GPUs Cuda & OpenCLIf anybody likes reading, I thought I would post a few notes
Future Hardware for Intel 2010
GP-GPU: General-purpose computing on graphics processing units
Though we have been hearing about GP-GPU foryears now, and academic research goes back quite a ways, it's still atechnology that is in its infancy. The latest graphics cards are pretty good atthis stream processing general compute stuff, but still have some seriousdrawbacks when it comes to task switching and alternating between graphics andGP-GPU computational tasks.
Over the next couple of years, we should see an increased emphasison how well GPUs run not only graphics applications, but general purposeapplications, with hardware features devoted to speeding up GP-GPU tasks. Thishas already begun, of course, and will accelerate over time. Ideally, yourgraphics card should be able to smoothly balance the demands of a graphicsapplication with general purpose computing tasks, whether it is video encoding,protein folding, game physics, or AI computation.
No discussion of GP-GPU's future would be complete withoutmentioning Intel's future GPU, code-named Larrabee.
Larrabee is the codename for a GPGPUchip that Intel is developing separately from its current line of integrated graphics accelerators.The chip was to be released in 2010 as the core of a consumer 3D graphics card,but these plans were cancelled due to delays and disappointing earlyperformance figures.[1] Larrabee will now be released as a platformfor research and development in computer graphics and HPC. Afuture version of Larrabee may eventually power a consumer graphics card, butIntel has not discussed specific plans.[2]
On May 25, 2010 the Technology@Intelblog announced that Larrabee would not be released as a GPU, but instead wouldbe released as a product for High Performance Computing competing with the Nvidia Tesla.
Microsoft beatsIntel, AMD to market with CPU/GPU combo chip Aug 24/10
At Hot Chips today, Microsoft's Xbox team unveiled details of thesystem-on-a-chip (SoC) that powers the newer, slimmer Xbox360 250GB model. Produced on the IBM/GlobalFoundries 45nm process, it'sfair to say that the new SoC (pictured above) is the first mass-market,desktop-class processor to combine a CPU, GPU, memory, and I/O logic onto asingle piece of silicon. The goal of the consolidation was, of course, to lowerthe cost of making the console by reducing the number of different chips neededfor the system, shrinking the motherboard, and reducing the number of expensivefans and heatsinks.
NVIDIA’s annual GPU Technology Conference If Intel has IDF and Apple has the World WideDevelopers Conference, then GTC is NVIDIA’s annual powwow to rally theirdevelopers and discuss their forthcoming plans.
The first GPU is called Kepler (as in Johannes Kepler themathematician), which will be released in the 2nd half of 2011. Atthis point the GPU is still a good year out, which is why NVIDIA is not talkingabout its details just yet. For now they’re merely talking about performance inan abstract manner, in this case Kepler should offer 3-4 times the amount ofdouble precision floating point performance per watt of Fermi. With GF100NVIDIA basically hit the wall for power consumption (and this is part of thereason current Tesla parts are running 448 out of 512 CUDA cores), so we’rebasically looking at NVIDIA having to earn their performance improvementswithout increasing power consumption. They’re also going to have to earn theirkeep in sales, as NVIDIA is already talking about Kepler taking 2 billiondollars to develop and it’s not out for another year.
Now a new technology in Mac OS X Snow Leopard (apple.com/macosx/technology/)
called OpenCL takes the powerof graphics processors and makes it available for general-purpose computing. Nolonger will graphics processors be limited to graphics-intensive applicationssuch as games and 3D modeling. Instead, once developers begin to use OpenCL intheir applications, you’ll experience greatly improved speed in a wide spectrumof applications.
OpenCL automatically optimizes for the kind of graphics processor in theMac, adjusting itself to the available processing power. OpenCLprovides consistent numeric precision and accuracy, fixing a problemthat has hampered GPU-based programming in the past.
NVIDIA GPU’s go beyond graphics by harnessing the CUDA architecturefor highly parallel GPU Computing or off-loading H.264 content with adedicated secure decode engine. These capabilities bring a whole newlevel of performance to Mac OS X based platforms at a much lower powerbudget.