md5 cracker benchmarks

Martyr

O/S: WinXP 64-bit
CPU: INTEL Q6600 2.4Ghz
GPU: XFX 8800GT 512MB
+++++++++++++++++++++++
SSE2 Code only

Per core: ~44 million k/s
Total: ~176 million k/s

---

CUDA 64-bit + SSE2

CUDA Total: ~365 million k/s
CUDA + SSE2 Total: ~543 million k/s

the interface is pretty crap, but its interesting benchmark for desktop PC.
most md5 implementations would calculate ~10 million k/s on same q6600.

http://3.14.by/en/read/md5_benchmark

Martyr

new version released yesterday:

http://3.14.by/en/md5

-? Prints this help
-h 21685d282d79098b89bdf5a916b66c90 Set hash to attack
-c 0aA~ Set charset. 0 - digits, a - small chars, A - capitals, ~ - special symbols
(special symbols are ~`'!@#$%;^&*()_+=- /?.,\|").
-C "abc23#" Add custom characters to charset.
-thread_n 128 | Might increase speed on GTX260 and later.
-grid_n 128 | Try 192, 256 for both
-gpu_time 1500 How long GPU kernel allowed to work, in milliseconds. 1500 - a bit faster, 50 - a bit slower,
but windows is more responsive. Default 500.

h57xiucj2z946q

Looks interesting, Im I right in saying that you can use your GPU to do work also in this?

Martyr

yes - i'm not sure if it works with an SLI setup, but any 8 series from Nvidia will work..although you don't see the impact with anything less than an 8800 model atm.

these cards are dropping in price and newer models are being released.

lots of algorithms can benefit from GPU's now since there are SDKs like CUDA for Nvidia and Stream for ATI.

Larrabee + AVX are around the corner too and security of passwords will get weaker as the hardware prices drop.

h57xiucj2z946q

This could have a huge impact to benchmarking? As GPU's specialise in math don't they?

Martyr

alot of benchmarks set for applications are gradually being reset since the release of CUDA.

for example TMPGEnc code was improved by almost 500% in some cases - encoding audio/video benefits alot..but alot is still under development.

ATI followed Nvidia with Stream

INTEL are planning to release Larrabee, its own GPU, and it just *might* destroy the competition in long run.(rumoured to support AVX - 256 bit vectors)

before CUDA was released, GPUs were only suitable for floating point calculations..now integer code has better support, so cryptography/audio/video and scientific applications are making use of this.

It would be great if Toshiba/IBM decided to release a multi-core Cell B.E on a PCI-E card. lol

I don't have the mathematic knowledge, unfortunately - but an interesting experiment would be to create a distributed DLP solver using something like Pollard-Rho algorithm. and run it on a cluster of GPUs

some people (even cryptographers) dismiss GPU's for solving cryptography problems..but if you think about the amount of computers in the world infected with malware and part of a botnet, owned or rented by criminals.

its more common to find a PC with SSE2 + GPU than an FPGA, so a botnet could be serious threat to e-commerce in the wrong hands. (as if it wasn't already)

who knows what kind of crypto could be cracked, but we won't really know the possibilities until some research is done - count me out.