Desktop Cray brings the ultrafast world of Supercomputing to your Windows, Macintosh or Linux computer... without all the trouble or expense. From business to pleasure, office work to game play... Desktop Cray delivers "All the speed you need!"
The Need for Speed
Todays modern applications demand the maximum power available... and more! Now, Desktop Cray is here to deliver on the promise of the desktop and beyond. Early alpha testers report seeing acceleration improvements of multiple orders of magnitude throughout the computing environment. Graphics, System Ops, and other applications are equally addressed by Desktop Cray's blend
of our sleek Know-Code™ parallel code deployment. Music apps, Video programs and C++ compilers will all benefit from the acceleration via Desktop Cray.
Auto System Selection
Desktop Cray's Platform Equalization Technology (PET) will auto select from all Windows, Macintosh and Linux systems - then configure and optimize itself. Desktop Cray utilizes its proprietary Advanced System Profiler Builder data engine (ASPB) configuring the user's real desktop model with the system listed on screen.

Cooling is critical in true supercomputing environments for hardware failure is a primary source of high speed computer malfunction. Four different subsystems for cooling are on offer: Blown Air, Immersion, Conduction, and a user's defined External. Please see the website for details on downloading additional cooling modules.

Fully implemented RAM caching though our Super Advanced Cache Memory Architecture (SMCMA) providing virtual dedicated memory delivering enhanced accelerated throughput. Accessing data from main memory is magnitudes faster with the fastest memory boosting onboard virtual L4 cache. The L4 cache boosts processor function by providing fast access to data and application code at unheard of speeds.
Tech Note: Desktop Cray prefetch mode is the world's fastest FIDO fetch.

Operating Environment
User selectable 256, 128 or 64 bit operational environment on demand. Through careful selecting and balancing of options, simulation of a 512 bit full bandwidth system is deliverable. Optimum performance and efficient operation from all levels of the system architecture is maintained automatically by Desktop Cray. Accordingly, to enhance performance at the system level, our architecture has been designed to accommodate the ultra high volume of system traffic required for mega complex processing. The prominent features of this balanced design include a high-speed system bus fast DDR main memory, virtual quad ATA hard drive interfaces (see HiPPI), a direct PCI2 bus, and fully integrated duplexing high-speed I/O interfaces.

Vectoring (VGPU)
Desktop Cray does graphics acceleration far beyond the original Crays. The extreme power of the Virtual Graphics Processing Unit (VGPU) simulating a 10124 MB frame buffer populated with ultra fast DDR RAM, the subsystem can perform 32.3 trillion operations per second. We incorporate a new WarpSpeed™ Memory Architecture III, an advanced DDR memory interface that delivers 100.26-gigabyte-per-second memory bandwidth and pumps out a groundbreaking 512 million triangles — and 12.8 billion textured pixels — per second. Our Phantom FX 12 Engine with quad programmable Matrix-X Shaders, advanced Voxel Shaders and 3D+ textures brings to life special effects on a metascale.

Load Balancing
Desktop Cray features a pipelined parallelization program that is able to perform segments of a complicated problem on many virtual processors. Two approaches exist in Desktop Cray, periodic and automatic. Periodic load-balancing occurs after a set number of steps while automatic occurs when the particle number is unacceptable. Our Load-Balancing algorithms broadcast the current numbers of particles on each process required for determining the numbers sent and received by each process.

Course or fine, user selectable, infinitely adjustable. The granularity algorithms are a natural outgrowth of our pioneering work with some of the finest Supercomputing facilities in the world. Adjusted properly, the granularity slider will amaze you with it's results and repay continued manipulations and fine tuning. Like all of Desktop Cray's controls, it's built in Memory Retention Architecture (MRA) will insure that every move made is remembered by the master system, and contributes to overall system load balancing.

Message Passing
Three Message Passing models are available with this release: Barrier, Gather, and Scatter. They work in very different fashion to deliver various low level messages deep within the DC codebase. For general purpose, most users will find the Barrier method to work fine. Scatter is best when intertwined and de-fragmented methodologies are needed to accommodate ultra rapid delivery of random computational messages.

Perhaps the most interesting and flexible variable available to the advanced Desktop Cray user. Upon activation, the Matrix works in two ways. Through immediate visual feedback, the user can see the various processes operating at any moment on all the systems, both real and virtual. With this latest technology, the entire ultra high speed activities of all processors are shown in an aggregated slow motion, in effect, allowing the user great insight into the various ongoing routines in progress. Secondly, the user can put the Matrix into active mode and if quick enough can bring the system up to one or two magnitudes faster through manipulations of various defined parameters in real time.

HIPPI is an abbreviation for "High Performance Parallel Interface". Desktop Cray follows the ANSI standard ( ANSI X3T9.3 document number X3T9.3/90-043). Our implementation of HIPPI utilizes a virtual point to point link for high speed data transfer at 100 MBytes/sec or 200 MBytes/sec . Using on screen non blocking crossbar switches, very powerful HIPPI networks can be build. This creates a network allowing 6.4 GBits/sec in each direction. Desktop Cray's HIPPI specifications can be found as working documents from the HIPPI Network Forum.

Block strided DMA fully supported. See technotes for more details.

Cray Emulation (Modeling)
All basic models, historical and current are represented in the emulation subsection of the software. With a simple user definable selection, one can change the processing configuration of the machine to that of the Cray of choice. Experimentation with the various models, and ad-hoc testing by the user is the best way to feel which emulation model suits the job at hand. For more information on Cray models see

Contact Us | Forums
Copyright © 2003 XOSX. All rights reserved.