A Few Questions:
#1
I have never spent much time with large scale decryption. Indeed, I have always deferred general cryptographic work to my colleagues and they likewise forward me penetration, topography, and forensics work. Though I am well versed in concepts, I have very little hands on when it comes to cracking with GPU and coprocessors.

I am putting together a system to run various password recovery jobs and have a few questions that I could use some help with. I thank you in advance for your thoughts and suggestions.

My build will consist of a four socket, 4U server utilizing Magny-Cours (I dislike the Bulldozer architecture and I lose sleep over Intel's vPro being used as a backdoor. Anyone else share that paranoid delusion?) I am aware that x86-64 CPUs have little effect on hash speed but I will be using the system for more than password recovery. Anyway, the important point is the GPU. I know AMD has many advantages but are there certain encryption/algorithms that work only on Nvidia architecture (either from actual architecture or from certain software only being written to handle Nvidias architecture for whatever reason)? If there are rare cases in which an Nvidia platform would be advantageous I would be very pleased to know as I would like to develop a rig that can handle as many algorithms as possible. Is there a specific hash that fares better with something like a Tesla? (whether because of software support or architecture)

Regarding AMD, what is the absolute best card to use at present? Is there anything coming up within 3-6 months that might be worth waiting for? I know that a 7970 ranks highly but I remember reading something about an older card being better because it had more cores. Another question that I can't find any answer for is regarding FirePro cards. Would a card like the S9000 or S7000 be better than the 7970? I can't find a core count. I know that extra GPU memory is largely unneeded but do the FirePro cards also have superior processors or functions that benefit OpenCL? In short, is there a FirePro card that is better than the 7970 for our purposes?

I am a UNIX man. I am most at home with AIX (and other UNIX offerings from other vendors) and BSDs but Linux naturally plays a huge part. My question: is there a time that Windows (please say no) ever surpasses Linux in a specific hash type (be it from architecture or simply because developers focused on Windows when writing the decryption software)?

Something that I don't think has been mentioned in the forums is Tableau's TACC1441. I am guessing because there is no support under hashcat for the device but I am nevertheless interested in the option (be it using hashcat or other password recovery software). The graphs show that it can be up to 10-30 times faster than a quad core Intel processor but I am very interested in how the unit fares today (the system is over two years old and was compared against Core 2 architecture) against today's Xeons but more particularly against GPUs. You can see the results at the following: http://www.lostpassword.com/tacc-acceleration.htm and http://jchblue.blogspot.com/2009/09/tabl...rator.html . So, who fares better? The TACC1441 or something like the AMD 7970?


I realize that many of my questions are far reaching beyond being specific to hashcat but I have not been able to find the answers to the questions and there is no forum that I know of that would be better suited than this one. I thank you very much for your time.

-Host
#2
Welcome to cracking world. There is no hash know on which Nvidia is faster than AMD, however there are hashes on which the distance on their performance ratio is smaller than compared to others. For example SHA512, since NVidia has better support for native 64 bit instructions than AMD. Also DES, since the memory access works faster. However, in total, they still are behind AMDs performance. The Teslas however are totally worthless for cracking since they are focuses on floating point calculations while in crypto we only do integer calculations.

The best AMD card is the hd7970. When the hd7990 will come out, this one might be better. You should totally not buy any other card if your are using it for professional stuff. Take a look at:

this: http://www.gat3way.eu/est
and this: http://www.golubev.com/gpuest.htm

What the linux and windows thing is about there is no difference in performance. They are both at the same speed, however, there are better overclocking utilities on windows than on linux. If you are not going to overclock it doesnt matter.

Your last question was about the TACC1441, which is a FPGA based system. Yes, FPGAs are cool, especially when it comes to electricity costs, but they are damn unflexible. Advanced attacks like rule based attacks or markov-chains for hybrid-attacks wont work with them. If you are going to crack real-life passwords you will totally use these attack-modes.

HTH
#3
atom,

Firstly, thank you very much indeed for your reply.

I looked over the charts that you linked to and it appears as though the 6990 reigns supreme (the 5970 coming in second, the 7970, third). Why do you suggest the 7970? I know the 7970 has the newest architecture but if the 6990 is better suited to cracking I would naturally prefer that architecture regardless of what benefit to gaming the 7970 provides. As for the 7990, when do you expect that to be arriving? Shall it have more cores, memory?

Also, you didn't mention anything about the FirePro. Is it generally not supported by cracking software? Is there nothing beneficial to cracking in the FirePro architecture?

Is there any time frame on when AMD's next architecture will be released ("8000 series")?

Thank you for clearing up my questions on NVIDIA's chips. Something still hangs in my mind about there being a single complex hash type that was only able to run on the TESLA (for what reason, I have no clue). Oh well.

My sincerest thanks for the help. I'll soon be making an investment and this information really helps.
#4
if you are limited with power and space dual gpu cards like the 6990 would be ideal as you can fit them in a full size ATX tower and still have 8 GPUs to play with. however, with the proper server chassis, 8x6970 (which is what the 6990 is based on) will be faster than 4x6990, and 8x7970 is faster than that.

there are a couple members here who have 8x7970 in a single server.

https://hashcat.net/forum/thread-1515.html
http://www.tmto.org/docs/HardwareComparisonGTXvsHD.pdf
http://ob-security.info/

note: in a single system, AMD drivers are limited to a max of 8 GPUs where Nvidia allows max of 16 GPUs

also, powercolor just announced their new dual gpu card based on the 7970 and should be released soon.
http://www.powercolor.com/global/product...asp?id=423
#5
forumhero,

Thank you for the reply.

I'm not intimately familiar with GPU architecture. Will the official 7990's from AMD be integrated so much so that a user is able to connect four in series or will it be such that only two are able to be "crossfired"?

Also, I see that the AMD system you linked above incorporates eight cards. Where does "crossfire" come into play, if at all? Is there ever a time where Crossfire is used in cracking? Also, is one able to cluster multiple systems together (via FoE or any such protocol) to utilize greater processing capabilities? Again, I am very new to the cracking field. If not hashcat, then what software allows this? Perhaps something can be contrived outside of the cracking software entirely - for instance, at a lower lever in the stack. That's not my department.

Atom - I just realized that you are the writer of "hashcat". Thanks for the fine work, mate!

Thanks,

Host
#6
oclHashcat-plus will support up to 128 GPU with next release. you can cluster them using VCL, but that works only on Linux. crossfire should stay disabled when it comes to cracking. Both oclHashcat-* programs have specially designed workload distributors that no generic solution can do better. I think it was blandyuk who made some comparisons on crossfire here on this forum.
#7
Great thread, and very helpful responses. Thanks, Atom. Once question for myself. I've got a 6990, running on Ubuntu 12.04 with whatever proprietary "restricted" drivers that Ubuntu told me to install. I previously had Windows 7 on there. My Ubuntu machine is getting about 1/10 of the cracking speed that I previously had with Windows 7. Any ideas? Do I need to do something special to install the drivers on Linux?
#8
try using AMD drivers directly. i've has problems using native ubuntu drivers in the past
#9
Yeah this sounds like the driver is not correctly installed. Make sure you install catalyst 12.8 and also make sure when you run clinfo it should use the following versions:

Driver version: CAL 1.4.1741
Version: OpenCL 1.2 AMD-APP (938.2)
#10
So I reformatted the machine, and installed native AMD drivers. It still is far slower than it should be. clinfo shows the following (I can print the entire thing if you want):

Driver version: CAL 1.4.1741
Version: OpenCL 1.2 AMD-APP (938.2)

For example, I do the following test, and you'll see the speeds aren't what they should be (roughly 1/5 of where they should be):


./oclHashcat-plus64.bin --attack-mode 3 --runtime 30 --force --custom-charset1 ?l?d?s?u --hash-type 500 '$1$aaaaaaaa$aaaaaaaaaaaaaaaaaaaaa1' ?1?1?1?1?1?1?1?1
oclHashcat-plus v0.08 by atom starting...

Hashes: 1
Unique salts: 1
Unique digests: 1
Bitmaps: 8 bits, 256 entries, 0x000000ff mask, 1024 bytes
GPU-Loops: 64
GPU-Accel: 16
Password lengths range: 1 - 15
Platform: AMD compatible platform found
Watchdog: Temperature limit set to 90c
Device #1: Cayman, 2048MB, 0Mhz, 24MCU
Device #2: Cayman, 2048MB, 0Mhz, 24MCU
Device #1: Allocating 57MB host-memory
Device #1: Kernel ./kernels/4098/m0500.Cayman.64.kernel (2484660 bytes)
Device #2: Allocating 57MB host-memory
Device #2: Kernel ./kernels/4098/m0500.Cayman.64.kernel (2484660 bytes)

[s]tatus [p]ause [r]esume [q]uit =>
NOTE: Runtime limit reached, aborting...


Status.......: Aborted
Input.Mode...: Mask (?1?1?1?1?1?1?1?1)
Hash.Target..: $1$aaaaaaaa$aaaaaaaaaaaaaaaaaaaaa1
Hash.Type....: md5crypt, MD5(Unix), FreeBSD MD5, Cisco-IOS MD5
Time.Running.: 33 secs
Time.Left....: > 10 Years
Time.Util....: 33899.8ms/950.1ms Real/CPU, 2.9% idle
Speed........: 1188.9k c/s Real, 47098 c/s GPU
Recovered....: 0/1 Digests, 0/1 Salts
Progress.....: 40304640/6634204312890625 (0.00%)
Rejected.....: 0/40304640 (0.00%)
HW.Monitor.#1: 0% GPU, 54c Temp
HW.Monitor.#2: 0% GPU, 54c Temp

Any ideas? I have a AMD 6990 card, running on Ubuntu 12.04 64 bit. I'd rather not go back to Windows, but if it's running my cracking faster, I might have to.