Author Topic: Cuda 2.0 available  (Read 4113 times)

0 Members and 1 Guest are viewing this topic.

Stefan

  • Consigliere
  • ****
  • Posts: 619
Cuda 2.0 available
« on: August 19, 2008, 05:46:46 PM »
Quote
New Features
--------------------------------------------------------------------------------

  Hardware Support
  o  Additional hardware support:
     - GeForce GTX 280
     - GeForce GTX 260
     - GeForce 9800 GX2
     - GeForce 9800 GTX
     - GeForce 9600 GT
     - GeForce 8800 GS
     - GeForce 8600 GTS
     - Quadro FX 3700
     - Quadro NVS 130M
     - Quadro NVS 135M
     - Quadro NVS 140M
     - Quadro NVS 140M
     - Quadro NVS 135M
     - Quadro NVS 130M
     - Quadro FX 3600M
     
  Platform Support
  o  Additional OS support
     - Windows Vista 32-bit
     - Windows Vista 64-bit

  New ISA Support SM 1.2
  o Support for any() and all() intrinsics
  o Support for atomic operations on shared memory
  o Support for 64-bit atomic operations on global memory
     - atomicAdd()   (unsigned 64-bit int)
     - atomicExch()  (unsigned 64-bit int)
     - atomicCAS()   (unsigned 64-bit int)

  Double Precision Computing with SM 1.3
  o Compiler support for double-precision math
     - Datatype 'double' compiles to native FP64 types when using SM 1.3
       Note that math functions in the CUDA math library are overloaded.
       In general, there are three prototypes for each math function:
         (1) double <func-name>(double), e.g. double log(double)
         (2) float  <func-name>(float),  e.g. float log(float)
         (3) float  <func-name>f(float), e.g. float logf(float)
       In particular, note that passing a float argument always results
       in a float result [variants (2) and (3) above].
  o CUBLAS Library Support
     - Added the BLAS1 functions:
        * cublasIdamax()
        * cublasIdamin()
        * cublasDasum()
        * cublasDaxpy()
        * cublasDcopy()
        * cublasDdot()
        * cublasDnrm2()
        * cublasDrot()
        * cublasDrotg()
        * cublasDrotm()
        * cublasDrotmg()
        * cublasDscal()
        * cublasDswap()
     - Added the BLAS2 functions:
        * cublasDgemv()
        * cublasDger()
        * cublasDsyr()
        * cublasDtrsv()
     - Added the BLAS3 functions:
        * cublasDgemm()
        * cublasDsymm()
        * cublasDsyrk()
        * cublasDsyr2k()
        * cublasDtrmm()
        * cublasDtrsm()
        * cublasZgemm()

  API Features
  o 3D texture API   
     - cudaMalloc3D
     - cudaMalloc3DArray
     - cudaMemset3D
     - cu(da)Memcpy3D
     - cu(da)Memcpy3DAsync
     - cuArray3DCreate
     - cuArray3DGetDescriptor
     - CUDA_MEMCPY3D and CUDA_ARRAY3D_DESCRIPTOR structures
  o Improved Direct3D interoperability API
     - cudaD3D9SetDirect3DDevice
     - cu(da)D3D9GetDirect3DDevice
     - cu(da)D3D9RegisterResource
     - cu(da)D3D9UnregisterResource
     - cu(da)D3D9MapResources
     - cu(da)D3D9UnmapResources
     - cu(da)D3D9ResourceSetMapFlags
     - cu(da)D3D9ResourceGetMappedPointer
     - cu(da)D3D9ResourceGetMappedSize
     - cu(da)D3D9ResourceGetMappedPitch
     - cu(da)D3D9ResourceGetSurfaceDimensions
     - cuD3D9CtxCreate
  o Improved OpenGL interoperability API
     - cuGLCtxCreate
  o Context migration API
     - cuCtxAttach
     - cuCtxDestroy
     - cuCtxDetach
     - cuCtxSynchronize
     - cuCtxPushCurrent
     - cuCtxPopCurrent
  o Async constant memory update
     - cudaMemcpyToSymbolAsync
  o Improved device attribute query
     - cuDeviceGetAttribute
     
  Performance Enhancements
  o Improved device->array memcpy performance

Geforce 177.89 is OK, no need to roll-back to 177.84
Samples are now working with 8800GTX  :thumbup:

Source: Nvidia

« Last Edit: August 19, 2008, 05:51:44 PM by Stefan »
Systeminfo: CPU-Z - GPU-Z - GPUCapsViewer