L’éditeur SiSoftware fait à nouveau évoluer son logiciel SANDRA qui est désormais disponible en version 2013.07.19.50 (SP4). Un numéro de version bien pompeux et bien compliquée tout ça pour dire que Sandra intègre de nouveaux benchmarks et apporte plusieurs correctifs.
-> Télécharger Sandra 2013 SP4 (version 2013.07.19.50)
Voilà le changelog complet (et en anglais) :
* New benchmarks: Multi-Buffer hashing using AVX2, AVX and SSE4
Tests the native CPU SIMD performance when running all 3 current hashing algorithms (SHA1, SHA2-256, SHA2-512) using each lane to operate on different memory buffers (9 tests in total):
– SSE (128-bit width): 4x SHA1, SHA2-256; 2x SHA2-512
– AVX (128-bit width): 4x buffers SHA1, SHA2-256; 2x buffers SHA2-512
– AVX2 (256-bit width): 8x buffers SHA1, SHA2-256; 4x buffers SHA2-512
If you were underwhelmed by Haswell – finally a reason to cheer: it is 2x as fast hashing blocks using AVX2 than AVX/SSE!
* Fixes: Multi-processor group systems with unfilled groups
Fixed affinity when running on systems with multiple processor groups (e.g. 1 package = 1 group) but with groups that are not full (e.g. less than 64 threads per group). Note that Sandra supports 1 group of 32 threads in 32-bit mode and 5 groups of 64 threads (320 total) in 64-bit mode.
* Updated benchmark: Memory latency
Now tests each core (and thus physical CPU) and aggregates the results; useful when running on NUMA systems or mutiple processor group systems. This will take X-times as long where X is the number of cores in the system (e.g. 6 cores – will take 6 times as long)
* Note: Running on Hypervisor / VM and AVX2, FMA3, AVX instruction sets
Latest Hyper-V and some VMs do not enable OS-XSAVE, i.e. does not save/restore eXtended CPU mode – thus it is not possible for applications to use AVX2, FMA3, AVX instructions that use the 256-bit SIMD registers. It is done in order to facilitate VM migration across systems.
Désolé, les commentaires sont fermés.