ATTO Disk Benchmark uses a queue depth of 4 by default so results will be better than a single threaded. The Vertex 4 manages to peak out at about 540 MBps read speeds and hits about 425 MBps writes. Dirty performance was poor overall except in the small file tests which is most important for a Windows Operating System. Here Dirty performance was equal to or better than clean and fill performance. This shows that despite moving as fast as possible Garbage collection works well on this drive.
Crystal Diskmark is a great application for showcasing a drives 4K IOPs by using up to 32 threads to maximize 4K performance. The limit for every controller currently made is 16 or 32 thread capable so any more would decrease performance (a la AS-SSD). By default it uses a random pattern, all drives benefit from using an easier 0 fill data test so results for that test pattern are included as well.
Once again the grouping is really close. Dirty performance overall still takes a hit. Real world performance will not be so severely impacted. This is a worst case situation remember (anyone with an SSD should know not to quick format and bench it). We do it to show case a worst possible situation. It is good because the vertex 4 drops from 55K IOPs 4K writes to just over 41K. Both respectable numbers one might see on a SandForce drive. Both numbers are far above other 128 GB drives based on other controllers.
AIDA64 is return of the AIDA32 to Tomas Miklos. sold for seems like forever to Lavasys As Everest System software. Finalwire is the new license holder and they are rocking it, constant updates making it better all the time. Really looking forward to version 3.
The read tests show how a drive performs in fill states exceptionally well. Specifically the difference between beginning middle and End. One can see how NAND performance drops in Fill tests in the beginning and then explode back to clean performance in the end tests. What do we see? AMD performs identically from start to end. This is as expected. What was not was that LSI performed no differently in any of the tests (even clean tests). Clean LSI tests max out at 500 MBps, so this might show an issue with the secure erase feature OCZ toolbox provides…But it doesn’t because middle and End performance returns to 500 MBps. *sigh Benchmarks suck sometimes.
Like the read tests AIDA64 provides, the write tests offer a different take on performance than other apps. Once again the charts needs to change to see the truth behind the vertex 4. This is not some marketing attempt to obfuscate the truth. Conversely it is an attempt to spread some truth on a popular misconception that Vertex 4s have poor write performance after a drive is half filled. Well they do, but that is true of all drives not just the Vertex 4. The difference is in a sustained write the firmware Indilinx/OCZ created optimizes for performance and it does that very well. the trade-off is if you are trying to fill the entire drive in one shot, after you hit 50% it will drop in performance. It actually drops below 100 MBps when it hits storage mode.
There is a reason for this. OCZ says they recognized a growing trend among SSD users that do not fully utilize an SSD’s space. Well, you are right, there is a reason for that, because everyone has been told that filling an SSD is bad. OCZ might even be one of the culprits (keep it under 80%!)
Anywho, what they did do is leverage that open space and an SSD’s ability to handle a fragmented drive. Why would they tell how? I’m sure Marvell will have something identical soon unless it is in the contract. It performs really well all the time except the dirty test once again. The GC tests is not as good as a standard Marvell. It performs better in every other way though so, 50/50?
.11 latency avg? It starts at .03 and goes up pretty quickly. Still what most users see is .03 range.
Overall the performance was pretty terrible. But it boils down to a few things. Overall yes a sub 60K score is terrible. However, the thing not to accept is that except for some outliers dirty and fill performance were just as terrible as clean performance. Media center loves clean performance everything else stayed grouped nicely.