would never get even 48 power4 processors(g4-g5) to run at the same energy requirements.

That said, it would be a crap-load more powerful.

Re:More Importantly...

[imroy]

Erm, no. The G5 (from IBM) is power hungry. The G4 (from Motorola/Freescale) is not. The Mac mini has a 1.xGHz G4 in it and only needs one little fan, which is mostly for the Radeon graphics chip anyway. So you can put a G3/G4 in a CDROM-sized enclosure with 3.5" hard disk, or put 4 processors on a single PCI card.

Underwhelmed

[Eunuch]

Isn't Cell supposed to hit a teraflop or something? And we should be using binary milestones like a 2^32-flop.

Re:Underwhelmed

[Gil-galad55]

Why on earth would you use a binary milestone for something like a performance metric? Measuring memory, storage, etc. in binary format makes sense as it is byte addressable. Measuring other quantities in anything but decimal is adds complexity.

Re:Standard comments

[jim_v2000]

You forgot to mention:

12. List of common Slashdot posts

Re:Standard comments

[thegamerformelyknown]

No, you missed something about "I, for one, welcome our new -insert comment here- overloards"

Interesting practical uses for Personal Cluster ..

[torpor]

.. Computing?

I mean, does Blender run on it at least? Can I do anything interesting from an 'immediate-personal' perspective with 96 nodes, and I don't just mean run Quake, or fire up "make -j 96" and such things..

What sort of interesting modelling software is around? Could I use it to design stuff on a personal, non-hard-core science perspective? What are the practical uses for personal cluster computing?

Needs silencing!

[Slowleggs]

From TA "Sound power 55 bels"

550 dBel noise? Perhaps the producers should look into Metal cooling ? :)

...and/or put the box in another room.

Re:Needs silencing!

[mrjb]

Your observation is correct, that obviously must be a mistake. Among sounds heard by humans, the firing of military rifles can reach 150 dB. Sounds this powerful can apparently break bones in the ear, and so I assume would have to be the loudest sounds that we can hear. 550 dB would be just ever so slightly over the line.

Re:Needs silencing!

[hcdejong]

The loudest sounds possible (at 1 bar ambient air pressure) are about 180 dB. At that level, the pressure minimum of the sound wave is 0 bar (ie vacuum).

Re:Needs silencing!

[madaxe42]

Also, bearing in mind that the bel scale is logarithmic - 550dBel is enough to probably tear a chunk out of the planet - imagine nuclear detonation shockwave scale noise!

Hmm... 55db in W/m^2 is about... 1x10^30. Yeah, that's a lot of energy. We could use these as a future energy solution - if they take 1500W in, and put out millions of exawatts...

Re:Needs silencing!

[gnuman99]

From TFM:
Sound pressure level 50dBA at operator position
Sound power 55 bels

There is a difference.

Re:Needs silencing!

[Vanth Dreadstar]

I would imagine a nuclear detonation would have a much lower dB rating * snix *

Re:Needs silencing!

[Jozer99]

Isn't that enough to liquify your eardrums from 1/2 KM away? Not to mention any nearby walls or floors...

Re:Needs silencing!

[belmolis]

Yes, there's a difference between sound pressure level and sound power, but the relationship is not such as to make 55 bels a plausible level. That has to be an error for decibels. Sound power is proportional to the square of the pressure. Since these are logarithmic measures, the result is to introduce a factor of two. Here is a handout that explains the relationship.

FPS

[strider44]

How many FPS can you get on Doom 3? I've got to plan my future purchasing decisions.

Re:Interesting practical uses for Personal Cluster

[Jeffrey Baker]

It can't even do a make -j 96, can it? It's not a single system with 96 CPUs, it's a network system of 96 nodes, each with 1 CPU and up to 2GB of memory and ethernet and an optional 2.5" disk.

distcc, yeah. Parallel make, no.

Inefficient ?

[zymano]

Merrimac 2 terraflop workstation for $20,000

General CPU's just don't have the punch that special purpose or Fpga processors do.

Re:Inefficient ?

[Slashcrap]

General CPU's just don't have the punch that special purpose or Fpga processors do.

And FPGAs or special purpose CPUs don't have the generality that normal CPUs have. There's also the small point about the Merrimac system not actually exisitng.

PS. Thanks for linking to Roland Piquepaille's fucking blog. He doesn't get nearly enough links on Slashdot in my opinion.

Re:Inefficient ?

[timeOday]

That's not even a product, it's just a schematic. Talking about building a computer than only has $20K of parts, and running an actual business by selling those computers for $20K each are two very different things.

Wanna know what it looks like?

[io333]

picture here

Re:Wanna know what it looks like?

[venicebeach]

Funny, I assumed it looked like this.

Slashdotted

[AndyCap]

I guess they haven't got one running their website...

Re:Slashdotted

[Vanth Dreadstar]

Since when did website performance depend on processing power? A P-II pure file server can saturate a 100Base-T LAN.

Re:Interesting practical uses for Personal Cluster

[lachlan76]

It can if you set CC="distcc"

Re:Just imagine...

[Adult film producer]

and how many Libraries of Congress ... ahh, I dunno.. What sort of LoC metric can we apply to this machine ?

pong

[Es02]

I wonder, with all that power, if it would play pong?

Re:pong

[moro_666]

actually it really can't win that many rounds as you may think, as all the cpu's are not directly attached to eachother, enormous latencys are created between them (in terms of computing, not in human terms) ...

so basically if it plays one pong at the time, it really can't beat a single chip machine doing the same (as it tries to scale a simple computation over multiple chips it may even be worse) ...

but ofcourse if you make it play 96 pongs at the same time ... it rocks !

ofcourse ... playing 96 doom3-s at the same time would be interesting ... (96keyboards & mice needed ? :p)

Re:pong

[Vo0k]

96 Dooms - no. But 1 Doom using "dual CPU" and 94 Really Smart Bots.

Re:pong

[moro_666]

i'm affraid that even doom3 doesn't have a "fluid" engine but a ticking one (e.g. 250 ticks are done every second, and each bot/player decides if or not it does a move within some of these ticks).

bots have the same knowledge base (they read the same memory about eachothers and your position) , only some limitations (visuality emulation filtering) on it. on a same machine with a single/dual processor accessing the same memory this technology is pretty fast. and also, bot engines are usually quite simple (they don't have to analyze a visual display to see the enemies, they just have to work on some !cached! matrixes of objects). if every bot would run on it's own memory then the memory latency would punch the whole idea into a big black nothing (the matrixes are quite huge and "links" between machines literally 'sucks' in such unit amounts)

so making the bots run on different threads won't really give you that much cause you'll get latency when talking to each bots and they get latency on "talking" to eachother ...

ofcourse it is possible to create one such supergame with a fluid engine and independent smart bots but doom3 surely isn't there yet (no need for it on the marketplace, yet) ...

on the other hand, if they would build a machine that has 96 cpu's working on one memory that they all share (and that still has a reasonable speed under such circumstances) it would be veeerryyy impressive to see 94 bots having fun :)

Re:pong

[Vo0k]

There are 4 areas where computational power is required in such FPS games.
1) Visual rendering - what the player sees. VERY heavy. That's what standard networked "client" does.
2) Bot AI - Way too "lightweight" as for my liking. Generally quite light.
3) Token world - actual "physical model of the world", managing players and bots, calculating collisions, area damage etc, where the real game goes on. The "display client" gets only camera position and a handful of stats, while all the rest goes on behind the scenes. Kind of "Dungeon Master" task. Listen to players, tell them the outcome of their actions, let them imagine what it looks like. This is actually a pretty light job.
4) Management - transferring data between all the above. Sending updates on player and item positions to the rendering client so it could combine them into the local cache of the world map, forwarding player actions to the "world management", all the networking stuff etc.

All the above are quite neatly separable into distinct tasks. The natural place for "Management" is together with the "token world" to make them work really fast - they together make a "dedicated server". It's usually much lighter than client+server like in a standard single player game, so you can squeeze more of it on one CPU. 100 clients isn't all that much...

The "client" task is the most heavyweight, but it's very heavy on the GPU, not the CPU. CPU has very little to do unless GPU lacks some feature which must be software-emulated (and when things start to suck).

And the "bare bones" of the bot task are very lightweight. You get 98% of the CPU time left to perform all the "bot thinking" stuff. From which less than 1% is being used... Make the bot REALLY smart by taking all the CPU power you want, but accessing only as much data as a player would see - local cache of the map plus updates.

Of course bots could communicate, but this could be event-driven, making it life-like, not each bot "talking" 2500 times a second by sending message to each bot, each tick, but only when they -do- have something to communicate, i.e. "I see the enemy at..."

What?!

[Anonymous Coward]

I can't hear you, the guy two cubicles over just fired up his new Opteron cluster. I'm just trying to hold on to my desk!

/.ed

[ZosX]

Warning: mysql_pconnect(): Too many connections in /home/www/php/functions/executequery.php on line 21

Clearly 96 processors just isn't enough!

Not a very good product endorsement if you ask me.

Re:Just imagine...

[X0563511]

I can benchpress that. It wouldn't be incredibly hard to move.

...what? A geek with STR > 10?

Re:Interesting practical uses for Personal Cluster

[tonsofpcs]

Forget Blender, How about Maya or Lightwave [screamernet] rendering?

Suspicious

[cyberfunk2]

I gotta say.. I'm a tad suspicious here.. there seems to be a lot of marketing flash (no pun intended) and scarce details.

What kind of CPUs are we talking about ? I'm assuming we're talking non-shared memory here, and therefore nodes that "retain" their own identies. But then isnt each cpu running it's own kernal ? That is.. This ISNT SMP , right ?

I think the details could be a lot clearer here. The lack of tech specs or simple explinations, and excessive use of buisness speak "Efficiency" "unprecendented power" etc. makes me a tad nervous.

Re:Suspicious

[julesh]

I'm sorry, but since when is "efficiency" (a scientific concept that can be quantified as the amount of work performed by a system divided by the energy input into it) "business speak"?

Wha...?

[raehl]

General purpose processors have *WAY* more punch. Especially punch per dollar, as FPGAs are fairly expensive.

They're just general purpose, whether they be scalar (CPU) or vector (GPU), so an FPGA that is specifically optimized for a specific problem will kick the general purpose processor's butt - in that specific problem.

But try running Quake III on an FPGA - it will be killed by the CPU in processing and killed by the GPU in graphics. Assuming you can even cram everything you need to be a CPU or GPU into the limited real estate of the FPGA in the first place.

Re:Wha...?

[Daniel Phillips]

try running Quake III on an FPGA - it will be killed by the CPU in processing and killed by the GPU in graphics

Oh, you mean, like this?

Re:a) it was quiter

[6800]

Boise Noise Cancelling Earphones aprox $300US.

Perhaps one pair should be included for the $100000!

Re:Suspicious [Transmeta]

[Anonymous Coward]

All the current Orion systems, including this one, use Transmeta Efficeon CPUs. Not surprising since Orion was founded by a Transmeta co-founder.

Actually, Efficeon performance is quite good on the type of repetitive loop-based code this system is intended for. It may not surpass an equivalent Athlon 64 or P4 based system, but in terms of bang per watt, it's not bad.

Re:Just imagine...

[0x461FAB0BD7D2]

How fast can it organize the LoC, using bubble sort and the Dewey system?

Leave the OCR'ing for Google.

Re:Seti

[Vo0k]

No worries. He can set up WINE on it and run the Windows version of SETI@home with the screensaver, emulated :)

A personal renderfarm?

[gbulmash]

When they get these down to say 30-40k, and maybe beef up the processing power of the chips another generation or two (all entirely within the realm of possibility), I could see some of the big animation studios slapping them in deskside and clearing out the big renderfarm racks.

- Greg

p\/\/n3d

[tod_miller]

http://en.wikipedia.org/wiki/Blue_Gene

Now I just need a desk big enough, and a power lead heavy enough to let me class this as a desktop machine.

Re:p\/\/n3d

[utlemming]

Yeah, until the Department of Homeland Security wants to know why you're power consumption has jumped and why you have a 96 node machine running. Could you imagine having to justify one of these things for a personal use to a federal agent? Some how explaining that you want to compile KDE really fast, or play Tux Racer at full-frame rate might not fly.

Re:p\/\/n3d

[Locke2005]

Funny, I just can't think of any apps that I need to run that require 64,000 nodes... not even sure what I'd do with 96, although it sure would make compiling faster.

Question

[roxtar]

From the spec sheet [i] Makes a standard Linux cluster into a standalone computer [/i] I wonder how it does that...

Only on /. (Re:$100,000)

[po8]

You can tell you're on /. when dividing 1e5 by 1e2 to get 1e3 gets modded up to +5 insightful. :-)

On a practical note..

[seamonster]

It looks pretty easy to walk out of the building with - although admittedly probably a limited resale audience. Supervillains spring to mind.

Re:yeah

[phalse phace]

It runs Windows?

Millions of Exawatts

[Anonymous Coward]

But that's only the peak power. :-)

Too expensive

[Buzh]

Although having 96 nodes in a single box makes it quite cute, from what I can interpret from the specs, you would get more bang for your $100K by getting what the beowulf crowd like to call MMCOTS (Mass-Market-Common-Off-The-Shelf, i.e. mass produced computers from Dell or the like), hooked toghether with a specialty high-bandwidth low-latency interconnect like Infiniband, Myrinet or SCI. Running a free beowulf cluster OS like for instance ROCKS would mean that a normal linux admin could maintain it quite effectively.

I expect this thing to be marketed towards scientists in small or medium businesses that aren't employing many/any IT staff, who use commercial computer models to do things like theoretical chemistry (Gaussian, ADF etc), bioinformatics (Phase, BLAS, Paralign etc), fluid dynamics, statistics, crypto, you name it. I don't expect to see any of these types of systems used in normal supercomputing sites, where people write their own (parallel) code and skilled staff maintain the cluster.

Re:Too expensive

[Tucan]

This system is not designed to deliver the most FLOPS per dollar. It aims to address the heat dissipation, space, and noise concerns that arise using lots of MMCOTS boxes. Factor in the Watts and it starts to look really good.

My big concern is that Transmeta recently announced that it was getting out of the chip making business. Unless another company licenses Transmeta's silicon design, Orion is going to run into serious supply-line shortages.

Re:Too expensive

[LWATCDR]

There choice of video really gets me.
"ATi Mobility(TM) Radeon(TM) 9000--64MB integrated DDR" What??? This is not a workstation class card. Why not an nVidia Quadro card? I am sure heat and power are an issue but if it supposed to be an all in one desktop machine it seems like a poor choice.

Re:Too expensive

[CommieOverlord]

Having built and ran clusters, this computer may very well be an option for many people.

Clusters need A/C. A $100,000 cluster might generate 100K BTU/h. That's going to generate serious heat.

A large cluster requires renovating a room to accommodate a number of things. High electricity draw, lots of weight on a small area of floor space, security whatnot. Building a server room from scratch could easily run 100K.

A $100K cluster might suck back 15-20KWh of electricity. Over a couple years that will set you back a couple tens of thousands of dollars.

This Orion system won't be ideal for everyone, or even a lot of people, but it will have a market.

Re:Too expensive

[Buzh]

There would be no reason to have a powerful graphics card in such a system, the applications you would run on that kind of cluster return mostly return their results in the form of (mostly) plaintext, numbers. Forget running your OpenGL screensaver at supercomputer speed, HPC systems like this doesn't do that.

For kick-ass gaming and that lot, another class of clusters entirely are required, called "viz clusters" (viz for visualization). You basically have a number of computers with powerful graphics controllers hooked up to one display each, and the displays are then tiled.

Then you run an application capable of displaying on tiles, like Doom for instance, to divide the rendering of the individual tiles. So if you have say 6 high-end workstations and 6 video projectors running at 1280x1024, you could install a clustered linux and the required libraries and stuff on them, and play Doom in 3840x2048!

Re:Too expensive

[Buzh]

Sure, in terms of heat and power consumption it can make sense. As well as in terms of interior design, as compared to 2 racks full of 1u boxes and interconnects. But there must surely be more factors in the total cost equation when getting such an odd-bird system, unless Orion is planning on having inexpensive first-class support for many years to come.

the Transmeta business raises big questions, where is that CPU going, what sort of compilers are provided and will your existing code run well?

the choice of OS raises some questions too, what sort of management framwork are they providing to make this cluster as easy to maintain as a single node computer?

And a big whatif is, you really need a close optimization for your particular CPU when running HPC applications. Companies like Intel and Portland Group are selling loads of expensive licenses for their proprietary linux/x86/x86_64 compilers to HPC sites, simply because they produce binaries more optimized than what gcc can do.

The commercial HPC apps usually come pre-compiled for various CPUs, but I can't recall ever seeing Transmeta mentioned. Maybe you could get some help if your software vendor or creator is still supporting the software, maybe you can't. Some apps are open-sourced, and sometimes you can get the source code for proprietary apps if you are an academic user or whatever, but what amount of work and computer expertise is needed to get the damn thing to compile?

My gut feeling says this server is a real one-off oddity, I'm not sure the Orion stockholders are going to be sleeping too well in the near future..

Re:Too expensive

[LWATCDR]

"There would be no reason to have a powerful graphics card in such a system, the applications you would run on that kind of cluster return mostly return their results in the form of (mostly) plaintext, numbers."
They why not have it in a 4 or 6u rack? To me the desktop aspect of it means interactive. I was thinking of a Cad system that does real time stress analysis and simulation. Or even for single screen visualization.
No I was not thinking of OpenGL screen savers.

Rubbish

[fleppir]

It's up and running right now.

Cluster != SMP

[fleppir]

Not neccesarily at least. You could have a cluster of SMP machines thou ....

150 pounds of equipment ...

[fleppir]

... is not something you carry under your arm like a desktop.

Re:150 pounds of equipment ...

[seamonster]

Hey, I work out.... I've yet to be challenged when I've a trolley full of gear and a PO'd expression. In fact, people hold doors open for you

Slashdotted

[fabu10u$]

Apparently they can't be bothered to run their website on one of these babies... I guess the cobbler's children never get any shoes.

Re:Apple?

[EvilAndrew]

Parent poster messed up on their calcs. Current XServes are 18.4 GFlops peak, not 35 eg Virginia Tech currently at #7/500 is 20240 GFlops peak for 1100 XServes. So 7 would only be ~129 GFlops peak, and 33 would be 607 GFlops peak. But not exactly fitting in a single tower case - though 8 would fit nicely one of those mini sound-padded racks which would be almost as good. And at least the last time I saw a price comparison made, the G5s were far cheaper than comparable rack P4s. (The G5 has 2x the FP hardware).

Altitude

[hey]

It has Altitude restrictions:

Altitude -300 meters to +3000 meters
-1000 feet to +10,000 feet

I've never seen that before.
So you cann't use it on a plane.

Re:Altitude

[hrieke]

Fairly standard disclamer text there- I've seen that text in everything from HD to stereo equipment manuals.

Re:Altitude

[anno1602]

So you cann't use it on a plane.

The air pressure in planes is higher than the air pressure outside at high altitudes... You can use it on a plane (why would you want to).

Re:Altitude

[omega9]

From the site itself:

They're fully scalable so you can add performance as your needs expand. It can be used on site: in the office, the laboratory, on a boat, or even aloft in a plane.

Ain't that sumpin.

Re:Altitude

[Analogy Man]

Dumping heat at higher altitudes will be an issue (thin air).

An airplane cabin is pressurized so you could as they say us this machine on a flight test or atmospheric research machine. Most data acquisition experiments would likely gather data with more modest equipment and do the analysis in a tera-ferma environment.

Re:Altitude

[imsabbel]

Then you never looked closely.

Something like that can be found on most crt monitors and nearly every harddisk.

Re:Altitude

[Locke2005]

So you cann't use it on a plane.
Heck, it probably wouldn't fit on the seat-back "table" anyway...

Re:Altitude

[flaming-opus]

Hard drive heads are tiny air-foils. They depend on a certain barometric pressure to keep from crashing into the platter. I had a device that measured weather characteristics in the mountains of antarctica. We had to use magneto-optical drives, because the heads on a disk drive kept crashing. This was many years ago though. I'd think it has improved since then. Maybe not.

Re:Altitude

[hey!]

I have, it's pretty common. Years ago I remember laughing at th especs for a computer that had a maximum altitude of 6561.66 feet. Obviously, an engineer told the marketing people that ti was good for about 2km.

By the way, I like your account name, but you need to do a little something to jazz it up a bit.

Re:Altitude

[Pharmboy]

for the record, the air pressure in a typical plane at 30k feet is about the same as normal air pressure at 8k feet, which you would think is still out of spec. This is one reason my knees hurt like hell after a 4 or 5+ hour flight: they swell, even if I move around alot.

Boeing is working on composites for the 787 to get the internal air pressure the same as at 6k above sea level in the near future. Read about it in last month's Popular Mechanics.

Is it just me..

[sucati]

or does that seem a bit odd to bundle it on Fedora? I like Fedora as much as the next guy, but maybe such an expensive solution would be better suited for commercial distribution on a more predictable release schedule e.g. RHEL, Suse Enterprise etc?

And?

[daVinci1980]

Who cares? Modern graphics cards are capable of (sorry it's a PDF, it was all I could find) 40 GFLOPS. That's not even in SLI mode, which actually does push you to about a 98% over a single card (in terms of raw processing power).

Why would you buy a 96-CPU setup when you could buy a 6-GPU setup and match the same theoretical performance? (All jokes aside about the costs being roughly equivalent, they're nowhere near the same.) 6 top of the line 6800s would run you about $3600. Even if you added top of the line parts for the rest of the system, you'd be looking at about $1600 per system. Add $0 for the linux distribution to drive the whole thing, and you're at a grand total of $10K.

I'm not impressed.

Re:And?

[sjwaste]

Serious question here. Does production software exist to drive arbitrary computation across a GPU? I've seen articles about software on its way, etc. Does it exist, either as an application or integrated into some OS? Man, if I could push some of my statistical computing off to the GPU...

Re:And?

[daVinci1980]

Right now, I do not believe there are any general-purpose software packages available. That being said, the article I referenced is a discussion on exactly that topic.

Which means that right now, you have to do a little bit of leg work yourself in terms of getting the data to and from the GPU (in textures). I can find out if there are any toolkits later today and let you know, though.

Re:And?

[ravenwing_np]

Look for GPGPU. They are trying to use the graphics processor for general purpose operations. It runs as any CG script would run. Just realize that it is focused more on parallel math operations then procedural. Please note that I have nothing to do with this project and haven't tried it yet.

Re:And?

[graphicsguy]

you could try Brook

Re:And?

[timeOday]

Serious question here. Does production software exist to drive arbitrary computation across a GPU?

No, because graphics hardware cannot do arbitrary computation. At least not at anything like the FLOPS it achieves doing graphics.

I've attended a workshop on using graphics hardware for accelerating other computation and it's mostly hype IMHO. It amounts to rendering images of your problem, then doing feature extraction on the image. So the *effective* FLOPs, i.e. the amount dedicated to *your* task rather than the overhead of reducing from a rendering task, aren't all that impressive.

A more serious challenger to this transmeta-based system, IMHO, is the cell processor.

Re:And?

[daVinci1980]

There's actually a staggering amount of performance to be gained out of doing stuff on the GPU... Anything that can be thought of as a massively parallel application will work pretty well there.

For example, FFTs can be done on the GPU at roughly 1.5-2x the speed of doing them on the CPU. (Of course, doing them on an FPGA will still trounce the GPU performance by--I believe--a factor of ~8.) But given a few more GPU generations, I believe this margin will be decreased signficantly.

There's a demonstration and whitepaper on this over here.

Re:And?

[Arkaein]

It amounts to rendering images of your problem, then doing feature extraction on the image.

This may be one approach, but it is certainly not the only one. The Brook software (linked a few replies above) takes advantage of modern, programmable GPUs to write programs that have little to do with graphics rendering. I saw a talk by Pat Hanrahan, the guy leading the project, and for some mathematical algorithms they achieved something like 10x speedup over highly optimized, best of breed implementations compiled with Intel's C compiler.

There are issues, possibly the biggest being that GPU programs are limited in size and the GPU bandwidth on the AGP bus is assymetric.

The Islands of Patmos Super Computing

[Danzigism]

Yea, I agree that Fedora was definitely an odd choice.. Well, I can trust that the kind of person whom can build a 96 node super computer, makes very educated decisions.. I'm glad to hear that company's are still involved in making these clusters. Its a great way to build something powerful for a cheap price, and not having to lean towards Crays etc.. I worked for a company called Patmos International for the longest time, and we never shipped a single cluster.. We had tons of investors that seemed interested of course, but after 2 years of contiuous development, and no sales, the investors simply stopped investing, therefore my job was done for.. We advertised the "$99,000" super computer that would supposedly be in "everyones" garage one day.. Of course that was just a saying because of how cheap we could offer a 32 node system with all of our custom applications and linux operating system. Pretty sweet setup.. it sucks to see the big guys go down sometimes.. to this day, it was the best job I've ever had.. You can still read about Patmos if you search for James Gatzka on google.. They tried their hardest to bring some damn technology and culture to the podunk town of the Eastern Shore of Maryland..

Re:The Linux community and the notion of desktop..

[sjwaste]

It's a "desktop cluster" because its compact. The word "desktop" is a good choice here, because what else would you call it and still be able to market it? "footprint of a desktop but too much power for mundane tasks" cluster? Don't take things so literally. At the end of the day, the people signing the purchase orders for these things are management, not low level IT staff, so you have to market accordingly. Also, Quebec isn't a country. It's a province in Canada. You may not like it, but that doesn't change the fact that it is, just like this product being a "desktop cluster"

Re:The Linux community and the notion of desktop..

[NeedleSurfer]

technicaly speaking Tibet is not a country is a region of China, Taiwan isn't a country it's a region of China, and so on...

Quebec is a country, it's just not free yet...

Re:Fedora Core 2?

[blueser]

Anyway, choosing Fedora N for a production environment is a bold choice to say the least... although lots of people do this according to posts on the Fedora ML, its short lifecycle will most likely turn into a serious upgrade-headache for sysadmins. Despite that, specs are cool, this cluster delivers some respectable processing power =)

Intended uses

[mattr]

What would be the target market for this kind of thing? Genomics and biochemistry? Engineering workstations for the department? Rendering? How about to run a company's desktops? Seems like it might be useful for CAVE-like environments and videoconferencing throughout a distributed office.. also maybe for a service provider offering virtual linux pcs?

Re:Fedora Core 2?

[lachlan76]

It's good enough for kernel.org, so why not? It's got just as much support as Gentoo or Debian or any other (free) distro.

Re:The Linux community and the notion of desktop..

[Mistah Blue]

I beg to differ, but Taiwan IS a country.

in my country?

[volley_srfd]

Last time I checked, your country was CANADA. You had your chance to get out of the union a few years ago and messed it up. No more whining!

Re:a) it was quiter

[Mistah Blue]

No, the price for those headphones should be: priceless. Had you said that, you would probably already be modded +5 Funny.

Multichannel EEG processing in Canada vs Chile

[mindpixel]

One of my best friends just bought a tiny little house in downtown Toronto for $377,000. I left Toronto last November and moved to Santiago, Chile and live downtwon where my rent is $260/month, for quite a nice, though small place, in an excellent area.

So, if I spend $100K on the Orion DS-96, that leaves me more than enough for a 250 channel geodesic EEG system which would allow me to compute self-organizing maps of the human mind based on flashing the 1.6 million mindpixels I have collected over the past five years to various volunteers [english teachers], AND still have 56.73 years worth of rent left!

Too bad no bank will loan me $377,000 for a computer and an EEG system and the time to play with it...

Wow

[graphicsguy]

It has almost as much compute power as my 3D graphics card!

Eclipse Performance

[myyrk]

With this maybe eclipse can keep up with my 2 finger typing.

cooling & UPS too

[Kludge]

Your $100K cluster will require a $40K cooling unit and a $40K UPS on top of the power costs.

Re:Just imagine...

[tomhudson]

...a Beowulf Cluster of these...

Looks like they could use a cluster of clusters after all:

Warning: mysql_pconnect(): Too many connections in /home/www/php/functions/executequery.php on line 21

Don't forget IBM

[kayak334]

Their 1350 Clusters all run Linux and use Infiniband, Myrinet, or GigE as the high speed interconnect and can bring you anywhere from 4 to 1024 nodes. You actually can order more than 1024 nodes, it's just not listed on the website. IBM 1350 recently built a cluster for a customer that was ~8000 nodes.

Total waste of time and money

[Boogiesbunny]

This is just a gimmick to capitalize on the craze of HPC. People will sit back and watch to see how this "Cluster" performs. If multiple nodes at once fail you would have to shut the system down and crack the dumpster open. I call this system a very fast desktop not a cluster.

Post credibility test regexp

[cloudmaster]

if($post =~ /^(?:.+\s)?phat[\s\.,]/i){
lose_all_credibility($post);
}

Re:Post credibility test regexp

[brsmith4]

If the use of an urban slang in an informal post on an informal website completely destroys my credibility, then so be it, biatch ;)

Re:Post credibility test regexp

[cloudmaster]

Augh! Too much slang! :)

Re:Interesting practical uses for Personal Cluster

[Danzigism]

hahah i haven't heard someone mention Blender in freakin years.. i wonder if anyone still uses moray or povray

Slashdotted!

[3Suns]

Warning: mysql_pconnect(): Too many connections in /home/www/php/functions/executequery.php on line 21

How blissfully ironic!

Re:Interesting practical uses for Personal Cluster

[frank_adrian314159]

For folks who often work with parallel systems (e.g., simulations, bioinformatics, soft computing researchers) this is a great thing. I can see a huge number of research projects including these machines in their grant proposals.

Re:The Linux community and the notion of desktop..

[kyoko21]

Off-topic:

If Taiwan is a region of China, then why are there two different passports? When one visit China from Taiwan, they need to bring their passport. Being in the U.S. I do not need to bring my passport in order to cross statelines. If that was the case, that would be insane.

China may think Taiwan is their region, but they sure don't act like it. And being born in Taiwan, I say Taiwan is a country. Though it would be good to have China and Taiwan reunited again.... though I don't know when and how that is going to happen.

Re:Seti

[nmaeone]

Wine Is Not an Emulator, tyvm.

Cost increase reasons over individual clustering

[aixguru1]

Most people are not noting the primary reasons for the cost increase over building a cheap cluster. Their system can run off one power outlet. It is designed to pull a smaller amount of power than a typical cluster. Imagine the power bill from 96 350 watt power supplies. That's 33.6 kwh if my calculations are correct. They claim on the spec sheet a maximum consumption of 1.5 kwh total (1500 watt). That is considerably less in the grand scheme of things.

The other thing to consider is the cost of engineering in the nodes themselves.

Re:Only on /. (Re:$100,000)

[Soon-to-be Has-been]

Damn, if only I had some mod points with which I could raise this comment!

Doesn't Playstation 3 beat it?

[RhettLivingston]

Last I heard, they claimed a solid 250 GFLOP rating from the 9 core cell processor. So, a Playstation 3 should be able to beat this. Or, if you want to pay some more, one of the IBM workstations equipped with Cells that the Playstation 3 software will be developed on.

Another processor that may help in the classes of problems that these things hit would be the PhysX chip from Aegia, due this Christmas. I'm currently working a project that hopes to eventually be looking toward either PhysX chips or Cells on multiple PCI or PCI-E cards hosted in quad dual-core Opteron chassis for accelerating a simulation farm. Either way, Cell or PhysX, I'd expect a factor of 10 cost reduction versus this beast for even better performance on highly parallel problems, though more general purpose problems would go to the cluster.

Re:Doesn't Playstation 3 beat it?

[Daniel Phillips]

Last I heard, they claimed a solid 250 GFLOP rating from the 9 core cell processor. So, a Playstation 3 should be able to beat this

Does the Playstation 3 have 24 GB of memory, 960 GB of disk and 10 GigE interconnect?

Re:Doesn't Playstation 3 beat it?

[RhettLivingston]

No, but with 250 GFLOPs on a single chip, for certain types of problems, it will be much more efficient than the same number spread across 96 chips. Also, in a week, I'll have a quad dual-core Opteron, that's 8 64-bit cores, on order with 32GBs of memory, 10GigE and I could easily equip it with a TB of disk though I don't care for that. When the PhysX cards come out later this year, it should beat this machine hands down. And when Cells on expansion cards become available, probably late this year or early next, I could hit a more general purpose TFlop with 4 expansion cards. The total price would be

The perhaps badly made point is, if your application is something simplistic like web or DB serving, this beast might not be too bad. But if you're looking for something to perform highly parallel numerical processing, this is not the most cost effective approach. Specialized processors working in conjunction with general purpose processors will yield far better price/performance.

Re:Doesn't Playstation 3 beat it?

[RhettLivingston]

Should have used preview :-) That first paragraph should have "less than 1/3 that of this 96 processor beast." appended to it.

Re:The Linux community and the notion of desktop..

[Locke2005]

In my country (Quebec) we have this problem
Yeah, you do... and I think the problem is that you keep insisting Quebec is a separate country, and not a part of Canada!

Did you read any of the links?

[TamMan2000]

Or do you not know much about parallel computing?

They describe it here

They mention the version of message passing they use (message passing implies a distributed memory system). They mention what proc they are using. And nobody said this was SMP, they said it was a cluster (which as mentioned by another poster means that it isn't SMP).

I hope...

[Jozer99]

I hope they have a big desk. If you search around their site, you can find some pictures that show the scale of the DS-96. It is about as big as a squat refridgerator. Not that I am complaining. Their processing power/area is WAY better than anything else out there, but it is definitely not a desktop system.

An SGI Onyx for the 21st century

[heroine]

So if I went to an Indian college I would probably see these floortop clusters in the same places SGI Onyx's used to occupy in the 90's. It's the first time since 1995 you could actually see a quantum leap occuring in the same environment as standard PC's.

Since they don't name the CPU, it's probably a 32 bit Chinese x86 chip. Combined with the gigabit ethernet, it's hardly enthralling. If they upgraded it to 96 Opterons and 10 gigabit ethernet with an NVidia Quadro FX card, it would be something.

Re:An SGI Onyx for the 21st century

[blofeld42]

Gigabit Ethernet gets dissed excessively as a cluster interconnect. It works fine in most small-to-mid installations, though the latency is higher than some other options.

One thing people are forgetting......

[tempest69]

That a mini cluster is way easy to sneak out of the office... heck one janitor, one night, one car. and boom big fat take. makes laptops seem pretty smalltime.

Re:Interesting practical uses for Personal Cluster

[Vanth Dreadstar]

As long as Maya runs on it, I will be happy. I will be able to gain some room back in my house, as it will replace several existing boxes with a single one, possessing far in excess of my existing total processing power. I don't care if I can get more power for the same money, if it means losing my living room. This box will give me room to put my treadmill back up so I can burn off the pot belly I have developed sitting at a desk 10-14h a day :)

Re:Interesting practical uses for Personal Cluster

[Vanth Dreadstar]

...and besides, imagine the bragging rights from having 230teraflops peak under your desk...almost as good as one...nevermind ;p

Re:Interesting practical uses for Personal Cluster

[torpor]

At $100,000 I wouldn't call this personal.

If I could do something interesting with it, I would. Better than a car ..

Re:Apple?

[coopex]

The correct link is http://spl.haxial.net/apple-powermac-G5/ where it does document Apple's blatent lies.

I just wonder...

[wyldwyrm]

Is a computer that big and expensive a feasible alternative to a girlfriend? "I just like to rub up against it because it's warm". Don't short anything. Pun intended.

gentoo...

[KillShill]

still takes a week to compile.

You work out ....

[fleppir]

... and still frequent slashdot? I thought that brawn and brain was a sum of 1 proportioned by the amount of time frequenting slashdot ........

I bet they stop that hardware leak FAST...

[antispam_ben]

That a mini cluster is way easy to sneak out of the office... heck one janitor, one night, one car. and boom big fat take. makes laptops seem pretty smalltime.

I bet this thing phones home when it finds itself suddenly on the Internet or finds its IP address has changed, or anything. Heck, for the price it ought to have a GPS receiver and cellphone built-in that alerts authorities/owners/ClusterJack Central as soon as it detects itself going out the door.

Wasn't there a slashdot article on that sort of stuff (GPS/cellphone antitheft alert when an 'unauthorized location' is detected) being the next thing in laptops?

Re:I bet they stop that hardware leak FAST...

[PingPongBoy]

Janitors have the most access - I wonder what the policy is. Are they bonded? Does insurance cover theft?

One of the most valuable things per kilogram is information media like documents, disks, tapes. They can be copied or taken so easily.

Perhaps the easiest test is to leave temptation lying around and see what happens. Little cameras can be used to ferret out the schemers.

Then again, a bit of education may be the best course. Everyone should be made aware they are all together for the common good, that best place for all the expensive things has already been planned and it would only be suboptimal to diverge from this. If someone is in sudden need of liquid wealth or a toy, the employer should try to be as accommodating as possible. Things that cannot be lost should be locked up.

Re:Apple?

[coopex]

20240/2200 = 9.2 Gflops/CPU max for Virginia Tech.
2048/256 = 8 Gflops/CPU max for UCLA.

Gee, I was off by 1.2Gflops/CPU.

Maybe YOU should't be such a delusional apple fanboy, and notice that even using virginia tech's numbers, 35 Gflops/xserve = 1.9*9.2*2. Notice the 1.9 in there, that would seem to indicate that a dual xserve does not get 35 Glfops, and that apple lied about 8.3 Gflops/CPU.

And there still is the matter of an dual xserve being 2/3 as slow as a 3ghz dual p4, while costing twice as much.http://spl.haxial.net/apple-powermac-G5/

Re:Seti

[dark grep]

Hey, how did you know that's what I was going to do?

Re:Apple?

[EvilAndrew]

I read the rant you linked to and while I think it had some reasonable things to say, I think you'll find you're a bit off here.

As far as supercomputing goes, it's all about doubles rather than floats, and the G5 has a strong edge over the P4s here.

For example, the top P4 supercomputer ranks at #10 with 1250 3.06 GHz Dual Xeons.

Quoted Rpeak is 15300 Gflops, coming to 6.12 GFlops per processor, or 12.24 for each dual box.

As a reminder, the XServes are 9.2 Gflops per processor, or 18.4 per dual box.

As far as pricing is concerned, I configured a Dell PowerEdge 1850 (the successor to the 1750s used in the previously mentioned NCSA cluster) to be equivalent to the Dual Xserve G5.

It came to slightly over $4000, while the Dual Xserve G5 is obviously overpriced at $3000.

Though I must admit to comparison wasn't totally fair - to keep the price of the Dell down I configured it without an operating system.

So to summarize:
The Xserve G5s are $3000 for 18.4 Gflops - or $163/Gflop.
The Dell Xeon was $4000 for 12.24 Gflops - or $326/Gflop.

Reconfiguring the Dells with faster processors, 3.6Ghz instead of 3.0Ghz, raises the price to $4600 for 14.4 Gflops, or $314/Gflop

Re:Just imagine...

[julesh]

Warning: mysql_pconnect(): Too many connections in /home/www/php/functions/executequery.php on line 21

This is a _silly_ configuration error. Basically, apache is running with more processes than the mysql concurrent user limit. They probably upped the number of apache processes to handle the slashdotting, but didn't think about updating the mysql settings to match.