CUDA Toolkit and SDK 2.3 released


Because of the new support for cross-compilation, the library locations on Linux have changed. 32-bit libraries are located by default at /usr/local/cuda/lib and 64-bit libraries are located by default at /usr/local/cuda/lib64. This is a change from 2.2 and will necessitate changing /etc/ or LD_LIBRARY_PATH.

The CUDA Toolkit and SDK v2.3 are now released and available to all developers.

A brief overview of features–there are a lot:

    The CUFFT Library now supports double-precision transforms and includes significant performance improvements for single-precision transforms as well. See the CUDA Toolkit release notes for details.

    The CUDA-GDB hardware debugger and CUDA Visual Profiler are now included in the CUDA Toolkit installer, and the CUDA-GDB debugger is now available for all supported Linux distros. (see below)

    Each GPU in an SLI group is now enumerated individually, so compute applications can now take advantage of multi-GPU performance even when SLI is enabled for graphics.

    The 64-bit versions of the CUDA Toolkit now support compiling 32-bit applications. Please note that the installation location of the libraries has changed, so developers on 64-bit Linux must update their LD_LIBRARY_PATH to contain either /usr/local/cuda/lib or /usr/local/cuda/lib64.

    New support for fp16 <-> fp32 conversion intrinsics allows storage of data in fp16 format with computation in fp32. Use of fp16 format is ideal for applications that require higher numerical range than 16-bit integer but less precision than fp32 and reduces memory space and bandwidth consumption.

    The CUDA SDK has been updated to include:

      A new pitchLinearTexure code sample that shows how to efficiently texture from pitch linear memory.

      A new PTXJIT code sample illustrating how to use cuModuleLoadDataEx() to load PTX source from memory instead of loading a file.

      Two new code samples for Windows, showing how to use the NVCUVID library to decode MPEG-2, VC-1, and H.264 content and pass frames to OpenGL or Direct3D for display.

      Updated code samples showing how to properly align CUDA kernel function parameters so the same code works on both x32 and x64 systems.

    The Visual Profiler includes several enhancements:

      All memory transfer API calls are now reported

      Support for profiling multiple contexts per GPU

      Synchronized clocks for requested start time on the CPU and start/end times on the GPU for all kernel launches and memory transfers

      Global memory load and store efficiency metrics for GPUs with compute capability 1.2 and higher

    The CUDA Driver for MacOS is now packaged separately from the CUDA Toolkit.

    Support for major Linux distros, MacOS X, and Windows:

      MacOS X 10.5.6 and later (32-bit)

      Windows XP/Vista/7 with Visual Studio 8 (VC2005 SP1) and 9 (VC2008)

      Fedora 10, RHEL 4.7 & 5.3, SLED 10.2 & 11.0, OpenSUSE 11.1, and Ubuntu 8.10 & 9.04

Notes for MacOS developers

    The cudadriver_2.3.1_macos.pkg driver is for use with Quadro FX 4800 and GeForce GTX 285.

    The cudadriver_2.3.1_macos.pkg driver may also be used with any NVIDIA GPU on SnowLeopard.

    Use the cudadriver_2.3.0_macos.pkg driver for MacOS X 10.5.6 and later (pre-SnowLeoard) with all other GPUs.

Download Links

Mac developers: use the packages in this post.

Getting Started on Linux

Getting Started on OS X

Getting Started on Windows

190.38 for WinXP32

190.38 for WinXP64

190.38 for Vista32

190.38 for Vista64

190.18 for Linux 32

190.18 for Linux 64

CUDA Visual Profiler 1.3 for OS X

CUDA Toolkit 2.3 for Fedora 10 - 32-bit

CUDA Toolkit 2.3 for RHEL 4.7 - 32-bit

CUDA Toolkit 2.3 for RHEL 5.3 - 32-bit

CUDA Toolkit 2.3 for SLED 10.2 - 32-bit

CUDA Toolkit 2.3 for SLED 11.0 - 32-bit

CUDA Toolkit 2.3 for SuSE 11.1 - 32-bit

CUDA Toolkit 2.3 for Ubuntu 8.10 - 32-bit

CUDA Toolkit 2.3 for Ubuntu 9.04 - 32-bit

CUDA Toolkit 2.3 for Fedora 10 - 64-bit

CUDA Toolkit 2.3 for RHEL 4.7 - 64-bit

CUDA Toolkit 2.3 for RHEL 5.3 - 64-bit

CUDA Toolkit 2.3 for SLED 10.2 - 64-bit

CUDA Toolkit 2.3 for SLED 11.0 - 64-bit

CUDA Toolkit 2.3 for SuSE 11.1 - 64-bit

CUDA Toolkit 2.3 for Ubuntu 8.10 - 64-bit

CUDA Toolkit 2.3 for Ubuntu 9.04 - 64-bit

CUDA Toolkit 2.3 for Windows - 32-bit

CUDA Toolkit 2.3 for Windows - 64-bit


CUDA GDB 2.3 User Manual

CUDA C 2.3 Reference Manual

CUDA Visual Profiler 2.3 Readme

CUDA Visual Profiler for OS X 1.3

CUDA Toolkit 2.3 EULA

CUDA Toolkit 2.3 Release Notes for Linux

CUDA Toolkit 2.3 Release Notes for OS X

CUDA Toolkit 2.3 Release Notes for Windows

CUDA C Programming Guide 2.3

CUDA C Best Practices Guide 2.3

CUDA C Online Documentation

CUDA SDK 2.3 for Linux

CUDA SDK 2.3 for Win32

CUDA SDK 2.3 for Win64


CUDA SDK 2.3 Release Notes for Linux

CUDA SDK 2.3 Release Notes for OS X

CUDA SDK 2.3 Release Notes for Windows

This space reserved for other important things. Like unsupported/undocumented stream support that should work fine in CUFFT 2.3 :)

Notebook drivers:

XP32 190.38
XP64 190.38
Vista32 190.38
Vista64 190.38

Cool. I’m downloading it and installing now.

I’m looking forward to trying 2.3 out.

I have a previous version installed in Linux. Do I have to uninstall that before installing 2.3? I have checked the installation instructions in the Release Notes, but they do not mention this.


# define NumThreads infinite

for(int i=0;i<NumThreads;i++)

PRINTF(\n"Thank u VERY MUCH for this CUDA  2.3 and supporting 9.04 Linux ((64bit) " );


Installing it now…

Thank you, you just have ruined the rest of this summer for me :)
It’s ok, it’s raining all the time.

On my system (openSuSE 11.1) ./ hangs indefinitely, when trying to run it. I use tcsh as my login shell. The same happened before with 2.3 beta. Running it with --noexec --keep and then installing on foot works ok. Has anyone had a similar experience?

All installed. All compiled. All working very fast. I have to check if it is faster than 2.2
It looks like the compilation is faster.

Question: How can I get 135 GFLOPS with 12 processors @1.35Ghz sm1.1 running nbody simulation?

Linux x86_64 slackware.

There’s some weird thing with SuSE where it doesn’t flush output where everything else on the planet flushes. Hit enter, and it will then magically show output and work. Don’t ask me, but I’ve seen it on every SuSE machine I’ve ever used.

The link to the 2.3 Reference Manual above goes to a PDF with only 3 pages. Maybe it’s a bad copy? Or my PDF reader is borked again?

you’re right, it’s a bad copy. I’ll get that fixed ASAP.

not necessarily a good idea, I’ve seen it defaulting to ~/NVIDIA_CUDA_SDK after just hitting enter. Not really a show-stopper, I just let the SDK install whereever it wants to and move the whole directory tree afterwards.

Curious if the SDK now actually compiles on suse11.1 32bit. Previous releases all died on some of the driver API examples. Not a show-stopper either (which is why I never complained) because I just build cutil and then every example that I really need (deviceQuery mostly to check if new machines installed properly)

Great news!! Thanks! :thumbup:

Looking forward to try out but unfortunatly it seems that the links to the 64bits windows drivers are broken…



checking on it…

Does this only happen with tcsh as your shell?

So sorry, I installed 32 bit version instead of 64 bit :)

now is the same :D

I can’t find window7 32/64bit driver.

How can I find it?

Win7 and Vista share the same driver.

The Mac OS X version appears to still not support the GeForce 9400M in the new Macbooks (with either of the two driver versions). Is this correct?