In order to move to charts we could not represent all of the data supplied by ATTO Disk Benchmark. We have chosen 4K and 8K read and writes as these file sizes represent more than 50% of windows operations. 128K seems to be the next file size up that closes in on the maximum speeds represented in this test. 1MB and 8MB cluster sizes top out the performance charts in nearly all SSD tests.
ATTO results were very interesting with nearly perfect performance in Dirty, Fill and Clean testing. The only exception is 4KB & 8KB file sizes dirty performance. 4K Read speeds increased 17 MBps for a 9% improvement overall. Write speeds Improved only 10 MBps for a total 5% increase in performance in dirty condition. Larger file sizes maintained perfect scores across the board with the only exception being 8MB Fill Write speeds that performed 20 MBps slower.
Maximum Read speeds pushed 510 MBps overall, with write speeds hitting 325 MBps. While the Read speeds in ATTO performed better than manufacturer specs, write speeds pulled shy of the 340 MBps by 15 MBps or 5%.
Crystal Diskmark uses 512K, 1024K and two 4K tests (single and 64 Queue depth). By default CDM uses a random data pattern and all tests were performed with that setting. An additional “clean” test was performed with a 0 fill data pattern that compresses easily and highlights the maximum capability of the drive.
Results in CDM were fairly similar to ATTO results. Once again Dirty performance had a small performance increase over clean testing in small file sizes, although not to the degree ATTO displayed.
512K Writes come much closer to hitting advertised ratings pushing 330 MBps. Although in fill testing, this dropped to 310. 4K IOPS crushes advertised rating of 60K IOPs though with a total speed of 265 MBps for a total of 66K IOPS in 32 Queue Depth testing.
AIDA64 Read Test suite compares read speeds of various file sizes. More importantly the linear read test acts much like older HDD testing software to show how performance scales as you move across a drive. For platter drives, the beginning will always score highest. With SSDs if any data is on the drive it will score the lowest.
The read tests once again display a similar phenomenon we saw in the previous two benchmarks. Dirty performance significantly performs better than clean and fill tests.
The Linear read tests fill results, as expected, performing much better on the end of the test than where data was already written. Clean testing however was even lower than Fill tests.
Read access shows the same flip flop of results, with dirty performance scoring a miniscule .04 access time, increasing to .1 for fill test results and much slower .14 for clean testing.
Write tests determines the capability of the drive to handle sequential as well as random writes. Although, there is no longer a recommendation not to use buffered writes on the drive, that is still no longer used for the sake of continuity.
Random write tests perform much closer to expectations of an SSD. Clean performance peaks at 312 MBps, dropping down to just over 250 MBps for fill testing. Dirty tests scored higher than Fill but still significantly lower than clean tests.
Linear writes and Access time both performed very similarly in all tests. Linear writes in the fill test were marginally lower but only a few percentage points of difference.
AIDA write tests also display nice charts for monitoring consistency of performance across the test. Clean test results are included below.
PCMark Vantage runs a variety of tests attempting to evaluate “real world” events. It covers performance from Windows Startup, to importing music or pictures, application loading and gaming performance. With the move to a more digital media being common Media Center and Video Editing are also included.
Both clean and fill tests scored fairly low in all tests. Contrary to previous benchmarks the Performance Pro does not stand up well to other drives in this test. Until one considers “dirty” performance. Scores were off the charts scoring 300-400 MBps numbers in nearly every test! This is aberrant and we will see if the same holds true when stuck on a HW controller next week.