Graphics Card Selection for GPU Password Recovery

Importance of the GPU

Recovery of a stable password consists of consecutive checking of millions of possible versions of the password. One after the other, one after the other...

If this process is divided into parallel streams and several passwords are checked simultaneously, search time will be reduced by the number of streams.

An ideal task for graphics cards with their huge number of computer cores!

And although graphics card computation units are primitive by comparison with CPU cores, there are many of them, they work smart, and they are able to recover passwords. Therefore, in most cases, even a simple graphics card will be better than any top processor — it will be quicker and cheaper!

GPU NVIDIA/AMD rating

On Passcovery's site we rate the throughput of the NVIDIA/AMD graphics cards. We calculate and confirm (in our Passcovery Suite program) speed yardsticks for different families of graphics cards when working on files of different types.

We recommend using the data from tables only as a starting point in selecting a GPU for password recovery.

I say "only", because there are nuances...

Energy efficiency

High-quality programs operating with a GPU are capable of loading any graphics card to 100%, but this creates problems:

• the immense amount of heat released — good cooling is necessary (preferably supplemental)
• high noise level from fans — isolation of the system unit is desirable
• the increased power consumption by operating 24/7 (and with stable security this will be the case) — additional costs to pay the power bills are required

With respect to energy efficiency, NVIDIA graphics cards with Maxwell architecture were the successful solution. Even under maximum load they do not turn into a heated frying pan, and at the same time show excellent results (with reservations about ZIP archives with classic encryption).

Economic efficiency

From time to time we receive a question as to how much more powerful a system built on professional graphics cards such as NVIDIA Tesla is. Without needing to spend $10,000, we will answer that!

The cost of a graphics card is not in itself a guarantee that it is the most productive.

NVIDIA Tesla cards, for example, are not at all suitable for password recovery! The speed that they demonstrate is one of the lowest in the family. A similar story holds also with the AMD FirePro.

Password recovery consists of very specific calculations for which many parameters of the graphics card are not important. There is no sense overpaying for professional GPU and gamer bells and whistles of mass top-end graphics cards which will never be required for password recovery.

The optimal choice (with respect to price and throughput) would be graphics cards in the average range of the current family, or the top of the previous family. This pertains both to NVIDIA product (with the reservation regarding classic ZIP encryption), as well as AMD.

Of course, if peak speed of GPU password recovery is necessary, and the cost of graphics cards and energy efficiency are not important, the best choice is the most loaded graphics cards, with maximum number of cores and maximum frequency.

Classic ZIP encryption

Does this mean that we build a system using graphics cards from the average range of the previous line and gain optimum productivity on each dollar invested?

Everything is that way for almost all formats with which our Passcovery Suite works. Just not for ZIP archives with classic encryption!

Several most recent generations of NVIDIA graphics cards are the worst choice for such ZIP archives. This does not depend on the quality of password recovery programs. It is associated with the architectural features of these graphics cards.

Here is how the suitability rating of different families of GPU NVIDIA for ZIP password recovery with classic encryption looks:

Here is that reservation that I mentioned earlier — among NVIDIA graphics cards, the best selection for ZIP password recovery with classic encryption are still those of 2010-2011!

And the best selection (by speed, but not by energy efficiency!) for such archives is a GPU from AMD with GCN architecture — the Radeon HD 7970, with a speed on the order of 1.2 billion passwords a second (or the R9 290X which is 30-50% faster).

AMD graphics cards built on VLIW5 and VLIW4 architecture were not so successful.

One more important point which must be noted regarding password search for ZIP archives with classic encryption: GPU acceleration is possible only for a brute-force attack where the last character is not masked. Only in that case is password search possible on graphics cards. An attack using composite mask, and more so an attack using GPU dictionary, will not be accelerated.

CPU, storage, and disk requirements

Usually, GPU password recovery does not indicate any special requirements for other system resources.

But there are three exceptions:

• RAR archives with RAR3 format
• ZIP archives with classic encryption
• OpenOffice documents

In these cases, CPU productivity becomes important for large archive/document sizes.

This is connected to validation of the pool of passwords encoded on the graphics card. With a major imbalance of throughput, the CPU/GPU can turn out such that the powerful GPU will stand idle a significant portion of the time waiting, while the less powerful CPU will check the validity of the calculations. We wrote about this in more detail for RAR3 archives, but the same is true for classic ZIP and OpenOffice.

Results of GPU selection for password recovery

I reduced everything above to a summary table.

Here are all the formats that are supported by Passcovery Suite — our program for password recovery, with a recommendation of the optimum graphics card for this format and additional requirements for high-quality searching:

And as a result I will say that graphics cards do not guarantee success for the search, but they do considerably accelerate password recovery for many types of files and documents with permanent protection, and it is simply essential to use them when possible.

If you have questions-comments-additions, write us through our online helpdesk system.