Jump to content


Photo

Discrepancies in Samsung 840 Pro benchmark numbers


  • You cannot start a new topic
  • Please log in to reply
1 reply to this topic

#1 continuation

continuation

    Member

  • Member
  • 10 posts

Posted 26 April 2013 - 06:56 PM

I was looking at Samsung 840 Pro's performance after 4K 100% write preconditioning.

In the original benchmark (http://www.storagere..._840_pro_review), the numbers for Samsung 840 Pro are:
100% read: 32,588 IOPS
100% write: 7,926 IOPS

In this benchmark (http://www.storagere...rise_ssd_review), the numbers for Samsung 840 Pro became:
100% read: 71,622 IOPS
100% write: 7,973 IOPS

So the 100% write numbers are roughly the same, but the 100% read numbers are *very* different (went from ~33K to ~72K). Why? Which is the "real" number?


The same problem occurs for Intel DC S3700. After 4K 100% write preconditioning:

In the original benchmark (http://www.storagere...rise_ssd_review), the numbers for DC S3700 are:
100% read: 33,830 IOPS
100% write: 33,016 IOPS

In this benchmark (http://www.storagere...rise_ssd_review), the numbers for DC S3700 became:
100% read: 64,345 IOPS
100% write: 33,013 IOPS

Once again there's a huge difference in the 100% read numbers between the 2 benchmarks.

Edited by continuation, 26 April 2013 - 08:14 PM.

#2 Kevin OBrien

Kevin OBrien

    StorageReview Editor

  • Admin
  • 1,393 posts

Posted 27 April 2013 - 12:31 AM

Both are "real" numbers, the primary difference came in when we changed to our newer testing platform. Initially our linux test platform was geared towards a "plug n play" approach where we left the OS configuration bone stock, with queue schedulers set to the default cfg. Later on in our newer rd630 we geared it to be optimized per storage type, with flash being set to "noop". The later optimizations allowed for higher SATA performance, which is the platform all our latest reviews are being tested on.



0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users