Dave Laban

SSD Performance Gain - Was I Expecting Too Much?

Discussion created by Dave Laban on Apr 26, 2012
Latest reply on Sep 12, 2013 by Ryan Dowling

We have two identical workstations in the office that are used exclusively for analysis work.  Specs are HP xw8600, 2x Intel Xeon X5492 (3.40GHz) processors, 32Gb of RAM, NVidia Quadro FX5800, Win7 x64 SP1.


The Windows Experience Index showed 7.8 for Processor and RAM, and 7.3 for graphics.  Data transfer rate only managed a 6.1.  The SolidWorks performance benchmark resulted as near enough the same on each machine, and I set up some benchmark analyses (FEA and CFD) which also came out as close enough to the same times.


To try and improve performance, I asked if IT could set up a SSD as the primary (only) disk in one of the workstations, to see if we could get another step in performance to reduce analysis time – all that I had read on this forum and in general indicated this should be possible.


So, one of the workstations now has a Crucial 256Gb M4 SSD (http://www.crucial.com/uk/store/partspecs.aspx?IMODULE=CT256M4SSD2) and the other has a Seagate 450Gb 15k HDD (http://www.pcbuyit.co.uk/seagate-st3450856ss-450gb-30gbps-serial-scsi-hard-drive-0b23461-brand-p-1504.html).  The box with the original drive was re-imaged at the same time as the SSD box was setup to ensure consistency with Windows, and SW 2011 SP5 was installed using the same setup.


First, the Windows Experience Index was re-ran – data transfer rate is up to 7.8 from 6.1.  The SW Performance Benchmark I/O score improved from 95.5 to 78.1s – a 20% improvement was looking promising for my run times!  However, when it came to running some representative analysis work, things weren’t so good – CFD mesh time was reduced by around 10%, but solve time only came down by 1-3%.  With the FEA, with a single part mesh time came down by about 10%, while solve time ranged from 1% to 15% improvement.  However, with assemblies mesh time sometimes increased by as much as 35%, with run time sometimes up by 200% or more!


Running the NAFMES benchmarks also yielded unpredictable performance results, with some solutions 10% faster, some solutions 30% slower and others with identical run times.  Solutions using Abaqus showed virtually no difference in performance (typically 1% worse).


Finally, I downloaded the Punch Holder model that is recommended on here for performance testing of SW and found that the rebuild time on the SSD PC was 10% slower than the HDD PC.


So, does anyone have any idea what might be going on?  Have we ended up with a duff SSD?  Or is it somehow the wrong sort of SSD?  Or, were my expectations too high and SSDs aren’t the sort of ‘magic bullet’ for decreasing run times I thought they were?