add amd appeared ars attempts bring cases chip combined computing conference configuration core creative de directly documentation doing easily fine formats further given goes h https india job july lower mac nvidia oct order parallel pci possible press print problems processing programmer ray recorded regular smaller specification speeds super table wrapper
In 2007 the first toolkits appeared to let a programmer create programs that are executed directly by the graphics processing units on some high end NVIDIA and ATI graphics cards. The potential for speed improvement may be a factor of a hundred (or more) in certain cases.
- 2010-Jul-16: Boosting the performance of a Python Mandelbrot calculator up to 22937 times faster using pyCUDA. Some more notes on using pyCUDA. 
- 2010-Jun-30: CUDA graphics engines have been used to accelerate the calculation of MD5 hashes to speed up password cracking attempts. Based on the timings that are published here a password length of 10 characters is getting to be pretty weak - that would take a single machine 50 years to search, so a project that combined these machines in a distributed fashion could easily crack 10 character passwords in days or less.  
- 2009-Aug-13: A comparison between the NVIDIA CUDA and ATI Stream GPGPU systems with a focus on the video transcoding problem. 
- 2009-Jun-17: CUDA has been used to accelerate matrix calculations in MATLAB, with speed improvements of about 10 times on suitable problems. 
- 2009-May-19: NVIDIA has released some free PhysX and CUDA software for users of GeForce 8, 9 and 200 series graphics cards. This also includes some CUDA applications like a Folding@Home client and a trial version of the Badaboom video transcoder. There is a discussion of Badaboom here. When I tried this on a 5 minute MPEG2 clip of some recorded TV I found Badaboom taking 338 seconds while a StaxRip run took 240 seconds, this was on a Q6600 machine with a GeForce 8600 GT card, so not much use for me (except that it off loads the CPU during the encode). Perhaps they will speed things up by the time it is commercially released. MaximumPC takes a look at Badaboom and compares it to Handbrake. Tom's takes a look at five applications that use the CUDA engine to speed up processing. 
- 2009-Apr-20: The Caustic Graphics ray tracing processor board sounds like it might pack a lot of horse power, perhaps this could also be used as a general purpose processor? Discussed here on Slashdot. A picture of the prototype board along with some samples are here, note the current board is only doing 3-5 frames per second so this is a ways off the necessary 30-60 FPS. 
- 2009-Mar-13: pycuda (home page is here, the documentation is here) provides access to Nvidia's CUDA parallel computation API. So if you really want to crunch a lot of numbers, now is your chance to do it from the comforts of Python. The July'08 meeting of ChiPy is taking a look at CUDA. An introductory presentation on using PyCuda. PyopenCL is a Python wrapper for OpenCL.   
- 2008-Dec-10: The OpenCL API specification has been released (also on Engadget), this is an open API for executing general purpose code on GPUs. More on OpenCL here. 
- 2008-Nov-14: AMD is planning to make more of the ATI video cards available for GPGPU type use. 
- 2008-Oct-19: ATI's Radeon HD 4800 spec sheet mentions that their hardware can assist MPEG2 to H.264 encoding, improving speed by 1.8 times on full 1080P and up to 19 times on lower resolution video. They call this Accelerated Video Transcoding (AVT). In the fine print (which is rendered in a smaller font with a faint grey colour to make it illegible) they say:
This may vary depending on your system configuration and video formats. Using an Intel Core 2 Duo E8500 3.16 GHz based PC, AMD was able to achieve GPU accelerated transcoding speeds up to 19x faster using Cyberlink PowerDirector than when using the same CPU alone with MainConcept encoder in Adobe Premiere CS3. Using the same system, full 1080p files were converted 1.8x faster than real-time.
The Cyberlink PowerDirector pages don't say anything about this. 
- 2008-Oct-19: VIA is getting into the GPGPU game. Discussed here on Slashdot. 
- 2008-Oct-03: The Sony Cell chip is going to be available on a PCI-E card (the PxVC1100 from Leadtek) to add H.264 video compression/decompression acceleration to PCs, let's hope this does a bettr job than NVidia's recent attempts. 
- 2008-Sep-24: Adobe is adding GPGPU based acceleration to its Creative Suite 4. 
- 2008-Sep-16: Could the general purpose GPU be just a short phase heralding the return of the regular CPU? Discussed here on Slashdot. 
- 2008-Aug-15: Interactive ray tracing has been made possible (in Aug'08) by the assistance of NVIDIA's Quadro GPUs. So by about 2010 the sort of power needed to ray trace at 30fps at resolutions of up to 1920x1080 will be appearing in the high end gaming graphics cards. 
- 2008-Aug-12: NVIDIA's Quadro Plex D CUDA desktop systems bring supercomputer powers (up to 480 CUDA cores) into the $10K price range. 
- 2008-Feb-16: NVIDEA is looking at using a physics engine to CUDA port to allow any GeForce 8 GPU to provide PhysX support. 
- 2008-Feb-05: NVIDIA is going to buy AGEIA (who make the PhysX accelerator engines), could this mean the NVIDIA is getting more serious about using graphics engines to accelerate general purpose computing? 
- 2007-Nov-08: AMD announces the FireStream 9170, a dedicated stream processor capable of up to 500G Flops. 
A set of tutorials on GPGPU given at the ICCS 2006 conference. 
The Wikipedia article on GPGPU technology.
- 2007-Sep-17: An early article about this on Slashdot, the NVIDIA GeForce 8800 is mentioned. The NVIDIA press release announcing their CUDA development system is here.