French RC5 HW decoder ?

[LeBlatt]

There's a thread at http://ubb.francexpress.com/ubb/Forum20/HTML/000987.html about building a rc5 decoder PCI card with 4 200MHz DSPs. It should crack at 180 Mk/s.

Interesting, but I think it's a bluff, because the guy says they have modified the dnet client and created an optimized core for that DSP. How can one do that without dnet's sources.

Moreover, this guy's stats doesn't reflect the thousands of blocks he says he cracked for betatesting... http://stats.distributed.net/rc5-64/phistory.php3?id=184729

I'll keep an eye on this

 

[ViRGE]

It's definately a bluff. You can't get 180MK/sec out of 800mhz worth of DSPs, even if they were custom built.

 

[JonB]

I tried reading that thread, but I think they got the font messed up. Its like its in a different language or something.

 

[dkappos]

I had no problem reading the page with IE5 and the encoding set to Western European. I suppose that something like this can be done (it was done for "Deep Crack" http://www.eff.org/descracker.html) but there are not a lot of people who can do it.
There is no client for this, this means that they if they have done it they are using a hacked client (?).
"- le client fonctionne avec de nombreux bugs (il est en béta aussi)"
They also have to proove that the results (if they have results) are reliable.

 

[rc5]

They haven't hooked that HW cracker up yet.


<< toTOW
Member posted 28-06-2000 22:59
--------------------------------------------------------------------------------
Veiled the news:
- the card fontionne without problems
- the pilot functions under Win 9x (still in beta)
- the customer functions with many bugs (it is in beta also)
Veiled one to copy to stick of a log that my father sent to me (C the very first one on this system!) (my comment are in capital letters):

dnetc v2.8009-xxxB1-CTR-xxxxxxxx for Win32 (Windows 4.10). (X REPRESENTENT THE NUMBERS WHICH DISTRIBUTED.net WOULD PUT; B1 INDICATES THAT IT A VERSION BETA)
Using email address (distributed.net ID) ' Test@off.line' (the TESTS HAVE ETE CARRIED OUT OFF-line, the ADDRESS MALL EAST THUS CAN)

[ Jun 28 14:03:19 UTC ] Automatic processor detection found 1 NCV card with 4 processors. (THE CARD EAST CORRECTLY RECOGNIZED )
[ Jun 28 14:03:19 UTC ] Automatic processor standard detection found Unknown processors. (THERE IT THERE A STILL OF WORK, THE VERSION OF THE DSP IS NOT RECOGNIZED YET... )
[ Jun 28 14:03:19 UTC ] RC5: using core #7 (Thomson' S DSP optimized). (CA GOES )
[ Jun 28 14:03:19 UTC ] Loaded random RC5 1*2^28 packet B274563D:e0000000
[ Jun 09 14:03:19 UTC ] Loaded random RC5 1*2^28 packet ACBA6888:50000000
[ Jun 09 14:03:19 UTC ] Loaded random RC5 1*2^28 packet ACBC0068 0000000
[ Jun 09 14:03:19 UTC ] Loaded random RC5 1*2^28 packet AD0A6F56:60000000
[ Jun 28 14:03:19 UTC ] 0 RC5 packets (0 work units) remain in buff-in.rc5
[ Jun 28 14:03:19 UTC ] 0 RC5 packet (0 work units) is in buff-out.rc5
[ Jun 28 14:03:20 UTC ] 4 cruncher (has - d) cuts been started. (AS ONE CAN SEE IT HERE CUSTOMER CONSIDERE THE PRESENCE OF 4 PROCESSORS)

The card of test will function during 24h to see whether all functions, the total power and the number of packages carried out in 24h.

toTOW

P.s: the card for SETI does not have interest, it does not exist (at Thomson in any case) of DSP planned for this kind of calculations.

P.s2: as soon as the card and the software unit function it att&eacute;rit in my p3 for test of long duration and intensives


>>

 

[amok]

Deep Crack uses 1800 custom chips to crack DES at around 92GK/s, or at around 51MKeys/s/chip, and I believe each of those chips ran at around 150MHz. However, that is DES, and RC5 is a much different beast, and demands more out of hardware. If someone has developed these DSP's and this PCI board, then I know a lot of TA members who will be running out to line up their crack rack's spare PCI slots . Hehe, most crack racks consist of only a network card, therefore you could theoretically put 4 or 5 of those cards into one machine. Imagine if you will a single box that can crack RC5 at 720MKeys/s! Maybe somebody should just drop this guy an e-mail and find out?

 

[hutch]

On a serious note, if you were to use what I call &quot;real&quot; DSP processors, FPGAs, they have already built the cores to do both DES and RC5 decryption. There is no software with FPGAs, just gates. I have built up plenty of DSP boards based on FPGAs that would shame a Pentium III 1GHz on math functions with only a $40 processor.

Anyway, with 4 2 million gate FPGAs, I believe you could hit 180Mkeys/sec, without doing any serious computations on the topic.

 

[amok]

Hutch, could you build such a PCI card? Even if you don't have the capability to actually get these things produced I'm sure that we can gather enough to get some foundry to run 1,000 of them if you can put together the design . And considering the huge increase in output, and the fact that we have d.net personnel hanging around the board I'm sure that we could get a port of the client . Just think about it! If we had 1,000 of these and they could each do 180MK/s that would be a total output of 180GK/s added to the team, which is beyond the current keyrate of the entire project! We could exhaust the entire keyspace in a mere 1.5 years!

 

[rc5]

Yeah. But what after RC5 is done? All cards will be trashed.

 

[amok]

So what? Of all the stats junkies here, do you think anyone would mind paying ~$180/card? That's $1/MK/s. Even if RC5 only lasts another year they would probably consider that a bargain .

 

[SpaceWalker]

I was wondering? Aren't DSP's Programmable?...They just perform what ever task they are told to do....So they could be used for other task such as SETI or OGR...


MOOOOOOOOOOooooooooooooo...............

 

[Ken g6]

And besides, RC5 will go on to a 72-bit key, and eventually even a 128-bit key. I don't know how we'll crack those without some kind of special hardware.

And hey, if all else fails, we could ask Harvey Dubner how to make them do GIMPS! (For those who don't know, Harvey Dubner made a specialized board called the Dubner Cruncher to help find primes.)

 

[rc5]

I don't know how programmable a hardware crackers is. But I know EFF DES cracker can't do anything except cracking DES. Also, presumbly, the ability of reprogramming MUST hurt the performance terribly, let alone the extra cost on designing and manufacturing a &quot;programmable&quot; hardware cracker.

Anyway, if someone can keep the price below $100, I might buy myself one.