Archive notice: This article was originally published on October 26, 2007. Links and embedded videos are preserved as part of the historical record.
The Russian company ElcomSoft has applied for a patent for a technology with which passwords can be cracked 20 times faster. ElcomSoft specializes in software intended to enable access to data even when passwords have been lost or forgotten. The Russian developers have now received support from graphics-card manufacturer Nvidia, which presented a developer tool at the beginning of the year with which the GPU can also be used to calculate work processes. ElcomSoft uses the tool in conjunction with the GeForce 8 card. The resulting hardware/software package allows decryption professionals to assemble computers, the software company said in a release.
Until recently, it was not possible to use the graphics-card processor for applications such as password recovery. Older graphics chips could perform only floating-point calculations, but fixed-point operations are required for most cryptographic algorithms. Current graphics chips are now able to perform these calculations. With 1.5 gigabytes of memory and up to 128 processor units, many of these chips are even better suited to such operations than conventional CPUs, according to ElcomSoft. GPUs function as multiprocessors, with multiple registers, shared memory and cache. The password software now accesses this performance. Since high-end motherboards can already work with up to four graphics cards simultaneously today, even faster password-recovery options will be available in future, ElcomSoft is convinced.
As an example of the performance of the patented technology, the company cites the recovery of an eight-character login password for Windows Vista consisting of lower- and upper-case letters. With the brute-force technique, 53 trillion passwords would have to be checked. A current dual-core computer would be able to check around ten million passwords per second. It would therefore take about two months to find the correct password. By including the GPU, the period can be reduced to three to five days, according to ElcomSoft.
The manufacturers of the graphics cards are of course also aware of the performance of their chips. Under the name Tesla, Nvidia offers a product family for the field of high-performance computing. The products are aimed at scientists and are intended to help desktop PCs achieve significantly greater performance. “Two of these GPUs installed in a standard computer turn the workstation into a supercomputer that delivers up to eight teraflops of computing power,” Nvidia spokesman Jens Neuschäfer told pressetext. The Tesla GPU consists of 128 parallel processors and achieves a computing speed of up to 518 gigaflops. A current CPU contains around 80 million transistors that process computing operations one after another. A Tesla GPU has 700 million transistors, allowing it to process more computing steps in parallel.
