Manifold Sets New GIS Record for Supercomputing Desktop
For Immediate Release
NEWSFLASH! Yet another world record set in Denver! See the update to this press release for the latest record of 1,920 simultaneous processing cores!
Redmond USA — 16 March 2009 — Manifold.net today announced a new world record for the number of processors used in a personal computer for Geographic Information Systems (GIS) processing. At the company's 2009 European User Meeting in London, Manifold demonstrated an upcoming new software product that simultaneously utilized over 1,440 processor cores to perform a remote sensing image computation at supercomputer speed with over 3.5 teraflops of performance. Manifold demonstrated the new software on a desktop 64-bit Windows PC equipped with three NVIDIA GTX 295 GPU cards costing less than $500 each. (Illustration at right shows the demonstration hardware.)
The demonstration showed how a desktop Windows PC using inexpensive consumer components and running a $245 Manifold GIS package can run over one hundred times faster than any other GIS or remote sensing software package, even those costing tens of thousands of dollars more than Manifold.
Modern technology provides significantly lower cost as well as much greater performance. The desktop supercomputer technology utilized costs almost nothing to deploy, as little as $80 to start, yet it far outperforms any competitor's software, even packages costing 50 times as much. By using inexpensive, massively parallel GPUs Manifold can provide teraflops of computing performance for only a few hundred dollars.
NVIDIA GPUs (Graphics Processing Units) employ multiple parallel processing units for faster performance. Manifold's new software uses massively parallelized programming to execute general purpose computations within those parallel processing units for much faster performance than possible on the main CPU. Because of NVIDIA's success in consumer graphics markets, over 100 million NVIDIA GPUs have already been installed in computers throughout the world, ready to provide supercomputer speed to anyone using Manifold's new product. NVIDIA GPU cards are widely sold by many vendors for as little as $80 per card. (Illustration at left shows a GTX 295 card made by EVGA.)
The new Manifold technology previewed in London will ship in the first half of 2009 and will be offered to all Manifold Release 8 licensees for an update fee between $50 and $100 per license.
Manifold is the world leader in modern GIS, with a focus on providing advanced technology at an exceptionally low price.
Manifold is the first mainstream GIS company to have delivered production products that run native 64-bit in Windows, that support Windows Vista and Windows Server 2008, that support multi-core / multi-processor operations in Windows, that support SQL Server 2008 spatial enhancements and that utilize massively-parallel NVIDIA GPU architectures to achieve supercomputer performance on a desktop.
Is this a hardware product or a software product? - This announcement covers a preview of a new software product. This new software product takes advantage of recent advances in hardware technology that are becoming ubiquitous. Taking advantage of modern hardware allows Manifold software to run much faster than old-fashioned software.
Is this a new graphics standard like OpenGL or DirectX? - No. This is new applications software technology that uses hardware that was originally developed to speed up graphics but which now can be used to speed up non-graphics programs as well. The NVIDIA chips employed are still called "GPU" (Graphics Processing Unit) chips because they continue to be used to speed up graphics and because it helps to have a name that makes it clear these are accessory processing chips in addition to the main CPU on the motherboard. However, the same super-powerful, parallel computer machinery inside those GPU chips that makes graphics so much faster can now also be employed as general-purpose supercomputers to run almost any application faster that requires complex mathematical computations. This press release announces a new record using 1,440 processor cores within such GPUs to make Manifold's GIS application run at supercomputer speeds. The ability to run more cores inside a GPU allows an application like Manifold to run even faster.
Does this new capability require special hardware? - Yes and no. Because the hardware required is very widespread, with over 100 million NVIDIA chips already installed worldwide, it is not at all unusual or exotic, so in that sense no "special" hardware is required. However, the PC in use must have an NVIDIA GPU that supports NVIDIA's CUDA standard, either installed as a plug-in card or built into the motherboard. Over 100 million graphics cards have been shipped with NVIDIA GPUs on them and in addition millions of computer systems from laptops to specialized dedicated systems have been shipped that contain NVIDIA GPUs. If a PC does not already have an NVIDIA GPU installed, there are many NVIDIA-based cards ranging in price from $50 to around $500 that may be purchased from virtually any vendor of graphics cards.
Does this new Manifold technology work with ordinary multicore CPUs like Core i7? - Yes, although that is a related, but different, technology than the use of massively parallel GPUs. The record-setting demonstration in London launched eight threads in the host Intel Core i7 CPU (running two hyperthreads through each of the four physical cores in the Core i7) to dispatch a massive number of parallel threads into the 1,440 GPU cores. The new Manifold technology is heterogeneous parallel technology in that it can simultaneously use as many CPU cores as are available to dispatch many threads into as many GPU cores as are available, including dispatch of multiple threads into the same GPU core, to different GPU cores, or to GPU cores of different types. (Illustration at right shows how Manifold can dispatch heterogeneous multiple threads from CPU cores to GPU cores.)
What is CUDA? - CUDA stands for "Compute Unified Device Architecture." It is a software interface provided by NVIDIA to enable programmers of software applications to write parallelized code using familiar syntax. Programs written to the CUDA standard will work on any NVIDIA GPU supported by CUDA without any need to recompile the program or to re-write it to work on different NVIDIA GPUs. This makes it possible for applications programmers to create software that will take advantage of whatever CUDA-enabled NVIDIA GPU has been installed as well as ensuring that their software will continue to run correctly if the user upgrades hardware to a newer or faster NVIDIA GPU. Manifold uses CUDA to assure correct function with any CUDA-enabled NVIDIA GPU. For more information on CUDA, see the NVIDIA CUDA home page.
Why is the hardware used so inexpensive? - Although they are used in this case in a non-graphics application, the GPU cards employed in this demonstration are primarily sold for use as graphics cards. In ordinary consumer usage the 480 stream processors in each GTX 295 card are used to render 3D graphics in applications such as computer gaming. The gaming market is a highly competitive consumer market where graphics cards must provide very high performance at a cost ordinary consumers can afford. It is also a high volume market with many millions of units being sold. The high volume of consumer graphics cards that are sold provides economy of scale that makes such powerful hardware inexpensive and widely available. NVIDIA's great innovation is to make it possible for applications developers like Manifold to use such affordable hardware to provide supercomputer speed in applications like GIS.
Does this new Manifold technology work with NVIDIA Tesla or Quadro products? - Yes. NVIDIA's Tesla line is a series of high performance computing products based upon NVIDIA GPUs that is marketed by NVIDIA. The Quadro line is a series of high performance professional graphics products based upon NVIDIA GPUs also marketed by NVIDIA. Both are supported flawlessly by this new Manifold technology.
Does this new Manifold technology work with AMD or Intel products? - The particular demo shown in London is aimed at NVIDIA GPUs. However, Manifold is working closely with AMD to assure that this new massively parallel software architecture also will work with AMD's Firestream massively parallel technology. Manifold has also designed the software to work with any future Intel massively parallel processors as well.
Can this new Manifold techology utilize G200 NVIDIA double-precision floating point? - Yes. Manifold's new technology automatically adapts to make the best use of whatever NVIDIA GPU is present, correctly taking into account different capabilities such as the availability of single or double precision floating point, available memory and other factors.
Can it work with four GTX 295 cards instead of just three? What are the limits? - The practical limit is how many GPU cards can be plugged into a computer and is not set by any fundamental limit in the software techology. Most high end NVIDIA cards are double-wide PCI-E cards so most PCs only have room for three such cards. Using a larger motherboard that provides enough PCI-E slots and enough room between slots to allow hosting four double-wide PCI-E cards would allow running four GTX 295 cards for a total of 1920 stream processors and over four and a half teraflops of performance. Manifold routinely runs such configurations during development. However, the public demonstration in London required use of a PC small enough to transport to London within an international standard 20-inch airline carry-on bag. Use of such a small form-factor PC as illustrated on this page provided room for a maximum of three GTX 295's with 1,440 processors yielding three and a half teraflops.
How is the new technology different from the CUDA support in Manifold's existing Release 8 GIS product? - Manifold's Release 8 GIS product was the first and is still the only GIS product to use NVIDIA CUDA for breathtaking speed increases, but it does so in a limited way. Manifold Release 8 provides almost three dozen CUDA-enabled functions for computations within an optional surface transform dialog. The new technology previewed in London runs automatically within hundreds of functions as a built-in part of Manifold and scales to many thousands of GPU stream processors. It utilizes a new, deeply parallel heterogeneous parallel processing architecture to take advantage of parallel processing within multicore CPUs to service massively parallel processing within GPUs. Since Release 8, Manifold has written millions of lines of code to make the new technology possible.
What was the remote sensing function computed? - The demonstration took a 23 GB image and sliced it into many thousands of checkerboard tiles for a convolution matrix computation using all 1,440 GPU processing cores at once. Convolution matrices are used in many GIS and remote sensing computations. There are hundreds of such functions supported by Manifold's new technology, with the point of the public demonstration simply being to show that it was indeed possible to utilize over 1400 processors to achieve dramatically faster speed in a GIS / Remote Sensing application.
What are the full specs on the hardware used? - The demo machine used an ASUS P6T6 WS Revolution motherboard with an Intel Core i7 Extreme processor, 12 GB of RAM, dual 1 TB disks in a RAID mirror, CoolerMaster 1250W power supply and three EVGA GTX 295 cards.
How can I get access to this new technology? - The fastest way will be to buy a copy of Manifold's existing Release 8 product and to then take advantage of the special upgrade offer when the new technology is released sometime in the first half of 2009. When launched, Manifold Release 8 checks for updates so as soon as updates to the new technology are available, Manifold will tell you they are available.
Buy Now via the Online Store!
This new technology will be made available to Manifold Release 8 licensees in the first half of 2009 at a low upgrade price. Buy Manifold System Release 8 today on the Online Store. The store is open 24 hours / seven days a week. Orders received by 10 AM Pacific time will process the same business day. Manifold System is delivered by download for rapid delivery and no shipping delays or shipping fees!
Please contact us at firstname.lastname@example.org. We would like you to be happy with your licensing of Manifold products, so please do not hesitate to ask any questions before placing an order.
All product and company names herein may be trademarks of their registered owners