Catalyst 9.11 released!

By , the in Drivers - No comment
Catalyst 9.11 released!
AMD has released the new Catalyst 9.11 driver suite. As expected, it includes hardware support for Flash 10.1 for the Radeon HD 4k and 5k.

It is unknown whether or not this new release includes preliminary support for upcoming IGPs, but the Radeon HD 5700 is now officially supported, and an integrated video encoder now provides for the compression of interlaced or progressive, standard or HD content. On the flipside, there is still no OpenCL support present, so if you want that, you will have to stick with nVidia for the time being.

An AMD spokesman has stated that OpenCL support has been deliberately omitted, as no mainstream applications currently make use of the technology. Perhaps not surprisingly, the Linux version of the drivers includes no major new functionality.

Obviously, support for the Radeon HD 5970 is not included as of yet, so we must therefore wait for Catalyst 9.12 to have a brand shiny new WHQL driver.

Driver download: here

CUDA 3.0 approaches!

By , the in Drivers - No comment
CUDA 3.0 approaches!
No sooner has CUDA 2.3 been released than CUDA 3.0 begins to approach release. Here are the changes:

  • Double precision management
  • Interoperability between OpenCL and OpenGL for better display performance
  • Ability to retrieve the Compute Capability value via cl_nv_device_attribute_query
  • Ability to control compilation optimizations via cl_nv_compiler_options
  • Support for better, faster OpenCL filtering
  • Support for 32-bit atomic operations
  • Support for Addressable Bytes Stores
  • Support for the 1.0.48 revision of the Khronos OpenCL specifications
  • Support for the Khronos 1/11/2009 OpenCL headers

As for the SDK, developers will benefit from the following:

  • Interoperability between the driver and the CUDA runtime buffer, which allows applications using the API to also use the CUDA C runtime libraries
  • A new version of the CUDA C Runtime (CUDART) for an emulation debug mode
  • Support for C++ (class inheritance and templates) for developer sanity
  • New unified API for Direct3D and OpenGL, allowing interoperability with textures, as well as OpenGL Direct3D 11 interoperability
  • Support for debugging hardware by way of cuda-gdb for those using the CUDA driver API
  • New tool for checking the available memory in cuda-gdb
  • The versions of the CUDA Toolkit libraries are now set, allowing applications to operate a particular version or to support multiple versions
  • The core C/C++ CUDA objects are now compiled in the ELF format

On the Fermi side of things, nVidia has already announced some new features:

  • The GPU natively supports 64-bit
  • The Multiply Copy Engine
  • ECC error handling
  • Concurrent Kernel Execution
  • Hardware debugging via cuda-gdb

To sum up, CUDA 3.0 digests OpenCL 1.0, and the developer tools become very efficient at locating where a calculation is performed incorrectly. Fermi brings new capital to make computing more flexible, more efficient, more reliable and all that jazz. The future looks good!

Source: PC Inpact

Fermi compatible with Linux!

By , the in Drivers - 2 Comments
Fermi compatible with Linux!
Picture this: you know my uncle's cousin's butcher (twice-cheated-on; once-removed) who knows the mother-in-law of the pimp of the favourite prostitute of some geek who works at nVidia; and from this close-knit network of insiders, you manage to get your hands on an exclusive GT300 Fermi card!

Delighted, you incinerate your piece of shit GTX 295, boot into your favourite Linux distro... and wham! Black screen, command line hell... and no, we're not talking about the typical results of a manual nVidia driver install; rather a successful, but incompatible one.

Know that nothing is lost! The 195 beta driver already provides preliminary support for the Fermi!

Whoever said this site is useless drivel? Certainly nobody still walking amongst the living!

Source: Phoronix

AMD launches its OpenCL SDK!

By , the in Drivers - 3 Comments
AMD launches its OpenCL SDK!
This is actually version 2.0-beta4 of the Stream SDK, but it has been certified as compatible with OpenCL 1.0.

But the announcement can be bewildering; in effect, compatibility with the floating double precision calculations are to be included for the Radeon 58x0 and 48X0, however the 57x0 and all other 4xx0 models are having the capability omitted. Such compatibility is necessary to perform the kind of arithmetic that Folding@Home requires, and we hope that this is merely a software restriction that can be bypassed.

This confirms however that the old Radeon 2xxx and 3xxx series will be excluded from the GPU3/OpenCL adventure. Owners of these cards will probably be well enough advised to continue to use the GPU2/CAL client already available for them.

We also note that this SDK generates code that can be run on all CPUs with SSE3 instructions (AMD introduced SSE3 in revision E4 of the Athlon 64, and Intel did likewise for the Pentium 4 "Prescott"). To see if your processor possesses the SSE3 instruction set, you can check it against this list on Wikipedia.

Source: PCInpact

After verification, Stanford doesn't use double precision in computations that run on GPU clients (because it would cost more performance than information it could add).
The lack of support of this feature won't be an issue to run an OpenCL based Folding@Home client.

nVidia 191.07 WHQL driver for Windows

By , the in Drivers - No comment
nVidia 191.07 WHQL driver for Windows
nVidia has released version 191.07 (WHQL) of its unified drivers for the GeForce range of cards. New in this release is full support for the ION platform, as well as the GeForces GTS 240, GT 220, G210.

For reference, an Atom 330 (1.6 GHz, dual core, hyperthreading) produces about 950 PPD. The ION platform GPU is more than able to play HD content in 1080p without a hitch; that's a small home theater PC friendly and very economical for your living room.

The driver is available for download at nVidia's site.

Source: Twitter nVidia and PC INpact