SiSoftware met en ligne Sandra 2014.08.20.42

par Sebastien le 17 Août 2014 à 08:55

L’éditeur SiSoftware a mis en ligne ce matin la version 2014.08.20.42 de son programme d’informations et de tests matériels SANDRA. Ce nouvel opus apporte surtout des correctifs et des améliorations par rapport à la version précédente (voir le changelog ci-dessous en anglais) ainsi que le support de CUDA 6.5. A noter par ailleurs que Windows XP et Windows Server 2003 ne sont plus supportés par le logiciel.

-> Télécharger Sandra 2014 SP3 (version 2014.08.20.42)

Changelog officiel de Sandra 2014 SP3 (en anglais) :

* Further Platform update: Windows XP, Server 2003/R2 no longer supported
In order to move forward we have to change the environment, more specifically to VS 2013; we took this « opportunity » to remove support for XP and thus Server 2003 – thus supporting Kernel 6.0 and later only.

* CUDA SDK 6.5
Now that recent WHQL video drivers support it, and due to VS2013, we changed to SDK 6.5 – which includes native Maxwell (SM50) support. Unfortunately SM10 support has been removed and SM11 support is deprecated.

* Internal Scheduler
– Fix for systems with multiple CPU groups (e.g. more than 64 threads) where affinity would be incorrect.
– Fix for benchmark failure when NUMA node would be incorrect due to above.
– Fix for benchmark lock-up when affinity would be incorrect due to above.

* CPU benchmark: Dhrystone
– Crash fix for AVX2 x64 code-path on some systems.

* CPU benchmark: Multi-Media
– Added Quad-Int (128-bit) implemented as 2x 64-bit integers or 4x 32-bit integers depending on platform.
– Added Double-Int (64-bit) SIMD AVX2, AVX, SSE4 implementation.

* CPU benchmark: Scientific
– Improved FMA FGEMM/DGEMM and FFFT/DFFT implementation.

* GPGPU benchmark: GP Processing
– Added Quad-Float (128-bit) implemented as 4x single-float (4x 32-bit) for GPUs that do not support native double-float (aka 2x 64-bit) in CUDA, OpenCL.

* Video Shader Compute
– Added Quad-Float (128-bit) implemented as 4x single-float (4x 32-bit) for GPUs that do not support native double-float (aka 2x 64-bit) in DirectX 11, 10 and OpenGL 3.x+
– Reduced OpenGL shader requirements by using extensions; lower base GL/GLSL are now supported if the extensions are supported.