FASTRA is a special PC designed for performing large-scale scientific computations. In particular, we use FASTRA for our work on tomography. FASTRA contains four NVIDIA 9800GX2 graphics cards, that each contain two Graphics Processing Units (GPUs). By using these eight GPUs in parallel, we can obtain supercomputer performance (equal to a cluster of hundreds of PCs within a single PC.
We are the research group ASTRA (All Scale Tomographic Reconstruction Antwerp), part of the Vision Lab at the University of Antwerp, Belgium. Our research focus is on the development of new algorithms and techniques for 3D image reconstruction by tomography. Tomography has applications on all length-scales, down from the nanoscale up to lightyears. We seek to combine the general properties of all these specific problems, and develop general techniques for solving them.
Tomography is a technique for obtaining a three-dimensional image of an object, without taking the object apart. Instead of making the image directly, it is computed from a series of projection images, recorded while viewing the object from a range of angles. The most common example of tomography can be found in a medical CT-scanner. An assembly, consisting of an X-ray source and an X-ray detector, rotate around the patient, while acquiring X-ray photos from a full range of angles. A 3D image of the patient is then computed from the series of X-ray photos.
As FASTRA is not a general purpose computer, its power heavily depends on the particular application it is used for. In our research group, we focus on tomography computations. For these computations, which can easily take weeks on a normal PC, FASTRA performs as fast as more than 300 Intel CPU cores (Core Duo, running at 2.4GHz). Our local supercomputer cluster consisting of 512 Opteron cores, which cost millions of euros when constructed in 2005, is actually outpaced by FASTRA in some cases.
To the best of our knowledge, FASTRA is the first consumer hardware PC in the world that contains eight GPUs, resulting in tremendous computing power. For our application, FASTRA obtains similar performance to a large PC cluster, while fitting on top of an office desk. FASTRA is made exclusively of consumer hardware components, which makes it extremely affordable compared to other supercomputing solutions. To be exact, it cost us less than 4000 euro! FASTRA requires less than 1500W of power, which makes it a green computing platform compared to cluster solutions, that use orders of magnitude more power.
We develop new reconstruction techniques, that are, in some cases, much more powerful than conventional methods. They are capable of computing accurate 3D images from just a few X-ray photos, by exploiting prior knowledge on the object being scanned. However, our methods require far more computation time than methods typically used in practice. Reconstructing a 3D volume consisting of 1024x1024x1024 volume elements sometimes takes us several weeks on a single PC. We need a supercomputer to reduce the reconstruction time to less than an hour for such a volume.
Performing general computations on the graphics card is gradually becoming more common nowadays. As far as we know, FASTRA is the first regular (i.e., built from consumer hardware) single PC that combines the power of eight GPUs, in four graphics cards. Building such a system is not a trivial undertaking, as it requires careful consideration of hardware components, power and cooling. The fact that FASTRA consists of consumer hardware is highly important, as it creates a supercomputing platform for less than 4000 euros, so that any interested group can afford it.
We selected the components of FASTRA, performed the GPU programming and the actual testing and running. The system was assembled by Tones.be, a computer hardware shop in Belgium, who did a great job in delivering a tidy build to us.