Slashdot Mirror
Russian E2K cracking RC5
Tuna Phish writes "Apparently the new Russian E2K computer is being used to crack distributed.net's RC5 contest! The user has come out on top the past few days with 2.7GK/s, more than 6 times the keyrate of second place. His message states "Russian Elbrus E2K is REAL POWER!" " Great-now we just need to get them to join Team Slashdot, and he can get all the pr0n he needs. *grin* So, I've been doing some looking around, and am doubting the veracity of this: bogus client? First prototype? Anyone have more info? Post it in the comments.
>Russian Missile Defense System
:/
I thought the avionics on Russian missiles was all built with tubes and not silicon. Due to tech export restrictions in the cold war era, Russia was far behind in semiconductor tech for quite a while and thus continued developing vacuum tube tech. They actually got the tubes down to a pretty small size too and using much less power! Oddly enough, were there ever a nuclear war, much of the US defense would be decimated by EMPs from the first attack wave while Russian tube controlled systems would be far far less vulnerable to EMPs and thus would continue to function effectively. Good thing this was never put to the test, though.
I believe distributed.net has the ability to eliminate packets processed by particular users, so it shouldn't queer the contest -- IF it's a fraud, that is. (Nice data rate, though. :-)
and according to the guy who started it they will doble their rate during the next days..
:and according to the guy who started it they :will doble their rate during the next days
Quite an abuse to their customers, I say.
Chat.ru is an russian isp, they've trojaned their
customers' machines with a hacked client, perhaps
it runs under BO.
Why? You don't know that it can't spit out 3 keys per clock cycle.
Russia doesn't even have modern fabs.
If they do manage to design a fast processor, they still have to worry about routing, yield, etc.
In any case, the (post-Merced) McKinley design is supposedly nearing completion. It should match the same speed as the E2k.
..and we see how amazing their computers must ;)
have been, since Mir was _real_ stable.
Easy, it's not U.S. ... that is bad, it's Russian, that is even worse. It just must not be.
Transmeta has not made *any* claim on the performance of their chips. They won't even say what they are doing! Anything you hear is just wishful thinking.
The elbrus homepage claims that the speed of the elbrus supercomputers of the sovjet era was mainly accomplished with smart logic instead of increasing component density..
If they've later learned how to do modern processor manufacturing,thus combining their smartness with raw power, I belive that the E2K processor may very well beat Intel's designs.
BTW, If you know how to build a CPU that'll be two or three times as fast as any existing processor of today, i dont think you'd have a lot of problems porting a RC5 client to your machine. =)
The pentium instruction set is a CISC-like set. The underlying architecture breaks each instruction into one or more smaller instructions. These instructions are RISC-like and the architecture is RISC. I believe Intel started to do this with the Pentium Pro, but I'm not sure.
Oh, and whatever happened to SKYLAB? I do believe that it "crashed" quite a long time ago. The Russians deserve only credit for having made MIR last LOOOONNNGGGG beyond it's initial expected life. Or maybe you want to compare to SKYLAB???
there is a LOT of hype starting about this cpu, they said they test linux on it, how cool is that :) if this CPU is as good as they say it is everyone should want it, learn from it, if it is better impliment it, not bit*h about it being russian, after all they were first in space, so i guess they are capable of insight and breakthroughs before americans, amazing isnt it. *rolls eyes* :)
One question maybe someone can answer, would that mean there is a possability the technology from this CPU would be easier to export to countries that have a current technology ban through exports?
No, current pentium cores are more RISC than CISC. Micro-ops decoding, etc.
have a look at the partners theres a few big names in there.
ùíù _Shaman is now known as rc5whs :( :-(
12:28 hello
12:28 Hey dude...
12:28 yep?
12:29 rc5whs, are you on that Union of Cracking Gods team?
12:30 TWP - yes
12:31 How did y'all manage 900K+ blocks in one day?
12:32 huge lan - more then all AnandTech
12:32 Obviously...
12:32 ~20 networks work for us - all city!
12:32 So y'all are going to be able to sustain those numbers?
12:33 How did y'all get permission to saturate 20 whole networks?
12:33 ISP admins and university admins was join and talk about this project
12:33 What city?
12:33 university + commercial firms + ISP's
12:34 Stavropol
12:34 near Chechnya
12:34 Cool...
12:34 not so
12:35 It's NOT?
12:35 Why "not so"?
12:36 Russia has war against Cechnya...
12:36 many terrorists is near us...
12:36 bombs etc
12:37 Well of course I was aware of that (I'm a former US Army Officer).... So there's a
lot of terrorist activity in Stavropol?
12:37 yes... our police every day so exiting...
12:38 Sorry to here that. I spent a year or so in Eqypt and Israel.... Not too many bad
incidents, but a few close calls...
12:39 That stuff must play havoc with the network backbone!
12:41 yep...
By sending it a fake block that it knows is
has they key, and seeing if the client properly
answers. If they don't have it yet. They should. Good way to test various ports as well.
The Union of Cracking gods (5 members) doubled the rate of the next highest team (1387 members). This is difficult to belive. They have suddenly improved their performance 10x.... interesting.
.. thats a possibility. OR there are a few people that have hacked the client. but lets face it, that is a phenomal rate. Too good to be true.
Some one suggested a BO plug-in
Actually, given yesterday's average keyrate acceleration, and yesterdays keyrate, it should take approximately 3.3 years to check every last block, putting us around November 2002.
Actually, the current RC5 project strikes me as a waste of time, and in fact, might even be counterproductive.
Everybody knows that as long as there's not some huge problem, we'll eventually find the key. However, when we do, the FBI/NSA/BB/Gubnen't will say, "It took a jillion computers many years to crack a SINGLE message. Thus, 64 bits must be secure." We all know this isn't true--that a dedicated machine could race through the keyspace faster and more cheaply than the general purpose machines we're using. Still, this is a subtle and difficult-to-prove statement to most of the population.
The instructions for x86 are not necessarily 32 bits. You're probably coming from the RISC world where 32-bit processor = 32 bit instruction word. But with x86 the instructions are all different sizes, ranging from 8 bits to 64 bits (I think).
This is part of the problem with x86: CPU's can't easily look ahead at instructions because they aren't sure where they begin without looking at all the intermediate instructions. One of the big improvements that RISC and IA64 offer is the consistent-length instruction which makes all kinds of things like prefetching and decoding easier for the CPU.
The tagline and the raw computing power points to something like NSA or an equivalent organisation.
/007
They don't have the money, and have not had the time to finish a prototype. Look at how well they have handled ISS! If the union has the money to blow on its processor do you really think it wouldn't goto Mir first? How about thoses living in poverty there?
I laugh at all of you distributed.net freaks. You really think that their first move would be to waste processor time on a silly game.
deft
and what,.. no other cpus are being used to crack rc5? what about intel, alpha, sparc, ppc... w00p
correct me if i'm wrong, but i believe that once the key is found..all 64bit encyrption can be cracked.
Because it's one of the few cars out there that can be repaired in the field, with simple hand tools. No fancy electronic stuff, not sophisticated feedback systems, no obscure bits and pieces.
Well, what a great topic for nitpicking
--ac
when will there be team e2k?
i want on the winning team!
Deep crack only did DES, not RC5-64
I think Deep Crack was purpose-built (hardware optimized) to crack DES, not RC5. If you can find a copy of EFF's "Cracking DES" (O'Reilly) they actually give you the schematics and chip source code for the ASICs they used.
On the other hand, in one chapter they state that it is actually better to implement RC4 crackers (and I'm assuming RC5, though my extrapolation could be wrong) in software. Guess you'd have to read the book -- well I could, but I haven't got around to it yet.
We all know that RC5 can be cracked given enough CPU horsepower. So it takes a few years.
... your choice.
I'd rather be using SETI@home and doing something with those CPU cycles that is a bit more interesting.
Discovering ETI, or somebody's lame encrypted message
Microcode = CISC. RISC processers work without microcode. That's why Alphas and ARMs don't have divide instructions e.g.
They only call it RISC for marketing reasons.
If you could program Pentiums with those "Micro-ops" directly it would be RISC.
Believe me, everything that executes only x86 nativly is CISC. CISC, CISC nothing else.
Let's keep the word RISC clean from this intel dirt, ok?
Funny if this was Linus testing out the transmeta :)
-
stuff
(couldnt be bothered to log in)
PenguinII
-----------------------------------------------
"Annakin! Drop!"
"What was that mister qu-" *SPLAT*
Not a super-duper error; just a confusing state of exisitence.
The PPro did start a generation that takes CISC-lik instructions and turns them into RISC-like instructions.
Still though, we're getting nothing like a Sparc or PPC/G3 RISC going on here.
Well, it'd be nice if:
a) Their clients were decently optimized like distributed.net's are. Ask for help if you need to on how to design a decent client! The differences in speeds between platforms is apalling and seems to be linked to bad coding.
b) Teams? Are the teams better secured now? I certainly wouldn't want to "waste" my CPU cycles on a project that is sending me the SAME data over and over again without telling me. F that.
Anyone remember that reconfigurable "shoot a bullet throught it" super computer from ages back?. I know most of you branded it as fake.. but if it is real then it could chew more keys by it self than all of slashdot combined. If this thing is released in 2001 (as they stated) to the home pc market.. dnet may be finished overnight ;p. And as for the E2K... pfft :)
Thats IF this thing is real.
-
^Zer0^
Ive talked to several members of the 'union'. It appears somewhere around 20 systems city wide have joined the effort, after a meeting among the city sys-admins thousands of computers were all added.
/. pw]
#distributed: [06:29] ~20 networks work for us - all city!
#distributed: [06:30] ISP admins and university admins was join and talk about this project
#distributed: [06:31] university + commercial firms + ISP's
#distributed: [06:31] Stavropol
#distributed: [06:31] near Chechnya
I also talked to citizen of stavropol, seems that almost everyone there has heard of the project, although he had admitted to only have submitted a few blocks himself.
[Forgot my
Matthew Fisch
netsplit@tri-millenia.net
From my understanding, the distributed.net client doesn't do a final analysis; in fact, it can't without human intervention. The client simply looks for keys that, when run on the code block, produce output within the realm of English language. (I think it just looks for all characters within the ASCII normal-use range.) After eliminating those 99.999% or whatever packets, any positive results are sent to the distributed.net machine, where they're further analyzed automatically (language patterns, spaces in the right places, punctuation), and then passed to a human to do the final check.
Because of this, every umpteen packets or so will have at least one "hit" to send to distributed.net for further analysis. As a result, a fake client processing more than a few packets would be discovered and weeded out.
no defense against missiles?
what about the Patriot?
SETI@home only runs as a screensaver, so run both clients and give the RC5 client a high niceness setting. When you're at your computer, it will work on RC5, and when you're away it will work on SETI.
That "lame encrypted message" could get you $2000, you may as well give it your idle time while SETI is not running.
> SETI@home only runs as a screensaver
You must have some really, really bad stuff running your computer. Windows, right?
On real operating systems, seti@home runs in the background.
> The x86 option is just there for compatibility, it's impossible to optimize that much more than AMD and intel have done.
Well, maybe.
The magic is in this paragraph:
"The Elbrus team has also devised a mechanism to enable x86 programs (written for
Intel computers, e.g. Windows) to run on the E2k chip without infringing on existing
intellectual property. This approach, called binary compilation, will be used to build in
compatibility with other platforms, mainly with IA-64 (Intel's platform for Merced).
Elbrus technology does not infringe on any Western intellectual property and it is
protected by 70 US patent applications."
"binary compilation". For me, this means (roughly): They use the x86 code as "source code", feed an compiler and get their native E2k code. This way they can heavyly optimize it.
Very interresting IMHO.
I've never read anywhere how many CPUs are involved. Without this number, we can't tell anything about this.
After looking at the stats on dNet, i went to: www.chat.ru looks like an ISP of some sort (looking at the pictures) so i'm guessing it probably is a large LAN through an IPMASQ linux box and nothing to do with an E2K chip.
If the guy/guys behind "rc5whs@chat.ru" chose the "E2K power" comment line just to shake you americans up a bit, I'd say they pretty much succeded.. :)
Never seen so much debate come from so little provocation..
Do I smell the fear of american toes being stepped on?
--
da sajb
Goddam commie scum!
ROTFL!!!
Nice to see the good old Slashdot mentality at its best.
"xxx can't be better than our Linux boxes therefore xxx must be cheating".
Sorry guys, I got the scale off by a factor of 1,000. The Celeron cracks 1.3 Mkeys/sec., not 1.3 Kkey/sec, meaning that it takes roughly 360 clocks to crack a single rc5 key.
However, this is still a ludicrously high figure, one that could not possibly be attained by an efficiently pipelined EPIC implementation as of 1999.
~k.lee
Maybe the Russians are running Linux on that thing?!
The equipment on the Mir is over ten years old. It was equipment failure rather than computer failure that plagued Mir. The only time the computers really failed was when an overly fatigued cosmonaut accidently pulled the plug. I'm glad the Russians have that thing up there. It's excellent practice for what's to come.
Yes. This is the russian text. It mean's
"Guys, what are you doing here? - A Movie is over yet..."
What they did was to crack the RC5 protocol and they are automatically generating bogus blocks that have been checked. It looks like it might queer the whole contest.
that russian article is very very interesting. here are some interesting facts:
9 /articles/art_2who.htm#pentkovski
1. I translated a paragraph related to Transmeta (firm for which Linus Torvalds work):
During the period of 1992 till 1995 Elbrus was working together with a Sun microprocessor architector Dave Ditsel. As Babayan said "After that Dave started his own firm - Transmeta and started working on a computer very similar to ours. We still have close contacts with Dave. And he still wants to work with us." Very little is known about the Transmeta product. The only thing that is own is that it is going to be VLIW/EPIC
microprocessor with low energy consumption, binary compatibility with x86 is achieved using dynamic object code translation.
2. One of the lead PIII developers was one of the russian elbrus developers:
http://developer.intel.com/technology/itj/q2199
Jez... I've got a better chance of finding E.T. than hitting the winning code.
Patriots don't work very well. There was lots of hype about them in the Gulf War, but they were still only shooting down a few SCUDs. They are also fairly short-range, and need quite a bit of prepping. They are no defence against nukes at all.
but then again, you have to admitt that transmeta is in a place where investors are looking for projects while E2K (if it's for real, I don't know) is on the other side of the earth however you look at it.
On a general note, having met a number of rusian scientists I dare to say that rusian science deserves A LOT of respect. They have great tradition in science and education. Don't put them down because the political system during the post war era has destroyed the country. There is nothing wrong with the people.
Cheers,
/jarek
I would be very much inclined to think this is a fraud and here is why. First of all, I believe that the E2k used its own intruction set, there is not E2k client.
Secondly when we last heard they hadn't even put the chip in silicon yet. I'm sorry, but even for large companies it takes several months to go from the final design to silicon.
It just doesn't seem feasible. Here is another thing, lets do some math. Slashdot did 1.55 billion keys yesterday. That is done with 912 people. Assuming only 1500 computers doing RC5 on slashdot that would be about 1000kkeys a second for each computer (seems high, my k6 only gets 330 or something). That would put this thing approximately 2700 times faster than the current cpus. I'm sorry, thats not true.
My Slashdot account is old enough to drink...
From looking at the actual work being performed, there is no reason to believe that this person is using a hacked client. We've been in contact with the user and although the details are sketchy his story seems legitimate.
We're still talking with him, but for now the assumption is that this activity is legitimate. Our biggest fear is not that he's compromised the project but that he's using resources he doesn't have permission to use.
For what it's worth, he's not claiming that the blocks are being completed by an E2K. The motto is just his way of showing enthusiasm for the platform. In reality, it's win32 (and a few sparcs) all going through a linux proxy.
It's still way too early to draw any conclusions, but so far there's been nothing to set off any alarms on our end.
No, the Pentium II/IIIs are RISC based at their core with a CICS interpreter above them for the system to interact with.
Here's more info on the E2k:
t ml
:P
http://www.el2000.ru/press-releases/25031999-01.h
At least it won't be those AnandTech weenies who overtake us first!
Occasionally send out a key which is known, but unknown to the client. See if the client reports it as positive.
The fact that it's being used to crack RC5.
---
seumas.com
Distributed.net don't have one for download...
;-)
I guess they had to hack together their own
My understanding is that there are people who swear *by* (not at) their Lada Samara's.
:-)
Because it's one of the few cars out there that can be repaired in the field, with simple hand tools. No fancy electronic stuff, not sophisticated feedback systems, no obscure bits and pieces.
Which, when you're in the middle of the Australian desert, can make the difference between living and dying. A bit of duct tape, a wrench and a whack with a hammer, and you're off and running again.
Wouldn't want one, myself. Got garages and taxis and stuff 'round here.
--
Don't like it? Respond with words, not karma.
> You really think that their first move would be to waste processor time on a silly game.
Yes, that's quite likely. Consider, for example, how many people own machines with horsepower and RAM leaking out of the seams, and dual interleaved 3D video cards just to play Quake and such like.
BTW, he doesn't say "I am running an Elbrus/Elbrii," just "Elbrus is really cool." And I agree.
They get their speed by ignoring the derived-from-4004-25-years-ago Intel architecture to build the basic CPU core, then building a good on-the-fly Intel-to-reality translator for it. Which is essentially what everything since the PentiumPro has done.
AMD seem to finally be getting a process like this right with the K7, shpiing in weeks, whereas Intel might get around to shipping Merced before RC5-64 finishes. Perhaps instead of just selling out, Cyrix should have started making Elbrus?
Got time? Spend some of it coding or testing
>>Isn't it amazing that Trasmeta, without any
>>thing, has most people here believers. While
>>E2K, on the same boat, with evidence, has most
>>people here yell fraud?
Yeah, but here's the difference. Transmeta has been basically silent about their project(s)--what you hear about them is speculation from outsiders, whereas Elbrus is it's own PR machine. And, at the risk of sounding cynical, they do have a vested interest in everybody beliving their claims, true or not.
Elbrus is the one making the outrageous claims here, let them prove that what they say is true. And in the absence of such proof, why should I believe them?
Third-party benchmarks anyone?
ABSURDITY, n.: A statement or belief manifestly inconsistent with one's own opinion.
Eff's machine that was used for previous competitions? Is it too expensive to pay the power bill for or something?
Or is this something completely different?
a notice doesn't even need to be given out - because the client doesn't return jack squat to the user about the key that was just completed. that would be a load on the keyservers that might not work well though... good idea though
--onyx--
Patriot is a joke.
No no no, Dave Taylor, founder of crack.com.
HE used to word at id software. He's the one that did all the ports of id games to linux (until he quit and Zoid took over)
The thing about hacking up a fake client is that it can report "key not found in block X" really rapidly by not actually doing the search. This is an excellent heuristic which is correct in very close to 100% of all cases, with the minor drawback that it completely misses the point. If someone measures the size of a "not found" report message, we can use that to perform a meaningful calculation on their computer system: not the number of keys per second, but the throughput of the proxy server as measured from their client. Still, the hacked client thing is a bit of a problem. How do you tell whether the client at the other end of a net connection is legit? Too hard for me, Jimmy.
proof, n. A demonstration that a conclusion is implied by certain premises and axioms.
From one of their recent press-release Q&A's:
http://www.el2000.ru/press/press_faq-2502.html
Q: Have you already worked with OSs supporting multiprocessing?
A: We have compiled Kernel OS Linux 2.0.34 using Elbrus compiler and executed on the machine simulator.
And yes, I know, Elebrus runs x86 so we could figure that out ourselves, but it's nice to hear it from them!
What makes the E2k just soooooooooo much better then any other western design..
It's more a question of known quantities. People like Linus Torvalds? Dave Taylor? They've done amazing things in the past; most people are inclined to believe they may deliver amazing things in the future.
Anyone here who has more information than whats on their website?
Shri
transmeta may just have sneakier pr team ,-))
Here are translations of some of the things mentioned about E2K (i put [sic] wherever i did not know what the hell they were talking about):
-- the processor should triple or quintiple the performance, cut down the electrical consumption, and be cheaper than the Intel Merced.
-- the Elbrus team has a good enough reputation and experience to make that happen [sic]
-- Elbrus series computers were produced in Russia well before their architectural analogs even made their way into western development labs.
-- the Elbrus 3 was made in 1991 using old crystall [sic: not silicon?] technology but still outperformed the Cray 2-to-1.
-- the E2K will use even better technology than that of the today's record holder Alpha 21264.
-- the E2K EPIC technology with its low consumption of electricity will provide in the next 2-3 years a "supercomputer in a pocket calculator"
-- the compiler for the E2K is as innovative as the hardware used. The de-parallelizing [sic] compiler at its current state allows upto 10 operations per cycle which is more than 3 times higher than the Alpha.
-- the E2K is able to run Intel and Sun processor code only 10-30% slower than its own (in comparison to the FX!32 patch for execution of Intel code on the Alpha slows the computer down threefold). With this the E2K allows for a 100% dual compatibility [sic] of any Intel codes under any operating system, which is another improvement over FX!32.
-- another important innovation is the "bulletproof" defense of active code and data against viruses. Development of similar technology in the West stopped with the downfall of Intel 432 processor.
The logical development of the Elbrus has been completed and the team is now ready for the final phase -- making the crystal. [silicon or something else?]
According to B.A Babaian, "The first superscalar machine was Elbrus 1 in 1978, whose analog in the West appeared only in 1992. In fact, that design is analogous to the Pentium Pro which appeared in 1995."
According to Kit Difendorf, a Motorola developer, "in 1978, in Elbrus 1 the processor worked with execution of 2 operations per cycle, changing of operation order, renaming of registers, and execution of request" [sic].
Again according to Babaian, a number of western companies have been talking to Elbrus, including Sun and HP, but only Sun was able to become a partner. The Elbrus team has worked with Sun towards improvements in the UltraSparc processor, compilers, operating systems (including Solaris), Java, and multimedia libraries. [Amazing, huh?]
However, the partnership with Sun was discontinued because all of the intellectual property would move over to them even though 90% of the technology was done before Sun even appeared. Right now, around 70 US patents protect Elbrus's intellectual property.
...
All system programming for the E-1 (1978) and E-2 (1984) was done in hig-level language El-76 instead of assembly. El-76 has a lot of features of Algol-68, but the principal difference is in dynamic connection of types on hardware level. All programming on Elbrus is in El-76, there is no assembly for the Elbrus. El-76 is translated into bytecode, reminding of Java, which is interpreted by the hardware into simple machine commands "on-the-fly."
A lot of the rest (which i am skipping) is about E-90 which looks almost exactly like the Pentium Pro spec sheet. And the next bit about E2K includes comparisons of it with Intel Merced.
E2K Merced
Frequency, GHz: 1.2 0.8
Performance, Specint95/Specfp95: 135/350 45/70
Size in mm^2: 126 300
Electric consumption, Watts: 35 60
System bus throughoutput, Gb/sec: 15 n/a
Cache, Kb: 64/256 n/a
Peak performance, GFLOPS: 10.2 n/a
I was thinking of how to intentionally fail my drug test... It would make a good memoir story someday.
I certainly hope this thing is real, the RC5 project needs as much CPU power as it can get. A lot of people switched their computers over to the SETI project and I think it's impacting RC5 a lot, which is a shame. At yesterday's block rate the last RC5 block will be cracked on January 4, 2007! But of course, the secret key could be found at any time....
For them to say that the E2K will be 3-5 times faster than the Merced is silly, considering:
The Merced is not out yet
There are no benchmarks for the Merced yet
The E2K has only been SIMULATED with Verilog
I don't care just how explicitly parallel the E2K is, you are not going to get that kind of performance gain.
For those who are curious, a bit about how explicit parallellism works:
Every intel cpu since the 386 has had a 32 bit instruction word. A portion of this is the actual instruction, and the rest is the data that it operates on (which register, memory address, etc.)
With Merced's EPIC (explicitly parallel instruction computing) ISA, the instruction word is _128_ bits wide. In this the 128 bits are actually *3* instructions plus a template that contains extra information. This is how it is explicitly parallel.
You may have heard how EPIC requires a lot of code analysis during compilation. This is so that the compiler can find portions of code that do not depend on other parts. These are separated and executed in parallel by the Merced.
There are a lot of other features that I won't bother explaining (predication, speculation, etc.). Just be assured that the EPIC architecture is VERY fast. For the Elbrus guys to claim this kind of performance gain over a cpu that is still not out (using benchmarks from a SIMULATOR, mind you), is ludicrous.
Then there's the issue of the insanely high keyrate.
Take one look at those numbers again. 2.7 GIGA keys/second? The fastest cpus today still operate in the area of millions of keys per second. I'm curious how an unknown company using unknown technology on an unknown cpu could get a keyrate an order of magnitude larger than anything else. That alone should make anyone a little skeptical.
Can anybody make anything of this? "Ðåáÿòà, à ÷î ýòî âû òóò äåëàåòå? - Êèíî-òî óæå çàêîí÷èëîñü..." My guess is that it would look more like human writing if I switched to the cyrillic font. Go here for the original.
That's what the 'explicit parallel' architecture they're referring to is all about: completing several operations at the same time or during the same clock cycle. That was all that older computers and OS's could do (Original DOS for instance). Back then, a megahertz rating on a computer meant everything simply because of the fact that the computers were explicitly serial (only able to do one operation at a time).
Your confusing *application* parallelism, with CPU parallelism. CPU parallelism is what allows a CPU to execute more than one instruction per clock, but as far as the programmer, the OS, and everyone else is concerned, it's still a serial machine. Consider this x86 example
add bx, ax
add cx, ax
inc dx
On a 286, or something each one of those instructions would go sequentially, some of them taking more then one CPU Cycle. But, if you look at the code, you see that none of those operations require results from any of the others. So it would be possible to do them "out of order" or even at the same time. The CPU is still a serial device, but CPU parallelism is just a way to cram more code in there faster.
On the other hand, *application* parallelism has nothing to do with the CPU your using. You can linux on a 386 just fine. What happens there is, the Operating system "time slices" every once in a while, and changes to another application. The CPU only executes code from one app at a time, weather or not the CPU has any parallelism. The big advantage for application parallelism is that you can run *more than one* CPU at a time, and have them operate on more then one stream at a time.
SMID implementations like Intel's SSE and MMX and AMDs 3dnow instructions are a further step in CPU parallelism, but they don't really let you run more then one thread at a time
_
"Subtle mind control? Why do all these HTML buttons say 'Submit' ?"
ReadThe ReflectionEngine, a cyberpunk style n
From the press release:
The E2K project represents the latest commercial endeavor of the former Soviet Union's most talented computer scientists, many of whom have designed and delivered three generations of supercomputers, including those running the Russian Space Mission Control and the Russian Missile Defense System.
Ummm... so these are the same 'computer scientists' that have us worried about Y2K problems with their missile defense system? Yet they can design a killer chip like the E2K?
There's a super duper huge error on the page with
the press release...
They state that current Pentium processors are based on a RISC architecture?!?!?
Isn't that more like CISC?
"Code free or die!"
I've got 6 computers here - all Intel, all overclocked. My keyrate is generally around 9.5-10million keys per second. Two of these boes are dual SMP Celerons - one running 2X464, the other 2X450 or so (Damn 366s!). My lowliest Celeron - a 266 Celeron runs 448mhz and cranks over 1million keys itself. Despite lacking a cache it flies on RC5 and DES. The dual boxes crank out over 2million keys per second each BTW.
:-)
Now - do I believe that this is a real team and number? Well - I'm very skeptical. 2Giga keys - ponder that a moment and compare it to what I'm running. 10 machines here crank say 10million keys - this guy is MANY times that. What does this company do? How many people does it have? How many workstations? Do all of these keys come through one Proxy? I figure he'd have to be running something like 2000 machines in the 450+mhz range to be getting this (unless my math is off and it's more). Mind you, my machines keyrate drops like a rock when I run something intensive so he must have more than what I calculated or be rolling in cash to let that many machines be dedicated to this task. Nugget said that this was a couple of SPARCs and the rest WIN32 machines so this isn't a CPU breakthrough here. That leaves one serious bunch of CPUs.
I dunno' - I guess if I had buy-in from the agency I work for and loaded RC5 on EVERY machine AND managed to coax the users NOT to turn off their machines at night I'd be able to have 2K+ machines doing this but what big company is going to allow this? In 3 days they went from zero to 2giga. Did that not ramp up over days as software was installed? Perhaps it was remotely installed but still that'd have to be one seriously organized company and administrative staff.
If this is real, and I'm skeptical, my hat's off to them. Sure wish I could get my employer to do this but we've got brickwalls instead of firewalls
Build it, Drive it, Improve it! Hybridz.org
I'm not running 10machines here so for my calcs I reduced the keyrate to about 1million per box. I also did some other figuring, if my current employer would do this and we've got 20K+ employees - all with machines - and assuming they all had decent CPUs (*cough*) then we'd blow these guys into the weeds.
Of course, getting the buy-in would take forever and would require leadership on the part of the managers. It would also require us to have recapped our computer hardware and be up to snuff on every box. Yup, the hope for us ever doing this is pretty dim huh? (smile) Maybe they've got 6 or 7K bokes laying around to get 2gigakeys. If the rate fluctuates up and down a good bit this could be more believable. For now I'm still a bit skeptical but if the d.net admins have confirmed this is correct it's a much needed boost in our overall keyrate and I'm happy to have them aboard. It's been a year plus and I'd really like to move onto something more fun to crunch without abandoning this project like so many have for SETI.
Build it, Drive it, Improve it! Hybridz.org
Now why dont you sell this computer to the U.S. so you can feed your people ;)
It's just 8k computers using the same email.
Isn't it amazing that Trasmeta, without any thing, has most people here believers.
While E2K, on the same boat, with evidence, has most people here yell fraud?
What an idiotic thing to say. First of all, these E2K rc5 stats are not evidence. Team Slashdot could change its name tomorrow to Team McKinley and claim to be cracking all its keys on a single McKinley chip, but only an idiot would believe us, for many reasons, chief among them being that McKinley does not exist in silicon.
Elbrus's website has no press release endorsing these people, and you can bet a company as loudmouthed as Elbrus would be shouting from the rooftops if they had a working piece of silicon that cracked keys at that rate.
Second, as someone already noted, Transmeta has made no real claims about processor performance. Even if they were to make such claims, I would not believe them until I saw a machine that I could boot up and use for real applications. Most sensible people here (the silent majority) would not believe them either.
~k.lee
(remove nospam for email)
Superscalar x86 architectures have been around since the first Pentium, which could also theoretically do 3 instructions per clock cycle (in practice it almost never does). But superscalar does not mean that cracking an rc5 key, let alone several, in a single clock is plausible.
Let's do the math. Given that overclocked Celeron-464 will give you about 1.3 Kkeys/sec:
(4.64*10^8) / (1.3*10^3) =~ 3.57 * 10^5 clocks/key
Given the small working set of rc5, memory latency and cache effects are probably negligible, so I believe this figure is CPU-limited.
People who are suggesting that the E2K cracks one key per clock cycle are being utterly ridiculous. A "crack rc5" instruction would have to carry out the equivalent of 360,000 P2 instructions. A VLIW/EPIC architecture, like a RISC architecture, depends upon fairly elementary individual instructions, even during vector processing (floating point ops are an exception, but are not relevant since AFAIK rc5 uses integer instructions exclusively). A "crack rc5" instruction would royally fubar the pipeline.
BTW the above figure is attained using the MMX/bitslice rc5 core, which is explicitly optimized for the P2 architecture's MMX instructions. True, the E2K has a different architecture than a Celeron. However, if E2k were cracking keys, it would be using the x86 client. It is inconceivable that the E2K could execute the equivalent of hundreds of thousands of P2 instructions in a single clock.
Of course, there are other reasons to believe that this is a hoax, but the technical reasons alone are sufficient to dismiss it out of hand. That Russian team is getting a pretty good keyrate; too bad they had to dress it up in lies.
~k.lee
(remove nospam for email)
They will never be able to execute x86 code at that rate. Even if this E2K was this fast, it would only be this fast at executing its native machine code. The x86 option is just there for compatibility, it's impossible to optimize that much more than AMD and intel have done.
maybe it's the new playstation
>>>please remove "nospam" from email address
I'm surprised at the number of people speculating on the E2K processor from the evidence given. All we have is some guy processing a large number of keys putting "E2K is real power" in his tagline, and all of a sudden people are assuming he's running an E2K processor. I'm not saying it's a fraud, but maybe we're reading too much into this - we're all familiar with how non-English speakers can be easily misinterpreted. Hmmm...
heh.. the new rectal wart plug-in for back :) I can
orifice - a distributed.net client
just see it now..
Hmm, there should be a way to check if a client
is bogus, and there is:
They could just throw one block in where something should be found! (maybe even a few)
If there are still "nothing found" replies they know what's happening.
To avoid media hysteria they can even announce that they will do so in some cases.
R.
"By the way if anyone here is in advertising or marketing... kill yourself." -- Bill Hicks
Perhaps you meant Dave Ditzel, the ex-Sun microprocessor guru who is supposedly leading the Transmeta CPU design..
Here is an interesting link that points out Ditzel's relation to Elbrus..
Zigbee Central: A Zigbee weblog
my mother used to drive a 2108 model Samara... We had problems with the carburetor (didn't start very well) but other than that, the car could hold together on Moscow roads.. Which in itself is an achievement, since the highway system in Moscow lacks some serious funding... I'm talking about potholes that could easily ruin a Mercedes in a few months... :)
The gas mileage on them is pretty impressive, as well, even though Lada engines don't get much bigger than 2 liters..
--- sig moved for great justice.
Isn't it amazing that Trasmeta, without any thing, has most people here believers.
While E2K, on the same boat, with evidence, has most people here yell fraud?
The irony of slashdot...
It turned out that the incredibly high key-rates were generated by a bogus client. Read what happened in Dave McNett's .plan or read how the mystery was revealed by Maxxim Kochegarov from Russian Team.