News etichettate con "CUDA"

Venerdi 10 Agosto 2012     

I ricercatori del Computer Science and Artificial Intelligence Laboratory (CSAIL), del MIT (Massachusetts Institute of Technology), hanno annunciato di aver sviluppato un nuovo linguaggio di programmazione chiamato "Halide".

 Halide è stato progettato per semplificare lo sviluppo di applicazioni riguardanti l'Image Processing.

Adding a new function to an image-processing program, or modifying it to run on a different device, often requires rethinking and revising it from top to bottom.

Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) aim to change that, with a new programming language called Halide. Not only are Halide programs easier to read, write and revise than image-processing programs written in a conventional language, but because Halide automates code-optimization procedures that would ordinarily take hours to perform by hand, they’re also significantly faster.

Rilasciato sotto licenza MIT, Halide è disponibile per piattaforme x86-64/SSE, ARM v7/NEON e CUDA. Il compilatore è attualmente disponibile per le recenti release di Mac OS X e Linux. Il supporto per sistemi Windows è tecnicamente fattibile, ma per ora non è stato testato su sistemi Windows.

 
Venerdi 07 Ottobre 2011     

NVidia ha rilasciato il SDK 3.1 di PhysX, download a partire da questa pagina del PhysX Developer Support

General

  • VC10 support has been introduced.
  • VC8 support has been discontinued.
  • Namespaces cleaned up.
  • Extensions, Character Controller and Vehicle source code made available in binary distribution.
  • Added x86,x64 suffix to PxTaskCUDA.dll
  • Removed boolean return value from PxScene::addActor(…), and similar API calls.
  • Added Mac OS X, Android and Linux to the list of supported platforms. See Supported Platforms below for details.
  • Upgraded GPU tech to CUDA 4.
  • Cleaned up a large number of warnings at C++ warning level 4, and set SDK to compile with warnings as errors.
  • Removed individual sample executables in favor of SampleAllInOne from PC and console builds.
  • Fixed alpha blending in samples.
  • Simplified some code in samples.
  • Improved ambient lighting in samples.
  • Made samples work with older graphics cards.
  • Renamed some XBOX 360 specific files and folders.
  • Improved and added more content the user’s guide.
  • No longer passing NULL pointers to user allocator to deallocate.
  • Various improvements to Foundation and classes shared with APEX.

Rigid Bodies

  • Rigid Body: High performance alternative convex narrow phase code available to source licensees. See PERSISTENT_CONTACT_MANIFOLD in the code.
  • Significant advancements in the continuous collision detection algorithm.
  • Optimizations and robustness improvements for articulations.
  • Added some helper code to the API.
  • Added sleep code for articulations.
  • Added support for vehicles with more than one chassis shape.
  • Solver iteration count for articulations.
  • Articulation limit padding configurable.
  • The reference count of meshes does now take the application’s reference into acount as well and thus has increased by 1 (it used to count the number of objects referencing the mesh only). Note that a mesh does only get destroyed and removed from the list of meshes once the reference count reaches 0.
  • Fixed autowake parameter sometimes being ignored.
  • Constraint solver optimizations.
  • Improved behavior of character controller on steep slopes.
  • Binary serialization now saves names.
  • Removed some descriptors from API.
  • Removed the angular velocity term in the joint positional drive error formula.
  • Fixed bug in capsule sweep versus mesh.
  • Fixed a crash bug in the tire model.
  • Fixed crashing of single link articulations.
  • Fixed bug related to removing elements of an aggregate.
  • Fixed swapped wheel graphs in sample vehicle.
  • Fixed some slow moving bodies falling asleep in midair.
  • Fixed missing collisions after a call to resetFiltering.
  • Fixed broken autowake option in setAngularVelocity.
  • Fixed D6 joint linear limits being uninitialized.
  • A large number of misc. bug fixes and optimizations.
  • Improved documentation and error messages associated with running out of narrow phase buffer blocks.
  • Added articulation documentation.
  • Expanded manual sections on joints.
  • Improved reference doc for PxSceneQueryHitType.
  • Added reference doc for PxSerializable.

Particles

  • Particle index allocation removed from SDK. Added index allocation pool to extensions.
  • Replaced GPU specific side band API PxPhysicsGpu and PxPhysics::getPhysicsGpu() with PxParticleGpu.
  • Memory optimizations on all platforms and options to reduce memory usage according to use case with new per particle system flags:
    • PxParticleBaseFlag::eCOLLISION_WITH_DYNAMIC_ACTORS
    • PxParticleBaseFlag::ePER_PARTICLE_COLLISION_CACHE_HINT
  • Fixed rare crash appearing with multi-threaded non-GPU particle systems and rigid bodies.
  • Fixed particles leaking through triangle mesh geometry on GPU.
  • Fixed fast particles tunneling through scene geometry in some cases.
  • Fixed erroneous collision of particles with teleporting rigid shapes (setGlobalPose).
  • Fixed particle sample behavior with some older GPU models.
  • Fixed a GPU particle crash bug.

Cloth

  • A new solution for simulating cloth and clothing.

Deformables

  • Deformables are deprecated and will be removed in the next release. There is a new optimized solution for clothing simulation (see documentation on PxCloth for details).
 
Martedi 07 Dicembre 2010     

nVidia ha presentato la sua nuova scheda video GeForce GTX 570, http://bit.ly/g1N9YD. La nuova scheda supporta le librerie DirectX 11 con Shader Model 5.0, le librerie OpenGL 4.1, HDMI 1.4a. La dotazione di RAM è di 1280 MB GDDR5, la memory bandwidth è di 152 GB/sec.

GeForce GTX 570 Features
GPU: GF110 @ 732MHz
Shader processors: 480
Streaming multiprocessors: 15
Memory: five memory controllers (5x256MB) for a total of 1280MB of GDDR5 @ 950MHz real speed (or 1900MHz DDR or 3800MHz effective, see here for details:Graphics Cards Memory Speed Demystified), 320-bit
ROPs: 40
Texture units: 60
TDP: 219W
Power connectors: two 6-pin
3-way SLI support
3D API: OpenGL 4.1 and Direct3D 11
GPU Computing: CUDA, PhysX, OpenCL 1.1 and DirectCompute
Price: around US $350
 
Mercoledi 24 Novembre 2010     
nVIDIA ha rilasciato il CUDA (Compute Unified Device Architecture) Toolkit 3.2, http://bit.ly/cplqBW, per sistemi Windows, Linux e Mac OS X. La nuova versione introduce nuove librerie matematiche, insieme alle librerie per la codifica e la decodifica H.264.