The developer provides some plain numbers to show the effectiveness of using the GPU to recover a rar password with four characters:

* ~168 passwords per second on single core of Q6600 @ 2.4Ghz
* ~325 passwords per second on 8600 GT
* ~3120 passwords per second on ATI HD4850
* ~2075 passwords per second on GTX260/192SP

The performance of the listed ATI card is almost 20 times that of a password recovery where only the cpu is used. The password recovery software is a command line utility and the developer is offering extensive information on the possible parameters that can be used to recover the password. The suggested length of the password should not exceed six characters although it is theoretically possible to start a password recovery for a password with up to 17 chars.
http://www.golubev.com/rargpu.htm

GPU-Accelerated Wi-Fi Password Recovery

The latest generation of ElcomSoft’s proprietary GPU acceleration
technology supports, for the first time, both ATI and NVIDIA hardware. The
technology offloads parts of computational-heavy processing onto the fast and
highly scalable processors featured in the latest ATI and NVIDIA boards. When
one or more compatible ATI or NVIDIA graphics cards are present, the
patent-pending GPU acceleration technology kicks in automatically.

This time, ElcomSoft has used the technology to accelerate the recovery
of WPA/WPA2 PSK passwords, allowing network administrators to perform timed
attacks on their wireless networks in order to determine how secure exactly
their networks are. Considering WPA/WPA2 strong security with the minimum
password length of 8 characters, Elcomsoft Wireless Security Auditor employs
the highest-performance dictionary-attack with advanced mutations to allow
carrying out a password audit within a limited timeframe.
For more information visit: http://ewsa.elcomsoft.com

Breaking Wi-Fi Protection with Elcomsoft Distributed Password Recovery

With growing numbers of Wi-Fi networks used by businesses and individuals all over the world, security becomes utterly important. There are currently two methods of protecting Wi-Fi networks, WEP and WPA/WPA2. Unlike enterprise, RADIUS protected networks, consumer-grade WPA and WPA2 protection methods rely on passwords and encryption to protect traffic transferred between users and network access points. However, WEP, the older protection method, is no longer considered secure even for home users, as sometimes it can be broken in less than two minutes due to security flaws discovered in the algorithm.

The newer WPA/WPA2 encryption is inherently more secure than WEP. The only way to break WPA and WPA2 encryption is to use a brute force attack, which involves trying all possible passwords in the hope to discover the only correct one. With billions of possible combinations, it can take years to break into a WPA/WPA2 protected network. However, WPA/WPA2 protected networks are not immune against distributed attacks performed with GPU-accelerated algorithms.

With the latest version of Elcomsoft Distributed Password Recovery, it is now possible to crack WPA and WPA2 protection on Wi-Fi networks up to 100 times quicker with the use of massively parallel computational power of the newest NVIDIA chips. Elcomsoft Distributed Password Recovery only needs a few packets intercepted in order to perform the attack. The new product of ElcomSoft Co. Ltd. makes it possible to quickly perform security audit of corporate Wi-Fi networks, allowing to test network security against threats such as inappropriate WLAN security policy.

Using NVIDIA Cards to Break Wi-Fi Protection Faster

Today’s video cards such as NVIDIA GeForce GTX280 can process hundreds of billions fixed-point calculations per second. Add as much as 1 GB of onboard video memory and up to 240 processing units, multiply it by two by using a couple of NVIDIA cards, and enter the whole new world of super-parallel computational power for just a few hundred dollars.

Until recently, all the power of highly parallel, super-scalar processors in 3D graphic accelerators could only be used for gaming. ElcomSoft Co. Ltd. has invented a way to utilize the massively parallel computational power of NVIDIA gaming cards for increasing the speed of password recovery . Elcomsoft Distributed Password Recovery, its flagship password recovery tool, is able to fully utilize recent NVIDIA chips used in laptop, desktop and server computers, increasing the speed of Wi-Fi password recovery up to 100 times compared to conventional CPUs.

ElcomSoft has released a new version its Distributed Password Recovery program for recovering system and document passwords at speeds of up to 1 billion passwords per second. Among the passwords the software can recover are system passwords such as NTLM (Windows logon passwords) and startup passwords, MD5 hashes, password-protected documents created by Microsoft Office 97-2007, PDF files created by Adobe Acrobat, as well as PGP, UNIX, and Oracle.

What’s interesting about the ElcomSoft approach is that the company is using multiple GPU-based video cards such as NVIDIA’s GeForce GTX280 in parallel to process hundreds of billions fixed-point calculations per second. This means, says ElcomSoft, that this release of the Distributed Password Recovery program can try around 5,000 passwords per second for Office 2007 documents with a single GeForce GTX260, while regular Core2Duo processors can only try up to 200 passwords per second.

ElcomSoft claims that all users have to do is insert into a PC video cards (like the GeForce GTX280) to take advantage of the capabilities. Unlike NVIDIA SLI mode (Scan Line Interleaving) that enables transparent use of multiple GPUs, ElcomSoft uses the computational power of several NVIDIA cards no matter if they are of the same kind. Currently supporting all GeForce 8 and GeForce 9 boards, the acceleration technology offloads parts of computational-heavy processing onto the fast and highly scalable processors featured in the NVIDIA’s graphic accelerators.

The acceleration technology developed by ElcomSoft allows the execution of mathematically intensive password recovery code on the massively parallel computational elements found in NVIDIA graphic accelerators. The GPU acceleration is unique to Elcomsoft Distributed Password Recovery, making password recovery up to 50 times faster compared to password recovery methods that only use the computer’s main CPU.

• CUDA version only:nVidia GeForce 8xxx and up, at least 256mb of video memory.
• LATEST nVidia-driver with CUDA support.Standard drivers might be a bit older (as CUDA 2.0 is still beta)
• CPU with SSE2 support (P4, Core2Duo, Athlon64, Sempron64, Phenom).
• Recommended 64-bit OS (WinXP 64 or Vista64). 32-bit version is also available.

heres our test

MD5 bruteforce benchmark

Competitors comparison: CUDA

Competitors comparison: CPU

You may download competitors here: