Slashdot Mirror

http://www.top500.org/lists/2005/06/overtime.php?c =8#

You can somewhat sort out by operating system by looking at this pie chart:

http://www.top500.org/lists/2005/06/overtime.php?c =8

It's now about less than 350 systems that run Linux (of the 500)!

Well I beat you there sister!! I am logged on to #38

Supercomputer #72, at the Chinese Academy of Science, comes from Lenovo. I wonder how far ahead IBM's sale of its "PC" business to Lenovo has put China's computing industry. And I wonder just what kinds of simulations they actually run on the beast.

So who here has access to one of these monsters?

I'm logged into #367 right now.

With all the more pressing issues for which supercomputers can be used, I don't believe that China is using the 18th fastest computer for weather forecasts. At least not the ones they publish in Xinhua, anyway. Is there any verifiable way to tell what that machine really does?

Look at the trends, though. Power's Top 500 system share peaked in 2000. For better or worse, Intel is on the rise...

You'd think that it would be a good idea to actually link to the html list, or the xml list, or the pretty charts.

The press release is interesting too.

You'd think that it would be a good idea to actually link to the html list, or the xml list, or the pretty charts.

The press release is interesting too.

You'd think that it would be a good idea to actually link to the html list, or the xml list, or the pretty charts.

The press release is interesting too.

You'd think that it would be a good idea to actually link to the html list, or the xml list, or the pretty charts.

The press release is interesting too.

You'd think that it would be a good idea to actually link to the html list, or the xml list, or the pretty charts.

The press release is interesting too.

For those that thought they might like an actual link to said list:

Here ya go.

And here's a link to the actual list. Also interesting is the historical chart of the TOP500 by manufacturer, which tells a story in itself -- the decline of Cray and rise of IBM and Hitachi, for one.

And here's a link to the actual list. Also interesting is the historical chart of the TOP500 by manufacturer, which tells a story in itself -- the decline of Cray and rise of IBM and Hitachi, for one.

How about a link to the new list? Yes, it can be found from the TOP500 website, but that link was missing as well.. clicky clicky.

How about a link to the new list? Yes, it can be found from the TOP500 website, but that link was missing as well.. clicky clicky.

The list can be found here:
http://www.top500.org/lists/plists.php?Y=2005&M=06

Take a look at the Linpack scores of the Top 500 supercomputer list. It's one of the fairest comparisons around, every processor optimised to the max, done by professional CS people proving only they're machine. You'll see that they are totally consistant between listing of the same processor type too.

Take the Rpeak and divide by number of procs.
What do you get?

2.3 GHz 970 (G5) = 9.2 GFlops (#7 on list)
2.2 GHz Opteron = 4.2 GFlops (#17 on list)

G5 smokes the Opteron by 2-to-1.

Are you sure they actually "switched" Tandem to Alpha? As I recall, the Compaq plan was to include the hardware-specific requirements for the Tandem "CPU-HA" process in EV8. Needless to say, with EV7z released as the last Alpha CPU, how did Compaq/HP ever use an earlier Alpha for the Tandem CPU cluster?

Of course, I remember some sort of Itanium discussion saying Itanium3 (or whatever it will be named) was to have this capability; so I fully expect Tandem under Itanium to go forward, just as OpenVMS under Itanium has gone forward and HP-UX under Itanium has gone forward; see the trend?

For those who live the PA line (I no longer am one of them), while 64MB sounds like the worst kludge I have ever heard of for fixing a latency problem, at least it is being shared between two CPU cores. And to be fair, Power5 uses a 36MB L3 cache - http://www.top500.org/ORSC/2004/power5.html - for similar reasons. Earlier PA CPUs had as much as 8MB of cache for a single core when other CPUs had 512KB to 2MB - so I would say that this just continues a long history of HP creating CPUs that do well in benchmarks by loading the test configuration with as much cache as possible - usually much more than the competition.

Yes, this is a troll for PA lovers; my only consolidation for both you and the Alpha lovers is your engineering will continue to live on in various Intel CPUs, especially Itanium. So neither CPU is dead, just "outsourced". What else would you call sending all your skilled CPU design engineers to Intel?

Y.A.C.C.

Take a look at top500.org This is the top 500 supercomputers benchmarked on linpack. I use it because it is one of the fairest benchmarks, since everyone on the list is optimised to the max. Now take the Rmax number and divide by number of procs. You'll find from system to system using the same proc that they're very consistant.

I gave the ibm paper before. Look at the power curve (second graph down), follow the speed down to 1.6 GHz, now what is its power? ~22Watts.

Now from the curve you can see that cranking the GHz on the 970 makes the power increase exponentially (like all processors). So at 2.7 GHz it really is hot, not P4 hot, but very hot.

The reason for water cooling is low noise, not that it is sucks more heat than a P4. The reason no G5 in a laptop is??? If you look at the same power curve the G5 at 1.25 GHz is around 15 watts not much different than a G4. Perhaps Apple thinks paying more for a G5 without a really big performance or GHz boost isn't worth it.

Well, it ain't UNIX that runs on those things

Of course its not Unix. But Pttttbbbttttppphhhttthh anyway.

If the X86 hardware is faster, then why doesn't it outperform the Xserve clusters using Mac OS X and standard Apple hardware? Don't mistake bad benchmarking for good data. Don't believe me, go to the Super Computer web site (http://www.top500.org/lists/plists.php?Y=2004&M=1 1) where 5 of the top nine are PowerPC hardware. Number 7 on the list is built by using stock Xserves ordered directly from the on-line Apple Store and running Macintosh OS X. And, there are even faster Apple clusters in the wild that didn't even bother to try to compete for a ranking. These are not marketing ploys. These are very expensive machines that are optimized to go as fast as they can to get the job done. No political bias...just go fast. The only X86 cluster in the list above a ranking of 9 has nearly twice the number of processors as the Virginia Tech cluster. Maybe you should spend a little less time pounding out Visual Basic programs and do some actual research before making statements like that.

The PlayStation 3 will feature the much-vaunted Cell processor, which will run at 3.2GHz, giving the whole system 2 teraflops of overall performance.

Microsoft has it written on their official Xbox site. I don't know.

This is a top list of super computers. Speed there is in GFlops, so divide by 1000 to convert to TFlops.

Going down to the first super-computer that has about 1TFlop of THEORETICAL maximum speed, I see this:

SGI Altix 1.3 GHz
208 processors (!)
Year: 2003

That's not shabby for the Xbox 360 to match up to. Or, put another way, it's only 1/70th the power of the fastest super-computer in the world.

Maybe YOU shouldn't try to troll against Apple - you chose the lowest system that you could find on the list rather than the Virginia Tech one that they were advertising in the first place. The one that Apple actually boasts about is number 7, NOT number 444! http://www.top500.org/sublist/System.php?id=7286 is the actual listing you should have shown. You may call me a troll for posting this, but I think the parent is much more of a troll for posting misleading stats

It might be useful to buy the macs if:
You didn't want one box.
Apple didn't put compulsively lie about it's performance http://www.top500.org/sublist/System.php?id=7309

If the object was to have the cheapest system, you'd be using racks of dual 1U p4, at 1.5x the performance of xserves for half the price.

Ease of use means very little when you're trying to get work distrubuted to > 1 processor effectively, it's like adding a picture to the finished product in a 300 page instruction manual.

Once again, lapack isn't used to rank supercomputers as you assert. It's a linear algebra library. As far as I know, that's it. Linpack is what you're thinking of. More specificly the high performance linpack benchmark. If you don't believe me, refer to the top500 site. It's right there on the front page.

And cmon man, it's not about hate, I'm just trying to correct some misinformation here. Libraries don't just up and do something by themselves. Or what, you expect the guy to write his own cpu burn in program? Why reinvent the wheel?

P.S. It's called blas, not lablas.

It's even cooler to look at the performance improvements of supercomputers. They double their speed faster than every year with amazing regularity. The top 500 supercomputers had a total processing power of 1.12 Teraflops in 1993. By mid-2004 they were at 1127.41 Teraflops. Look at the graph, it really is impressive.

It's even cooler to look at the performance improvements of supercomputers. They double their speed faster than every year with amazing regularity. The top 500 supercomputers had a total processing power of 1.12 Teraflops in 1993. By mid-2004 they were at 1127.41 Teraflops. Look at the graph, it really is impressive.

Were you correcting my spelling? Because I always make that mistake (myranet... it's myrinet damn it!). You know what I meant though ;) It looks like BlueGene/L is using a hybrid backplane/hypertorus interconnect where a whole bunch of "machines" (more like system-on-a-chip) are connected via a backplane, then that case of "machines" is connected to another case in the same rack on some number of layers of interconnect. Then the racks are connected using some other protocol. Though you may not "be at liberty" to discuss this, the top500 site already disclosed an ample amount of information on the subject for any beowulfer to get the general idea of what type of interconnect topology/setup BlueGene/L is using.

And I quote:

The nodes are interconnected through multiple complementary high-speed low-latency networks, including a 3D torus network and a combining tree network. The physical machine architecture is targeted to be most closely tied to the 3D torus, a simple 3-dimensional nearest neighbor interconnect which is "wrapped" at the edges. An independent combining tree network provides for fast global operations, such as global max or global sum.

http://www.top500.org/sublist/System.php?TB=2&id=7 101

Enjoy.

FYI the top 500 supercomputers list is maintained at http://www.top500.org/.

And try this; go to your local CompUSA and ask for a PC mobo with dual independent 1.2ghz frontside busses and dual processors fast enough to use them. You'll need to go to the Apple part of the store for that.

I work on Macs and PCs and the Macs feel and act like a giant Swiss watch while the PCs feel like crap.

We buy/build plenty of computers in our company, hundreds in the last 10 years, Macs and PCs. With rare exceptions, PC laptops are broken between 8 and 18 months and Mac laptops generally outlast their usefulness (5 years) and we sell them to the staff. The PC laptops are in the dumpster way sooner just by trying to use them. The Mac laptops that broke got knocked off a table, so they're little headless servers now.

Every Mac desktop in our company has outlasted 2.7 PC desktops. The PCs are dumpstered or turned into metal cases ready for new hardware typically within 2-3 years. Some PC cases are on their 3rd life.Practically everything blew up in them from the power supply to the crappy bypass caps on the mobo. The Macs outlast their useful life and they're sold to the staff for $100. We've had exactly four Macs just totally up and die in the last 10 years compared to dozens of PCs. Macs and PCs died in lightning storms but most of the PCs just started flaking out, crashing and eating their faces.

I just removed an old Mac 6100/60 that's been running as a heavily used server since 1994... along with four others as workstations for that server, all still running, and replaced them with Mac Minis. I have a glut of ancient but still operational Macs and just don't have that problem with PCs.

We were getting sick and tired of PC problems and out of 100 or so current computer users, I've seeded 35 new Macs, most of them to the PC users. They bitched and moaned for about two weeks and I haven't heard from them since. Many have discarded their home PC and bought a Mac... and an iPod... and THANKED ME for their new freedom. Really. I'm a freakin' hero.

What happens when you stack a few thousand processors of all different kinds into a room, compare the performance and compare the price tags. I did that comparing the clustered "Supercomputers" measured in November of '03 (http://www.top500.org/lists/2003/11/top5.php) which has the VA Tech room full of 2.0ghz DP G5's at #3 in the supercomputer lineup. Not much has changed since then but if anyone wants to redo this with the current lineup, have at it, but this is basically what people argue about and this many processors kind of averages out what's real:

#1 Earth Simulator, Japan, 35.86 TFlops
5,120 (680 8-way nodes) 500 MHz NEC CPUs
Cost: $350 million

#2 Los Alamos National Laboratory ASCI Q, 13.88 TFlops
12,288 EV-68 1.25-GHz CPUs (3,072 HP AlphaServer SC machines)
Cost: $150 million

#3 Virginia Tech's X, Terascale Computing Facility, 10.28 TFlops
2,200 IBM PowerPC 970 2.0 GHz CPUs (1,100 Apple G5 machines on OS 10.2.7)
Cost: $5.2 million

#4 NCSA Tungsten, 9.819 TFlops
2,900 Intel Xeon 3.06 GHz processors (1,450 Dell PowerEdge 1750 servers on Red Hat Linux)
Cost: one of four NSF TeraGrid clusters totaling $53 million.

#5 MPP2 Pacific Northwest National Laboratory, 8.633 TFlops
1,960 Itanium 1.5 GHz processors (980 HP Longs Peak nodes, also known as the HP Integrity Server rx2600, on Red Hat Linux)
Cost: $24.5 million

Hmmm...

If they built up the original VA Tech Apple system to cost 10% as much as the Earth Simulator, it would theoretically run twice as fast as the Earth Simulator.

The 2.0ghz DP Apple used at VA Tech has about 75% the speed of the ASCI Q using only 18% as m

But what the heck did Larry Ellison smoke this morning? Larry, go read your own web page, look which platform makes the fastest growth for you at the moment, and think it over again.

Or EMC - why did you buy VMware to begin with, if you think that the ESX server base OS does not scale, and is insecure? Oh you added your own patches? Okay - where is the source code, and how about forks?

Hey Cisco, do you have a problem with Asterisk in the upcoming VoIP hype?

And Sun... well they advertise SUSE Linux and Red Hat Linux on their web page, but I didn't expect these free riders to actually *support* Linux. These folks turn any discussion from Linux as a teaser to Slowaris x86 within a wink of the eye. The fork discussion is particularly interesting. As far as I remember, it was not Sun Microsystems who prevented the UNIX fork from happening in the 80s and 90s, despite all the XPG4's, SPEC1170's, Single UNIX Specifications and all these other marketing smoke grenades. "McNealy ... finally announced he won the battle and had the one surviving Unix out there." Scott, go sleep again. Or get sued by SCO because they allegedly own Unix.

Scalability problems is a nice thing. Go read the top500.org list, see what OSses are at the top, and come back then. Hint: it's not Microsoft Windows, and it's not Solaris, especially not the system on rank 2 (has anybody ever seen a 10000+ CPU cluster even *boot* Sloaris or Windows?). The top 10 were not built by EDS, EMC or anybody else on this list. There is *one* Windows system on the list - the entry is 15 months old and fell from rank 84 to 194 within one year. Apparently, nobody dared to run another Windows HPC benchmark ever since. The fastest Sun machine is on rank 31, and guess what, it's not an UltraSPARC box, and did it run Solaris? So far for scalability. BTW: The fastest UltraSPARC machine is on rank 32, and it's not a Sun box, but from Fujitsu-Siemens. Watch your six, there is an Alpha machine behind you!

They assume their customers are deaf, dumb, and blind. Disgusting.

Yeah, nobody has this except Intel with Hyperthreading and IBM with SMT on the Power5 chips.

http://clusters.top500.org/ORSC/2004/power5.html/

I could either buy one Sun Fire V490 or two IBM P550s. Guess which way we went.

It's 70.72 teraflops.

It reminds me of what SUN was talking about in this.

Jonathan Schwartz must be happy to see that finally, his idea of selling cpu time is being realised (and how much he loves IBM ;))

Anyway, even if, I guess, the price will be lot higher than Jimi Hendrix (and that's something), the few people getting access to some of the best performing supercomputers is really nice.

To sum up : nice business plan.

You maybe slow so here's another link with less text so not to confuse as much as the previous link. Reading is hard isn't it?

Here's a top 500 server that runs windows. Buy a clue thanks.

• Designed
• Ment
• Capable

They haven't released any machines with dual core CPUs, because none are available.

None are available?

IBM's had multicore POWERs forever, at least since like 2002, I think before. The G4 and G5 have both had the technical capacity to be made in a multiple core configuration. I think Apple isn't interested in dual core because getting dual core PPCs would have been as simple as just asking IBM "hey, could you start making us some dual core PPCs", but they haven't bit.

you are right. Who would want to use an IBM powerpc processor anyway?

I happened to look at the Top 500 supercomputers site and I couln't help noticing out of the top 5 supercomputers almost half are in non-US countries like Spain and Japan. This is not to beat some kind of patriot act drum. Instead, it got me to thinking.

With supercomputing powers now avaible to any country or group with a few readily available components, it is only a matter of time before these supercomputing powers may be used by a rogue state or radical group to cause havoc among electronic communications using methods like denial of service attacks, spyware, and crapflooding message boards.

I think it is high time the nations of the world put their heads together and addressed this issue. For example, I don't think the US Federal Government even has any cabinet-level position like Secretary of Information Technology or something like that. When are they going to get with the times? It will probably take another terrorist attack or something.

I think you need to adjust your list...

Check here

Linux is not ready for mission-critical computing. There are fundamental things missing,' pointing out the lack of a development environment and no single 'sign-on system' giving reference to Microsoft's foundering .Net passport program.

Oh, shit! They could have said it before! We've just bought the forth most-powerful computer in the world from IBM with Linux and we won't even get MSN Messenger accounts!?

Let's just wait until a AMD64 cluster hits http://top500.org/ before we start saying things like "AMD64 is obviously better then the PPC chip"

Just use good passwords. Anyone using all lowercase letters for their passwords probably isn't too worried about it. I mean, if it's easier for you to add another character to your password than to use uppercase, then by all means do so. But lowercase + uppercase + numbers factors out to:
((26 + 26 + 10)^15) / (70.72 terahertz) = 344 539.008 years Add symbols to that and it will practically never be broken. You'd need a 19-character all lowercase password to almost match the possibilities of a 15 character alphanumeric case-sensitive password.

Keep in mind that these are pretty optimistic guesses. The only thing I forgot to add in that would cause lesser time is the statistic factor that your password will not be the last one they check, so that'll average out to half the amount of years. But I'd be extremely amazed if they could generate the next password, and check it in one floating point operation. Plus, most people worried about it use a combo upper/lower, plus symbols, plus numbers.

Also, you're saying they could store all the possibilities, that's really not mathematically possible either, assuming you're using a good crypto algorithm/good hashing algorithm(*NOT* MD5). The goal would be basically to find a password that hashed to the same value as your password, so for every possibility of the hash function, they'd have to store a password that would match it. Simply storing one byte for every possibility of that hash would be near impossible with today's storage: 2^64 * 1 byte = 16 exabytes (And that's for a 64-bit hash, even MD5 is 128-bit (though MD5 is insecure for other reasons)) I don't think there's that much storage in the world, and even if there is, there's certainly not enough to store even 8 character passwords, using a 256-bit hash, and especially not at one institution.

BTW, an exabyte = 1024 petabytes = 1024^2 terabytes

You also mentioned that the computer was "working alone". Looking through the top 500 list of supercomputers, there's 398 computers in the world above 1 teraflops. Let's be completely over-optimistic on the computing power, and say there's 400 computers operating at 70.72 teraflops working on your password alone. That'd still only reduce the amount of time it would take by a factor of 400. So you're looking at 861 years

It's just not going to happen that your password is brute forced if you follow password guidelines, and make it decently lengthy. I'd highly doubt that your data is worth the world working on your password for 100+ years.

Don't get me wrong, the first-world governments have a lot of computing power. But you're underestimating the time it takes to brute force. You'll be much better off securing your system in other ways than worrying about lengthening your password.

The difference here is a)This article is not a claim, its a poor distillation of a patent app and b) This other company is involved. This list seems to indicate they might know something about fast processors. And possibly more than the average slashdot poster.

True, you can build very large clusters from these bricks, but the bricks themselves don't scale beyond a relatively small number of CPUs.

If you want to build a supercomputer for floating-point-heavy scientific computations, and cost is not a factor, you really have two options: the Itanium, or a NEC vector processor.

I suppose all of those Power-based (Power4, G5, etc) machines in the top 10 of the http://www.top500.org/ are just makin' those numbers up, then eh? And at such dramatically low cost, too!