L’éditeur SiSoftware a publié une nouvelle version de son célèbre logiciel d’informations système et de benchmark, j’ai nommé : Sandra ! Celui-ci est maintenant disponible en version 2013.19.05.41 (SP3). Un numéro à rallonge mais qui n’apporte malheureusement que plusieurs changements assez mineurs puisqu’il s’agit exclusivement d’ajouter un indice financier au matériel analysé (CPU, APU ou GPU). Pour cela, Sandra s’appuie sur plusieurs comparateurs de prix et plusieurs modèles de référence pour délivrer un indice mesuré en OPT/s (options par seconde).
-> Télécharger Sandra 2013 SP3 (version 2013.05.19.41)
Le changelog complet (et en anglais) est disponible ci-dessous :
* New benchmark: GP (GPU/APU/CPU) Financial Analysis
Tests the performance of the three (3) standard financial models (options pricing); different models are used in organised option trading to determine their future worth (e.g. price at expiration date / « strike » price) and thus their worth:
– Binomial Option Pricing (European Model)
– Black-Scholes Option Pricing (European Model)
– Monte Carlo Option Pricing (European Model)
Measurement unit is OPT/s (options/second), i.e. how many price options can be calculated in 1 second.
– Float/FP32 & Float/FP64 versions (6 in total)
Can compare the drop in performance if higher precision is required (FP64 vs. FP32) to compute the results to higher accuracy. Now you can see why professionals pay big money on Tesla / FirePro / Phi cards that provide higher FP64/double performance!
Using the corresponding CPU Financial Analysis benchmark you can determine when/whether a GP(GPU) can surpass in performance the latest multi-core/treaded CPUs.
– CUDA, OpenCL versions (12 in total)
Can compare CUDA against OpenCL on nVidia as well as against vendors. See whether a Tesla card in CUDA can take on a FirePro card in OpenCL.
Note: DirectX 11 ComputeShader support coming soon!
– Up to 8 GP(GPUs) supported, including CPU+GPU (aka APU) in OpenCL
* New benchmark: CPU Financial Analysis
The corresponding native CPU benchmark to allow you to directly compare native GP(GPU) performance to CPU performance.
– Float/FP32 & Float/FP64 versions (2 in total)
– Up to 320-threads (5 groups of 64-threads) supported in 64-bit mode
* Updated benchmark: GP (GPU/APU/CPU) Crypto: SHA2-512 hashing
SHA2-512 is the 64-bit version of the SHA2-256 hashing algorithm and thus performs better on 64-bit arch while providing similar strength against attacks. (While SHA3 has been selected, it has not been standardised). It is useful to compare its performance against SHA2-256 and SHA1 especially for GP(GPU)s that don’t necessarily support native 64-bit operations in hardware.
* Updated benchmark: CPU Crypto: SHA2-512 hashing
The corresponding native CPU benchmark to allow you to directly compare native GP(GPU) performance to CPU performance.
– AVX2, AVX, SSE4 vectorised (x64) and standard CPU versions
Désolé, les commentaires sont fermés.