PC GAMESS/Firefly-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

are these results obtained for the same driver using
different settings (i.e., NCQ enabled/disabled)?
Or, the drivers are different as well? In the latter cases,
it would be difficult to compare results as driver's quality
can vary seriously affecting I/O performance...

regards,
Alex


On Wed Nov 25 '09 11:27pm, Sergey wrote
---------------------------------------
>Hi all,

>as promised, I keep you informed abouth the experiment.

>It seems this option makes little to no at all difference in WinXP in 1 core per 1 disk setup BUT,

>for the 2 core per 1 disk design moderate sized jobs (I tested ~ 800 basis functions MP2 mainly) the result is pretty significant: especially with increased number of buffers to 8 or even 16 if memory allows.

>Most likely this is due to some ordering occurs in the tagged queues of SCSI controller (*remember, I do NOT yous SAS or real SCSI, but usual budget solutions e.g. SATA II drives, yet they are identified as SCSI after proer installation of AHCI / RAID drivers.

>