Test Setup and Software
For testing Solid State drives we use the LSI MegaRAID 9265-8i RAID controller card. This provides seamless interaction between single and multidrive testing and allows platform agnostic results to be given. Additional testing was performed with the onboard SATA 6 Gbps controller of the SB950.
- ASUS Sabertooth 990FX Motherboard
- AMD FX-4100 Quad Core Bulldozer
- 8GB Corsair EVO 1866 9-10-9 DDR3
- EVGA 580GTX Superclock
- Corsair TX-750 PSU
- Western Digital 1TB Greenpower drive for OS
- Cooler Master 212+ Heatsink
- Custom Built Bench Station
For gauging performance the following software is used
- ATTO Disk Benchmark
- Crystal DiskMark X64
- PCMark Vantage
- AIDA64 Storage Read and Write Tests
The LSI card is set up as single drive RAID0, with disabled read ahead and Direct IO write methods to bypass cache performance.
Test Results
ATTO Disk Benchmark
ATTO uses compressible data and is a good indication of best possible results. It defaults to a 4 thread depth, so small file size will score better than other tests but not as well as the full AHCI 32 or 64 thread test other programs use. Still ATTO is more indicative of real world possible performance with real world max threading. We use the maximum 2 GB test length to provide more consistent results and alleviate any possible LSI caching issues.
ATTO testing showed excellent performance on the LSI card. The AMD controller struggled to keep up with the RAID card clean tests with a notable exception. 8KB writes scored a whopping 412MBps. 4K and 8K both saw a hit in dirty and fiull tests but larger file sizes were almost unaffected. Maxing out at 550 MBps read speed and about 513MBps writes. 4K write performance topped out at 77K IOPS
Crystal Diskmark
Crystal Diskmark is another popular SSD benching tool as it provides a nice neat output and uses a Queue Depth of 32 for 4K read/writes. As this is the limit on SSD controllers reviewed so far this test is perfect for finding the absolute maximum 4K performance. By default CDM uses a random test pattern, which is tested separately. In order to provide better comparisons with ATTO 0 fill test data is used for all other tests. A 4GB test data length is used in testing.
Crystal diskmark tests are a little different, the 0-fill data test is used throughout and because of this we also include another entry, random data test. This is usually the bane of SandForce driven drives, in truth it effects all SSDs just not nearly to the same degree. Overall random data punished this drive even worse then normal. Topping out at about 210 MBps when the drive usually performs in the 450 MBps range in these tests. 4K performance is affected as well but not nearly to he same degree. 25% differnce in 4K 32QD performance between the two. The AMD controller lags behind the LSI card in the larger file sizes but stays right with the LSI in 4K performance. Excatly the opposite of ATTO results.
AIDA64 Storage Tests
AIDA64 has been around for many years under different names and has been updated to keep up with the times. Starting off as AIDA32 the software was licensed to Lavalys for a number of years before being sold to FinalWire, updated and rebranded as AIDA64. It also performs CPU and memory benchmarks as well as storage. A read storage test suite is run as well as three of the four write tests.
Read tests were good overall for the SP900. Peaking at 540MBps, while normally we only include linear end results in charts as they are typically higher on drives showing all three was deemed important. If you look at fill testing under linear end, you see the same result whether clean or fill. Dirty of course drops don but not nearly as much as beginning or middle (this is because the drive was fill tested, quick formatted and run dirty). You will see a lot of reviews mentioning that TRIM does not function properly under Firmware 5, that maybe true, the SP900 reviewed has no firmware updates available and may have come after a fix from LSI. The dirty test linear begin scores much higher than fill or dirty tests throughout, which leads one to believe that TRIM or Garbage Collection is working. It did not have time to restore the drive completely (fast testing is done in order to prevent TRIM/GC from kicking in) but did result in improved performance.
In write testing the AMD controller rules the roost. This is most likely due to the vastly reduced latency. Rocking in at about 40% of the RAID cards latency.
PCMark Vantage
PCMark Vantage is a great whole system benchmark TechREACTION uses for motherboard reviews. It is also popular among enthusiasts looking to beef up their total score to compare against others with similar components. The Storage test focus on “real world usage” for the modern digital media manipulator, with highlights on digital photo, video and music manipulations as well as OS and application loading.
In the last few reviews with the LSI card, PCMark Vantage performance was less than stellar. Especially when RAID arrays were involved. Adding a second drive resulted in approximately 2% perofrmance increase adding 7 more drives only lead to about a 5% increase in scores. This is one of the quirks of running a RAID card combined with PCMark Vantage testing. With a single drive the LSI card scored a 61,690, very decent score for a sub $1/GB drive. Dirty testing took a massive hit though and knocked the score down to just under 30K. On an AMD controller however the drive scored a very respectful 69,314. This is only 4K less than the MAXIOPs drive did on an Intel PCH. Very impressive, and points out, somewhat inconsistently, that with the right controller none of the synthetic tests can compare to good old fashion moving data around tests.
