Slashdot Mirror

What I never understood is why someone like IBM didn't come along and cluster 10,000 dual P4 nodes together for fun to get on the top spot. I'm sure they have the inventory to write that off.

That would be ASCI White, which is currently #8 on the top 500. It's an 8192-cpu Power3 machine, and they didn't do it just for fun. It was #1 on the top 500 in Nov 2000.

Also, #10 on the top 500 is a 1920-node IBM Xeon 2.4Ghz cluster, but why should IBM use Intel processors when they make their own?

What I never understood is why someone like IBM didn't come along and cluster 10,000 dual P4 nodes together for fun to get on the top spot. I'm sure they have the inventory to write that off.

That would be ASCI White, which is currently #8 on the top 500. It's an 8192-cpu Power3 machine, and they didn't do it just for fun. It was #1 on the top 500 in Nov 2000.

Also, #10 on the top 500 is a 1920-node IBM Xeon 2.4Ghz cluster, but why should IBM use Intel processors when they make their own?

What I never understood is why someone like IBM didn't come along and cluster 10,000 dual P4 nodes together for fun to get on the top spot. I'm sure they have the inventory to write that off.

That would be ASCI White, which is currently #8 on the top 500. It's an 8192-cpu Power3 machine, and they didn't do it just for fun. It was #1 on the top 500 in Nov 2000.

Also, #10 on the top 500 is a 1920-node IBM Xeon 2.4Ghz cluster, but why should IBM use Intel processors when they make their own?

A actual avalanche is orders of magnitude more complicated. It'd probably be easier, and much more informative, to simulate it on a computer, actually.

The same thing could be said about an avalanche relative to most of our fluid dynamics models. It is still orders of magnitude more complicated. Which isn't to say that our models aren't quite accurate, they are. But at the same time, they're merely a 'good enough' oversimplification of what's really going on.

Fluid dynamics are an extremely difficult thing to model, and even more difficult to compute. They could challenge most of the supercomputers on the top500 list.

In fact, the top computer on there, Earth Simulator, (at well over double second place) has one of it's two primary objectives being the calculation of relatively simple fluid dynamics models across the Earth's entire oceans.

So, to put it summarize my point here, the best fluid dynamics models we have are extremely expensive to compute, and they are still not perfect. The best way to better understand, and therefore better model, what is happening, is still to experiment with real physics. This experiment will help us develop faster-but-still-accurate models, or extremely precise models for fluid behavior. Either way, recording the locations of each 'particle' as they flow is actually research and will provide a solid set of data for future research to build on.

Disclaimer: I work in the petroleum industry, and therefore only have experience with extremely high pressure/small scale fluid dynamics. My extrapolations may not hold true to the broader field of fluid dynamics.

In Apple's MacWorld presentation (and this film) they show how the VT supercomputer is #3 and they talk about the details, but barely touch on #2 and #1. I'm curious what operating systems primarily drive those two, but none of the searches on Google I'm doing are turning up the info that I'm seeking. Anyone have any links or resources to share that can clarify it?

Tried this?

Paragraphs, man. They're useful.

Anyway.. no OpenMosix here, this is using MPI. Specifically, on top of DK Panda's MPI libraries, they brought Kazushige Goto in to optimize the BLAS libraries in order to obtain the Top500 ranking of 10+ TF.

Incidentally, the Top500 rankings are based on a standardized LINPACK benchmark and formula, not "raw" processor rankings. I saw another comment that implied the latter.

Other interesting notes:

• With conventional air cooling, the airspeed throughout the facility would have been 60-70MPH+. Try working on a console in a hurricane.
• Dr. Varadarajan is a very very cool guy. He absolutely knows every detail that is going on in this machine, and knows how to make a good story out of it.
• The facility this is in was upgraded to handle 3MW. The current cluster takes around 1.5MW. And you thought your Athlon was hot. :)
• The #1 Top500 machine, the Earth Simulator, not only runs on custom Hitachi vector hardware, but required an entire new building to be built. The facility is a feat in itself, and is a big portion of the cost (for those of you extrapolating cost/performance if it was built at the same time as System X).

According to http://www.top500.org/list/2003/11/, number 1 and 2 were built all the way back in 2002. But neither uses off the shelf hardware. Both were custom built.

It's currently #3

It was finished a while ago. Currently it holds the number three spot in the world.

The biggest FLOPS can be found here.

I know this isn't quite on topic, but I wonder how the latest Alpha design would fair. The alpha was the first mass produced 64 bit chip that had any commercial success. It was introduced in the early nineties. IN fact Linus had one. Basically the curret EV78 is a 6 or 7 year old design, but in most serious tests of processor power it has done quite good. It's amazing that such an "old" design still works so well. The last SPEC numbers I can find are here. Considering the platorm has been ignored and basically orphaned, it's suprising that this chip still powers many of the worlds top rated super computers.

How does all this relate to the G5 and Opteron? Well AMD gets it's bus design from the Alpha lineage. The G5 is built by IBM, who I believe is building the alpha cores as well (I could be wrong, I can't keep up). The irony? Every current intel pentium chip is quality control checked by machines with alpha processors. Funny world huh?

AngryPeopleRule

considering that the most inefficient algorithm on the hardware layer is ANY floating point calculation. that's a major reason that Flops determine the Top 500 Supercomputers and as for distributed computing, it's needed for power because they use an operation called Discrete Fourier Transform (a Fast Fourier Transform on n points, rather than the entire real number system) plus compiling a program in C++ is quite inefficient. ANSI C is MUCH more efficient (decently close to GOOD hand assembly) this is true only because the wanted to work at it. From a management level, OOP (including C++) is good. C++ just sucks at implementing it. C++ code: Microsoft Windows* Microsoft Office Almost every game made recently(with some assembly inline) Xerces parser Windows system calls in POSIX mode and Win32 mode (as of NT4, windows could do POSIX). etc.

One advantage of a cube is stackability. The Virginia Tech's X is an example of how to build yourself a very powerful computer, but can you run that in the back of a van? Too much space is wasted. I can link cubes together, put some legs on them and end up with a computer centipede.

Official Site

Short Blurb from Time Magazine


Descriptive Article (with pictures)

Details Regarding its Supercomputer Status

Finally! A real use for the Earth-Simulator. Or maybe ASCI Q would be better suited?

And why can't the supercomputers have more intense names?

Number 44 - WETA Digital New Zealand/2003 BladeCenter Cluster Xeon 2.8 GHz, Gig-Ethernet / 1176 IBM

I recently did a search of top500.org which has specs back to June 1993 and up to June 2003

BTW WHO THE HELL BROKE top500.org!?!? This site used to be easy to use and informative, now it is a banner add hell, that obscures the info you used to be able to get to easily, with many broken links and apologies for works in progress.

Anyway I digress, the point is that in 1993 the fastest computer was the TMC at Los Alamos with GigaFlops ratings of 59.7 Rmax 131.0 Rpeak
My Dell XPS today would rate in the top half of fastest machines in the world for 1993 if I'm reading the stats right with just over a GigaFlop of power.

Todays fastest machine is Japan's Earth Simulator rated at 35860 Rmax 40960 Rpeak

If we define a super computer as the ability to get in the top500 then 245.1 Rmax 384.0 are the numbers that indicate your machine would be a super computer by 2003 standards.

I recently did a search of top500.org which has specs back to June 1993 and up to June 2003

BTW WHO THE HELL BROKE top500.org!?!? This site used to be easy to use and informative, now it is a banner add hell, that obscures the info you used to be able to get to easily, with many broken links and apologies for works in progress.

Anyway I digress, the point is that in 1993 the fastest computer was the TMC at Los Alamos with GigaFlops ratings of 59.7 Rmax 131.0 Rpeak
My Dell XPS today would rate in the top half of fastest machines in the world for 1993 if I'm reading the stats right with just over a GigaFlop of power.

Todays fastest machine is Japan's Earth Simulator rated at 35860 Rmax 40960 Rpeak

If we define a super computer as the ability to get in the top500 then 245.1 Rmax 384.0 are the numbers that indicate your machine would be a super computer by 2003 standards.

I recently did a search of top500.org which has specs back to June 1993 and up to June 2003

BTW WHO THE HELL BROKE top500.org!?!? This site used to be easy to use and informative, now it is a banner add hell, that obscures the info you used to be able to get to easily, with many broken links and apologies for works in progress.

Anyway I digress, the point is that in 1993 the fastest computer was the TMC at Los Alamos with GigaFlops ratings of 59.7 Rmax 131.0 Rpeak
My Dell XPS today would rate in the top half of fastest machines in the world for 1993 if I'm reading the stats right with just over a GigaFlop of power.

Todays fastest machine is Japan's Earth Simulator rated at 35860 Rmax 40960 Rpeak

If we define a super computer as the ability to get in the top500 then 245.1 Rmax 384.0 are the numbers that indicate your machine would be a super computer by 2003 standards.

I recently did a search of top500.org which has specs back to June 1993 and up to June 2003

BTW WHO THE HELL BROKE top500.org!?!? This site used to be easy to use and informative, now it is a banner add hell, that obscures the info you used to be able to get to easily, with many broken links and apologies for works in progress.

Anyway I digress, the point is that in 1993 the fastest computer was the TMC at Los Alamos with GigaFlops ratings of 59.7 Rmax 131.0 Rpeak
My Dell XPS today would rate in the top half of fastest machines in the world for 1993 if I'm reading the stats right with just over a GigaFlop of power.

Todays fastest machine is Japan's Earth Simulator rated at 35860 Rmax 40960 Rpeak

If we define a super computer as the ability to get in the top500 then 245.1 Rmax 384.0 are the numbers that indicate your machine would be a super computer by 2003 standards.

That being the case, why aren't distributed apps considered as part of the Super Computer list?

Most of the tasks you pick a supercomputer for aren't things you can cut up into a thousand chunks and let every computer finish it's chunk of the problem independently. In particular, the benchmarks (LINPACK) that determine who goes where on that supercomputer list generally measure a computer's performance at big linear algebra problems (which are what takes up most of the compute time for huge classes of real problems), and for those problems every node needs to share results with many other nodes after essentially every iteration: this means you need high bandwidth and very low latency connecting the nodes.

Now, the supercomputer benchmarks may make things worse than they have to be: according to this they're measuring performance on dense matrices (where every node needs to talk to every other node), whereas many real world problems can be discretized into very sparse matrices (where each node only has to talk directly to a few of the others) instead - still, even in the sparse situation you want your computers to be separated by microseconds across your high speed interconnect rather than milliseconds across the low bandwidth internet.

Top 500 Supercomputer list

Top500.org

My desktop machine is faster than a Cray 1, and it'll never be labelled "Supercomputer" by any rational being.

Unless their architecture actually hits the Top Ten, I'm not going to be impressed that it's overcoming its handicap. Unless you're running a Special Olympics for computers and "everyone's a winner."

That only shows how timely the definition of a supercomputer is. 100 common desktop machines are very uncommon and obsolete 3 years from now.

Right. I recall Apple making a big stink a few years back about being a desktop supercomputer when they hit 1 GFLOPS, or whatever the benchmark was that initially established the first supercomputers. What makes a computer "super" while Moore's law is still being met will definitely change over time.

I think energy efficiency (MOPS/Watt) is a very relevant metric.

Only inasmuch as electricity costs money, and those costs can be compared to the cost of pushing more horsepower at the problem. I mean, if you have some calculations that daily require a 1 TFLOPS cluster (say some weather prediction), it doesn't much matter if you can get a .5 TFLOPS cluster that consumes 75% less power. Yeah, in theory you could get two, but there are coordination costs and scaling inefficiencies that would have to be dealt with so that in order to actually see 1 TFLOPS from the the second architecture it might end up being less efficient and requiring more energy.

All I'm saying is that if they want to claim any sort of efficient supercomputer, they have to have something on the Top 500 list and then worry/brag about what kind of low power consumption it has.

If you look at the more recent November 2003 list instead of the older June 2003 one, this cluster would rate more like #84 than #38.

If you look at the more recent November 2003 list instead of the older June 2003 one, this cluster would rate more like #84 than #38.

"The machine is the first supercomputer based on Macs; it is one of the few supercomputers built entirely from off-the-shelf components and it cost a bargain-bucket price -- only $5.2 million. By comparison, most of the top 10 supercomputers cost about $40 million and up. The Earth Simulator cost $350 million."

"The $200 million (US$) computer is the fifth in the DOE's nuclear weapons simulation program, called Advanced Strategic Computing Initiative (ASCI), which allows the U.S. to keep its nuclear stockpile while complying with the nuclear test-ban treaty. Named "ASCI Q," the newest supercomputer will be capable of performing 30 trillion calculations per second.*"

The facility's name is Terascale Computing Facility. The cluster itself is called "X".

This is similar to NCSA. The facility is called the National Center for Supercomputing Applications (NCSA). Their new #4 cluster is called "Tungsten".

The facility's name is Terascale Computing Facility. The cluster itself is called "X".

This is similar to NCSA. The facility is called the National Center for Supercomputing Applications (NCSA). Their new #4 cluster is called "Tungsten".

Here is the query I ran (top 256 clusters).

I'm sure that no one has trademarked the letter X yet...

Linkie

Hmm, guess this means my submission a couple hours ago won't go through (dangit, Wired!)...

Here is the official press release and the list.

There is a lot of good points to note all around. The first is the G5 Terascale cluster at Virginia Tech at #3 (10.28 Tflops/s, 2200 CPU, Infiniband) is the first academic computer to break 10 teraflops/s. This extra performance was promised at Mac OS X Developer's conference last month. Not to sure if the price is a testament to Infiniband ($1.5 million cabling, cards, and routers) or the Macs ($4.2 million list).

Good thing too because in a surprise move the NCSA cluster made the list at #4 (9.82Tflops/s, 2500 CPU, Myrinet). This cluster is built using Dell's running Pentium 4 XEONs and Red Hat Linux! One subtle point to note is that they didn't get all the systems online in time (there should be 2900 CPUs, not 2500). I bet some programmer at PSC and an ex-Chief Scientist of SDSC is appreciating having a hand in edging out NCSA for #3--not to mention Apple beating Dell for #3.

The fastest Itanium cluster is at #5 (8.63 TFlops/s, 1936 CPU, Quadrics) which is looking like the odd man out boxed in by a PC based systems using Myrinet, the P4 Xeon above, and the most powerful Opteron system at #6 (8.05 Tflops/s, 2816 CPU, Myrinet). Another point of similarity:did I mention it's also using Linux?

And finally, It's easy to overlook #73, a single compute node of BlueGene/L (1.44 Tflops/s, 1024 CPU). Imagine 128 of these connected together and you have something that will easily take #1 when it's completed even if we handicap it 20-40%. As noted on SlashDot earlier, this will be running Linux.

Hmm, guess this means my submission a couple hours ago won't go through (dangit, Wired!)...

Here is the official press release and the list.

There is a lot of good points to note all around. The first is the G5 Terascale cluster at Virginia Tech at #3 (10.28 Tflops/s, 2200 CPU, Infiniband) is the first academic computer to break 10 teraflops/s. This extra performance was promised at Mac OS X Developer's conference last month. Not to sure if the price is a testament to Infiniband ($1.5 million cabling, cards, and routers) or the Macs ($4.2 million list).

Good thing too because in a surprise move the NCSA cluster made the list at #4 (9.82Tflops/s, 2500 CPU, Myrinet). This cluster is built using Dell's running Pentium 4 XEONs and Red Hat Linux! One subtle point to note is that they didn't get all the systems online in time (there should be 2900 CPUs, not 2500). I bet some programmer at PSC and an ex-Chief Scientist of SDSC is appreciating having a hand in edging out NCSA for #3--not to mention Apple beating Dell for #3.

The fastest Itanium cluster is at #5 (8.63 TFlops/s, 1936 CPU, Quadrics) which is looking like the odd man out boxed in by a PC based systems using Myrinet, the P4 Xeon above, and the most powerful Opteron system at #6 (8.05 Tflops/s, 2816 CPU, Myrinet). Another point of similarity:did I mention it's also using Linux?

And finally, It's easy to overlook #73, a single compute node of BlueGene/L (1.44 Tflops/s, 1024 CPU). Imagine 128 of these connected together and you have something that will easily take #1 when it's completed even if we handicap it 20-40%. As noted on SlashDot earlier, this will be running Linux.

The Top500 site lists two competing 64bits architectures-based clusters: the Integrity rx2600, with 1938 Itanium2 at 1.5GHz (must be pricey), and an 2816 Opteron 2 GHz cluster, that achieves only three fourths of Big Mac's performance. Now that's a defeat for AMD.

Also, the VirginiaTech cluster is the only "self-made" supercomputer in the Top50 (the next one is ranked 63th, based on SunFire V60). The original #3 slipped to the 7th position because of the new supercomputers. Competition for that third place was tough !

Now where's the G5 XServe ? It was supposed to be out when OS X Server 10.3 was released.

That said, for what is provided, the Earth Simulator seems to be the current king by about 2x. (Corrections appreciated.)

Clicky for the official November list

It should be interesting to see where this system will rank. The 22nd top 500 list will be announced this week at the Supercomputer conference. Further details on the top 500 list can be found at the Top 500 list.

This project is a way of grabbing market share in the supercomputer market, not any other market.

Oh, but wait a minute. That's right... IBM has over 30% of the server market. More than HP, Dell, and Sun. But they're 0.6% behind HP in supercomputers! They'd better getting working on that poor market share.

• #0 - 58 TFLOPS - SETI@HOME
  
• #1 - 35.8 TFLOPS - NEC Earth Simulator
  
• #2 - 13.8 TFLOPS - LANL ASCI Q - HP Alphaservers
  
• New #3 - 9.55 - Big Mac
  
• Old #3 - 7.6 LLNL - MCR - Xeon Cluster

I've always been sort of intrigued by Top500 Has there ever been a good comparison written about the similarities/differences between a 'supercomputer' and the regular pc sitting on my desk running Linux/2k? At what point does the computer in question earn the title "Super"?

If you examine the 21st top500 list at http://www.top500.org/, you'll see that the first AMD cluster is #80 with a Rmax score of 825 GFlops, which isn't a contender to anything really. The first AthlonMP cluster is #84 with 794 GFlops. IOW, AMD isn't in the race.

AMD has the most price effective clusters bar none

Before anyone can advance anything like this we will need to see something bigger than what's on this list, which does not exists or is simply too slow to reach top50. We're talking big boys here, not children toys. I don't think an AMD cluster with the potential of reaching higher ranks than the ones on the list could go unseen. Unless you want to put some $ on the table, build one and prove what you're claiming, or simply provide us with some references, i don't see any sense in your claim.

Do a little research before you start flapping about the Mac guys

A little research!? Did you do yours ? I don't think so. Nice troll.
(NB. I'm an AMD user myself, but i don't have senseless opinions like this.)

Why build a more expensive and slower cluster on the G5 when you can have it faster, cheaper, and far more scalable with Opteron
Zealot-like (vendor) statement... ;-)

The deadline for submission to the Nov 2003 Top 500 list was Oct. 1st (see call for proposals), so it has already passed. Any further improvements that they make to the scalability of the cluster should not be included. This is true for all the machines.

Here is top500.org's list, upto date as of 2003/06.

Everyone should take notice that an ES-class Alpha cluster is at 2nd place while the only Itanium2-type cluster to ever compare comes in at 8th place and then 35th place. The first model of Itanium doesn't even appear on the list until about 111th place! Itanium2 may be a huge step from the first Itanium, yet Top500.org clearly shows that Alpha technology is superior to Intel's Itanium. A few people in this forum are critical that Intel would start producing Alpha if the Itanium2 flops, yet I seriously doubt any Alpha workstations would be made for average availability. The only customers to receive an Alpha would be large coroporations willing to pay the high price, and I seriously doubt that Intel would want to put in the hands of customers a technology they want replaced with their in-house flop Itanium2. The purpose of Intel is to buy their competition, then the market, and raise the price. Intel doesn't want any complete computer being sold that doesn't have their brand of technology in it. Low-power technology included. I hear VIA's Nehemia architecture and Transmeta's Crusoe technology are verry good replacements if you don't like energy-sucking Athlon and Itanium, but there is a limit to how low power you can go with seriously suffering performance.

I only see one in the list, but I didn't look too hard.

Intel Xeon's can do 2 (64 bit) flops per clock.

Look at the 'Rpeak' reading on the top500.org site here. This is the theoretical maximum for the cluster. Take the #3 Intel machine. It has 1152 dual processor machines for a total of 2304 processors at 2.4 Ghz. So, 2304*2*2.4 = 11,059.

An Itanium, BTW can do 4 flops per clock, and that's why a 1.5 Ghz Itanium competes with a 3.06 Ghz Xeon on Linpack

This is tremendous advertising for apple, but what about clusters of Power4's and 5's? why wouldnt they out-perform this cluster?

Rmax is the best it's attained; Rpeak is the theoretical peak in a perfect world with 100% multiprocessing efficiency.

One of the really amazing things about the NEC Earth Simulator (aside from its sheer power) is that its Rmax is over 85% of its Rpeak. A lot of other systems only get Rmax of 60-70% of Rpeak.

I've read that PowerPC chips are pretty efficient in SMP scenarios, and if "Big Mac" is turning in an Rmax that's 80% of Rpeak (even on a fairly small number of processors), that would seem to validate that assertion.

One Power Mac G5 2.0GHz Dual (introduced in June 2003) has an Rpeak of 16 (if you look solely at the multiply-add operation pair, thus giving each FPU 2 operations/cycle). And Virginia Tech went and got 1100 of them. I guess they didn't want their system to "fall off" the bottom of the list for a while.

Isn't progress wonderful? We live in a world where you can go to the mall (if your mall has an Apple Store), call up a mail-order place, or even order on-line, and get yourself an aluminum box with enough goodies inside that a mere five years ago it would have been on the Top500 list.

Um, red face. I don't have a link. I worked it out a while back looking at benchmarks on multi cpu boards, but assumed it was common knowledge. Isn't it?

Have you tried your line of analysis on the clusters that exist in the top end of the top 500 supercomputers? If you want to find a real relationship, this is probably the best place to look, since the theoretical maximum will be approached asympototically by the teams that get numbers placed in this list.

It would appear that 15-20% more gets done in a G5 processor cycle by these figures.

If the "proportional to sqrt(processors)" relationship is true, then I guess I would agree. But I think that I might have to dispute that relationship.