Slashdot Mirror
Compaq to Build Alpha Supercomputer
kfarmer@tru64.org writes, "The French Atomic Energy Commission has placed an order for a supercomputer to simulate and analyze nuclear explosions. The supercomputer will use about 2,000 Alpha chips running in the 1.25-GHz range, or about 2,500 chips at the 1-GHz level."
No, it's the French, their testing grounds were in the Polynesian islands. All the crap from their nuclear testing is now spread pretty evenly around the southern Pacific Ocean. I certainly wouldn't want them to start up real testing again, let them have their computer :-)
----
----
Open mind, insert foot.
Aarrrggghhh...
It was not MS that gave up - it was Compaq!!!! Compaq decided to drop support for NT on Alpha!!
Who do you think had to maintain Alpha/NT? Clue: Not MS. Yes, DEC/Compaq had to pay for a complete NT software development dept., because DEC/Compaq had to do the maintenance. (same thing when NT used to run on PPC - IBM had to maintain it, until they realised it didn't sell).
Then some bright spark looked at the figures and realised that nobody was buying NT on Alpha. It's best marketshare was on workstations - ~15%!! On servers it was even worse - because people who tend to buy nice hardware like Alpha also tend to buy nice OS's like OpenVMS or Unix. People were not spending money on Alpha/NT.
They put 2 and 2 together and realised that paying for NT/Alpha was costing more than the revenues generated by Alpha/NT sales. And that's why it was scrapped. The biggest money maker on Alpha is Unix, closely followed by VMS.
Also, look how hard compaq is pushing Linux on Alpha. This is for the same reason as why NT was dropped - money. Linux sells a lot of Alpha's.. esp in the lower end, eg Linux marketshare on DS10's is about 40% or higher... it also does well on clusters. And Compaq is pushing linux/alpha clusters really hard.
(my mouse mat is a picture of tux on a fat motorcycle with the Compaq Alpha logo, and a banner saying "Linux SCREEEEAMMMS on Compaq Alpha".. this is an official compaq mousemat)
:)
in fact this Alpha cluster will most likely run linux..
I use Friend/Foe + mod-point modifiers as a karma/reputation system.
Furthermore, I am not saying that an Alpha chip is like a Pinto and an AMD chips is like a Porsche
I surely hope you don't... An Alpha at 150 MHz is roughly equivalent to a PPro/PII at 800 - 1000 MHz, for floating-point intensive programs. Partly because the Alpha is just blazingly fast for FP operations, and partly because the x86 FP architecture sucks. When they developed the 8087 originally, they chose a design which is flawed from beginning to end, but they couldn't get it to work any other way... And now we're stuck with it.
it kinda reminds me of an old Star Trek (1st gen) episode where some planet was in the midst of a war, but it was all computer simulated, and whoever lost a virtual 'battle' had to send a bunch of people to a death chamber.
So if China ran a simulation of an ICBM launched at LA, pitted against a US simulation of a 'star wars' ABM missile trying to knock it out of the sky, and the US missed, we'd have to bump off a bunch of LA residents, all with no messy radiation or destruction of property!
try { do() || do_not(); } catch (JediException err) { yoda(err); }
From what I have read, the tricky part of computing ballistic missile trajectories is having to use a high fidelity model of the gravitational field around the Earth. There are also questions about the accuracy of existing models. I'm not sure how atmospheric effects are handled, this was a major problem for the USA during World War II. Precision bombing missions by high altitude bombers often missed targets by large distances due to wind effects.
Mea navis aericumbens anguillis abundat
***
The Spanish intercepted these French intercepts and are still pondering a suspicious code phrase: "noleche"
***
Luckily the Spanish found their Franglish translator which revealed the secret message: I'm not licking.
(from lécher, to lick)
I was thinking noleche was a secret French message instead of a secret Spanish message. Don't speak Spanish so I defer to you on the he's not licking theorem
=)
it seems to me that Compaq are going to sell a computer that is going to analysis the destrution of the south pacific.
If this means I'll never hear 'there's nothing like a dame' again, that can only be a good thing...
Alternatively, is it a "real" supercomputer, with a scalable high bandwidth low latency interconnect?
If so, it makes me wonder why they just don't buy a Cray...
Chris
Don't you mean SPECfp? That's where alphas have the most advantage. At similar clock speeds, the SPECints aren't that far off for a P3 and Alpha (IIRC)...
"It's tough to be bilingual when you get hit in the head."
Something that breezes through nuclear calculation could probably brute-force crack most encryption methods in an afternoon...
That would be very bad crypto, in that case. The keyspace enumerable (ie, assume a key checked per cycle -- obviously a best case scenario) for a 1 TeraOPs machine is just shy of 40 bits per second. So in two seconds it does 41, in 4 seconds it does 42... in 24 hours, just over 51 bits. Since you on average only need to search the half the keyspace, the best this machine could hope to crack using brute force is a 52 bit key per day.
But they were talking about eventually scaling up to 100 odd TOPs. 128 == 2^7, which brings it up our searchable space to just under 60 bits.
At that rate, You'll need to search for 2^68 days to break a 128 bit key. That's a pretty long time. Go calculate it for yourself. Oh, I'll do it, it's 10^20 odd days.
Johan
Well, the first few millionths of a second. In a multi-stage device you may have 3+ seperate fission and fusion reactions triggering each other at exponentially greater yields.
Simple devices use primary with plutonium core compression fisson device with tritium core. X-rays escaping the primary tamper are focused via beryllium into plastic/foam/explosive waveguides to generate enough energy in second detonation to heat lithium-6 deuteride or other fussion fuel sufficiently, and outer bomb casing is often made of fissile material for free extra bang (and higher burn efficiency of lithium). An extremely complex system. I always thought it was cool how in "dial-a-yield" by simply varying the tritium in the primary core "pit", you can change the final yield by a good order of magnitude or two - for tactical situations vs. simple inventory.
Handling propogation of energy between stages introduces signifigant aspects of materials sciences, and is one of the more interesting problems with maintaining a stockpile - how things change after sitting and bombarding itself with low level radiaition for years on end. Lots of trace isotopes appearing, opacity to x-rays changing [the internal power exchange medium - "high temperature photon gas"), _very_ neat stuff.
As an aside, the utilities to work through these exchanges are what Dr. Lee is accused of losing his backup tapes of. Rather important stuff, as higher efficiency means more yield per launch platform.
I agree that Compaq is doing great work when it comes to servers, espcially with the alpha stuff.. but I cannot and will not forgive them for the poor craftsmanship, quality, and engineering behind any single Presario you can pull out of Compaq's arse. I personally have a Presario 4640 (P2 266) No fan on the CPU but a case fan in the front blowing across it. This is fine except that they put the shortest possible floppy cable they could in the box and to fit it was stretched across the surface of the fann completely blocking airflow. How is that for engineering? Speaking of Presario's has anyone gotten Linux to run on a model near mine? Any version of RedHat from 5.2-6.1 has a kernal oops when formatting a linux partition during the install. I got slackware 3.5 up on it but every 2-10 minutes it complains of dev/hda being out of sync and restarting the drive. YOu subsequently hear the drive spin down and back up, halting disk IO for a good 5-15 seconds. The drive is fine so I can't figure it out.... any suggestions would be greatly appreciated.
www.mp3.com/Undocumented
I'll do it for cheesy poofs.
You do this so you don't have to actually set off a nuclear device to predict how it will perform. You must do a number of real nuclear tests to get baseline information. After that, you can do computer simulations of similar explosions. These simulations are VERY processor-intensive (like weather prediction) and require these large parallel systems to compute.
The U.S. does the same thing with massively parallel systems at Sandia National Labs and Lawrence Livermore National Lab. Check out the list of the top 500 supercomuter sites in the world -- http://www.top500.org/ -- to see who's doing this kind of thing.
Even if this is a powerful machine, I'm afraid it will need a VERY long time to boot...
Especially on NT : Windows has detected a new CPU.. please insert Windows NT CD-ROM... Windows has detected a new CPU....
Fetchez la vache !
The machine will be built out of 4-CPU nodes based on the ES40. The OS will be a customized version of Tru64 5.0, which has VMS-style clustering. This will make the entire cluster appear as 1 machine, i.e. the OS only has to be installed one time. The CPUs will be the new 21364, aka EV7.
--
http://gammatron.weblogger.com
I would rather have them simulate it than to start nuking the Nevada and New Mexico plains again.
More race stuff in one place,
than any one place on the net.
Well, when I think about it it is amusing that I've been duped. Before amusement becomes irritation. Not with the origional Troll, but with oneself. I couldn't believe I could be so stupid as to respond to it.
I don't think that Trolls should be ignored. If there is a good Troll, I like to respond in a humourous way. This time I failed.
I know what you mean about the moderators not understanding insightful. I've posted about one insightful comment, and had dozens moderated up as insightful. What I don't quite follow is why you're so angry about it. Ordinary people get to moderate. People are fallable. Not everyone does know what insightful means. Hell, most people even get the meaning of the word "instantaneously" (to choose a random example) wrong too. This is because nobody knows everything. This does not give you just cause to correct minor errors with pedantry. It is no reason for swearing. And if you think good trolls shuld be encouraged, then GREAT! respond to them with anger. That's what they're there for.
And as you can see, I (Neil Sluman) have responded with my name. Happy now?
Once again, the difference between architectures and software slips through the grasp of the media...
Also, I could have sworn that there were restrictions on the computational power that we could export from the U.S. Something that breezes through nuclear calculation could probably brute-force crack most encryption methods in an afternoon...
It's not for analyzing, it's for simulation and emulation to verify designs without physical testing. The simultanious equations to do these things are pretty extensive and a hugely parallel processor is very useful. Remember, all the interesting things happen in the first millionth of a second, beyond that it's an expansion/compression issue of the blast propogation.
Where would this be placed in the current supercomputer ranking?
My question is:
Why would you want to put so much into analyzing nuclear explosions?
I can see for weapons testing and maybe just out of scientific curiosity. Are there any other reasons anyone can think of?
I'm wondering if we can get a Beowulf cluster going for a cheaper price with similar speeds.
These alpha boxen will problably run tru64 in a configuration similar to beowulf, that is a cluster.
I'll do it for cheesy poofs.
Quick translation:
..... The power of 5 teraflops is obtained by the use of the Compaq Alphaserver SC series of supercomputers..... ..... is the first of three steps in the realisation of the nuclear weapon simulation centre. The second step, towards the year 2005, will see an increase to a power of between 30 and 50 teraflops and the last step, 2009, to a machine of about a 100 teraflops.
.....
The installation of this supercomputer
I'll do it for cheesy poofs.
_______
2B1ASK1
Mhz only has any validity as a benchmark within an architecture. To compare across architectures, you must use bogomips!
Repeat after me: Mhz only has any validity as a benchmark within an architecture. And even that validity is limited. A 400Mhz PII is NOT 33% faster than a 300Mhz PII. It's maybe 10%. To talk about Ghz Alphas as though they are at all similar to Ghz Intels is crazy.
You want to share CPU benchmarks on something like this, talk about SPECint and SPECfp. Not Mhz.
--
-- Slashdot sucks.