CUDA

CUDA

Developer(s)	NVIDIA Corporation
Stable release	3.2 / September 17, 2010 (2010-09-17)
Operating system	Windows 7, Windows Vista, Windows XP, Windows Server 2008, Windows Server 2003, Linux, Mac OS X
Type	GPGPU
License	Proprietary, Freeware
Website	Nvidia's CUDA zone

CUDA (an acronym for Compute Unified Device Architecture) is a parallel computing architecture developed by NVIDIA. CUDA is the computing engine in NVIDIA graphics processing units (GPUs) that is accessible to software developers through variants of industry standard programming languages. Programmers use 'C for CUDA' (C with NVIDIA extensions and certain restrictions), compiled through a PathScale Open64 C compiler, to code algorithms for execution on the GPU. CUDA architecture shares a range of computational interfaces with two competitors -the Khronos Group's Open Computing Language and Microsoft's DirectCompute. Third party wrappers are also available for Python, Perl, Fortran, Java, Ruby, Lua, and MATLAB.

CUDA gives developers access to the virtual instruction set and memory of the parallel computational elements in CUDA GPUs. Using CUDA, the latest NVIDIA GPUs become accessible for computation like CPUs. Unlike CPUs however, GPUs have a parallel throughput architecture that emphasizes executing many concurrent threads slowly, rather than executing a single thread very fast. This approach of solving general purpose problems on GPUs is known as GPGPU.

In the computer game industry, in addition to graphics rendering, GPUs are used in game physics calculations (physical effects like debris, smoke, fire, fluids); examples include PhysX and Bullet. CUDA has also been used to accelerate non-graphical applications in computational biology, cryptography and other fields by an order of magnitude or more. An example of this is the BOINC distributed computing client.

CUDA provides both a low level API and a higher level API. The initial CUDA SDK was made public on 15 February 2007, for Microsoft Windows and Linux. Mac OS X support was later added in version 2.0, which supersedes the beta released February 14, 2008. CUDA works with all NVIDIA GPUs from the G8X series onwards, including GeForce, Quadro and the Tesla line. NVIDIA states that programs developed for the GeForce 8 series will also work without modification on all future NVIDIA video cards, due to binary compatibility.

Example of CUDA processing flow
1. Copy data from main mem to GPU mem
2. CPU instructs the process to GPU
3. GPU execute parallel in each core
4. Copy the result from GPU mem to main mem

Advantages

CUDA has several advantages over traditional general purpose computation on GPUs (GPGPU) using graphics APIs.

• Scattered reads – code can read from arbitrary addresses in memory.
• Shared memory – CUDA exposes a fast shared memory region (16KB in size) that can be shared amongst threads. This can be used as a user-managed cache, enabling higher bandwidth than is possible using texture lookups.
• Faster downloads and readbacks to and from the GPU
• Full support for integer and bitwise operations, including integer texture lookups.

Limitations

• CUDA (with compute capability 1.x) uses a recursion-free, function-pointer-free subset of the C language, plus some simple extensions. However, a single process must run spread across multiple disjoint memory spaces, unlike other C language runtime environments. Fermi GPUs now have (nearly) full support of C++. Exceptions as follows:
• Code compiled for devices with compute capability 2.0 (Fermi) and greater may make use of C++ classes, as long as none of the member functions are virtual (this restriction will be removed in some future release). [See CUDA C Programming Guide 3.1 - Appendix D.6]
• Texture rendering is not supported.
• For double precision (only supported in newer GPUs like GTX 260) there are some deviations from the IEEE 754 standard: round-to-nearest-even is the only supported rounding mode for reciprocal, division, and square root. In single precision, denormals and signalling NaNs are not supported; only two IEEE rounding modes are supported (chop and round-to-nearest even), and those are specified on a per-instruction basis rather than in a control word; and the precision of division/square root is slightly lower than single precision.
• The bus bandwidth and latency between the CPU and the GPU may be a bottleneck.
• Threads should be running in groups of at least 32 for best performance, with total number of threads numbering in the thousands. Branches in the program code do not impact performance significantly, provided that each of 32 threads takes the same execution path; the SIMD execution model becomes a significant limitation for any inherently divergent task (e.g. traversing a space partitioning data structure during raytracing).
• Unlike OpenCL, CUDA-enabled GPUs are only available from NVIDIA (GeForce 8 series and above, Quadro and Tesla).

Supported GPUs

Compute capability table (version of CUDA supported)

Compute capability (version)	GPUs
1.0	G80
1.1	G86, G84, G98, G96, G96b, G94, G94b, G92, G92b
1.2	GT218, GT216, GT215
1.3	GT200, GT200b
2.0	GF100, GF104, GF106, GF108

A table of devices officially supporting CUDA (Note that many applications require at least 256 MB of dedicated VRAM).

Nvidia GeForce GeForce GTX 480 GeForce GTX 470 GeForce GTX 465 GeForce GTX 460 GeForce GTS 455 GeForce GTS 450 GeForce GT 440 GeForce GT 430 GeForce GT 420 GeForce GTX 295 GeForce GTX 285 GeForce GTX 280 GeForce GTX 275 GeForce GTX 260 GeForce GTS 250 GeForce GTS 240 GeForce GT 240 GeForce GT 220 GeForce 210/G210 GeForce 9800 GX2 GeForce 9800 GTX+ GeForce 9800 GTX GeForce 9800 GT GeForce 9600 GSO GeForce 9600 GT GeForce 9500 GT GeForce 9400 GT GeForce 9400 mGPU GeForce 9300 mGPU GeForce 9100 mGPU GeForce 8800 Ultra GeForce 8800 GTX GeForce 8800 GTS GeForce 8800 GT GeForce 8800 GS GeForce 8600 GTS GeForce 8600 GT GeForce 8600 mGT GeForce 8500 GT GeForce 8400 GS GeForce 8300 mGPU GeForce 8200 mGPU GeForce 8100 mGPU

Nvidia GeForce Mobile GeForce GTX 480M GeForce GTX 470M GeForce GTX 460M GeForce GT 445M GeForce GT 435M GeForce GT 425M GeForce GT 420M GeForce GT 415M GeForce GTX 285M GeForce GTX 280M GeForce GTX 260M GeForce GTS 360M GeForce GTS 350M GeForce GTS 260M GeForce GTS 250M GeForce GT 335M GeForce GT 330M GeForce GT 325M GeForce GT 320M GeForce 310M GeForce GT 240M GeForce GT 230M GeForce GT 220M GeForce G210M GeForce GTS 160M GeForce GTS 150M GeForce GT 130M GeForce GT 120M GeForce G110M GeForce G105M GeForce G103M GeForce G102M GeForce G100 GeForce 9800M GTX GeForce 9800M GTS GeForce 9800M GT GeForce 9800M GS GeForce 9700M GTS GeForce 9700M GT GeForce 9650M GT GeForce 9650M GS GeForce 9600M GT GeForce 9600M GS GeForce 9500M GS GeForce 9500M G GeForce 9400M G GeForce 9300M GS GeForce 9300M G GeForce 9200M GS GeForce 9100M G GeForce 8800M GTX GeForce 8800M GTS GeForce 8700M GT GeForce 8600M GT GeForce 8600M GS GeForce 8400M GT GeForce 8400M GS GeForce 8400M G GeForce 8200M G

Nvidia Quadro Quadro 6000 Quadro 5000 Quadro 4000 Quadro FX 5800 Quadro FX 5600 Quadro FX 4800 Quadro FX 4700 X2 Quadro FX 4600 Quadro FX 3800 Quadro FX 3700 Quadro FX 1800 Quadro FX 1700 Quadro FX 580 Quadro FX 570 Quadro FX 380 Quadro FX 370 Quadro NVS 450 Quadro NVS 420 Quadro NVS 295 Quadro NVS 290 Quadro Plex 1000 Model IV Quadro Plex 1000 Model S4 Nvidia Quadro Mobile Quadro 5000M Quadro FX 3800M Quadro FX 3700M Quadro FX 3600M Quadro FX 2800M Quadro FX 2700M Quadro FX 1800M Quadro FX 1700M Quadro FX 1600M Quadro FX 880M Quadro FX 770M Quadro FX 570M Quadro FX 380M Quadro FX 370M Quadro FX 360M Quadro NVS 320M Quadro NVS 160M Quadro NVS 150M Quadro NVS 140M Quadro NVS 135M Quadro NVS 130M Nvidia Tesla Tesla C2050/2070 Tesla M2050/M2070 Tesla S2050 Tesla S1070 Tesla M1060 Tesla C1060 Tesla C870 Tesla D870 Tesla S870

See the Comparison of Nvidia graphics processing units for more information.

Example

This example code in C++ loads a texture from an image into an array on the GPU:

void foo()
{
  cudaArray* cu_array;
  texture<float, 2, cudaReadModeElementType> tex;

  // Allocate array
  cudaChannelFormatDesc description = cudaCreateChannelDesc<float>();
  cudaMallocArray(&cu_array, &description, width, height);

  // Copy image data to array
  cudaMemcpyToArray(cu_array, image, width*height*sizeof(float), cudaMemcpyHostToDevice);

  // Set texture parameters (default)
  tex.addressMode[0] = cudaAddressModeClamp;
  tex.addressMode[1] = cudaAddressModeClamp;
  tex.filterMode = cudaFilterModePoint;
  tex.normalized = false; // do not normalize coordinates

  // Bind the array to the texture
  cudaBindTextureToArray(tex, cu_array);

  // Run kernel
  dim3 blockDim(16, 16, 1);
  dim3 gridDim((width + blockDim.x - 1)/ blockDim.x, (height + blockDim.y - 1) / blockDim.y, 1);
  kernel<<< gridDim, blockDim, 0 >>>(d_odata, height, width);

  // Unbind the array from the texture
  cudaUnbindTexture(tex);
} //end foo()

__global__ void kernel(float* odata, int height, int width)
{
   unsigned int x = blockIdx.x*blockDim.x + threadIdx.x;
   unsigned int y = blockIdx.y*blockDim.y + threadIdx.y;
   if (x < width && y < height) {
      float c = tex2D(tex, x, y);
      odata[y*width+x] = c;
   }
}

Below is an example given in Python that computes the product of two arrays on the GPU. The unofficial Python language bindings can be obtained from PyCUDA.

import pycuda.compiler as comp
import pycuda.driver as drv
import numpy
import pycuda.autoinit

mod = comp.SourceModule("""
__global__ void multiply_them(float *dest, float *a, float *b)
{
  const int i = threadIdx.x;
  dest[i] = a[i] * b[i];
}
""")

multiply_them = mod.get_function("multiply_them")

a = numpy.random.randn(400).astype(numpy.float32)
b = numpy.random.randn(400).astype(numpy.float32)

dest = numpy.zeros_like(a)
multiply_them(
        drv.Out(dest), drv.In(a), drv.In(b),
        block=(400,1,1))

print dest-a*b

Additional Python bindings to simplify matrix multiplication operations can be found in the program pycublas.

import numpy
from pycublas import CUBLASMatrix
A = CUBLASMatrix( numpy.mat([[1,2,3],[4,5,6]],numpy.float32) )
B = CUBLASMatrix( numpy.mat([[2,3],[4,5],[6,7]],numpy.float32) )
C = A*B
print C.np_mat()

Language bindings

• Python - PyCUDA KappaCUDA
• Java - jCUDA, JCuda, JCublas, JCufft
• .NET - CUDA.NET
• MATLAB - Jacket, GPUmat
• Fortran - FORTRAN CUDA, PGI CUDA Fortran Compiler
• Perl - KappaCUDA
• Ruby - KappaCUDA
• Lua - KappaCUDA

Current CUDA architectures

The next generation CUDA architecture (codename: "Fermi") which is standard on NVIDIA's released (GeForce 400 Series [GF100] (GPU) 2010-03-27) GPU is designed from the ground up to natively support more programming languages such as C++. It has eight times the peak double-precision floating-point performance compared to Nvidia's previous-generation Tesla GPU. It also introduced several new features including:

• up to 512 CUDA cores and 3.0 billion transistors
• NVIDIA Parallel DataCache technology
• NVIDIA GigaThread engine
• ECC memory support
• Native support for Visual Studio

Current and future usages of CUDA architecture

• Accelerated rendering of 3D graphics
• Real Time Cloth Simulation OptiTex.com - Real Time Cloth Simulation
• Distributed Calculations, such as predicting the native conformation of proteins
• Medical analysis simulations, for example virtual reality based on CT and MRI scan images.
• Physical simulations, in particular in fluid dynamics.
• Environment statistics
• Accelerated encryption, decryption and compression
• Accelerated interconversion of video file formats
• Artificial intelligence

See also

• GeForce 8 series
• GeForce 9 series
• GeForce 200 Series
• GeForce 400 Series
• Nvidia Quadro - Nvidia's workstation graphics solution
• Nvidia Tesla - Nvidia's first dedicated general purpose GPU (graphics processing unit)
• GPGPU - general purpose computation on GPUs.
• OpenCL - The cross-platform standard supported by both NVidia and AMD/ATI
• DirectCompute - Microsoft API for GPU Computing in Windows Vista and Windows 7
• BrookGPU
• Vectorization
• Lib Sh
• Comparison of MPI, OpenMP, and Stream Processing
• Nvidia Corporation
• Graphics Processing Unit (GPU)
• Stream processing
• Shader
• Larrabee
• Molecular modeling on GPU
• AMD FireStream (ATI GPUs)
• Close to Metal

External links

• Official site
• Nvidia Parallel Nsight
• Nvidia CUDA developer registration for professional developers and researchers
• Nvidia CUDA GPU Computing developer forums
• Programming Massively Parallel Processors: A Hands-on Approach
• Intro to GPGPU computing featuring CUDA and OpenCL examples
• A conversation with Jen-Hsun Huang, CEO Nvidia Charlie Rose, February 5, 2009
• Scientific Publications, Videos and Software using CUDA
• How-to Guide: Running CUDA on Visual Studio 2008
• Beyond3D – Introducing CUDA Nvidia's Vision for GPU Computing March 10, 2007
• University of Illinois Nvidia CUDA Course taught by Wen-mei Hwu and David Kirk, Spring 2009
• CUDA: Breaking the Intel & AMD Dominance
• NVidia CUDA Tutorial Slides (from DoD HPCMP2009)
• Ascalaph Liquid GPU molecular dynamics.
• CUDA implementation for multi-core processors
• Integrate CUDA with Visual C++, September 26, 2008
• CUDA.NET - .NET library for CUDA, Linux/Windows compliant
• Using NVIDIA GPU for scientific computing with BOINC software
• CUDA.CS.MSU.SU Russian CUDA developer community
• Enable Intellisense for CUDA in Visual Studio 2008, April 29, 2009
• CUDA Tutorials for high performance computing
• An introduction to CUDA (French)
• NVidia CUDA Tutorial & Examples (from ISC2009)
• GPUBrasil.com, First website on GPGPU in Portuguese
• 3D cloth Simulation OptiTex.com, Implementation of CUDA in Cloth simulation
• DDJ CUDA series, First in the Doctor Dobb's Journal series teaching CUDA (over 13 articles by Rob Farber)