Benchmarks
I used four benchmark programs to test this drive:
- ATTO Disk Benchmark.
- AS SSD Benchmark.
- CrystalDiskMark x64.
- PCMark05’s XP Startup Test.
The PCMark05 test is most for the benchers out there as it is crucial if you’re competing in that benchmark. It is a decent indication of how fast a PC will boot as well, as it is a recorded XP boot sequence. Other than boot speed, it is largely irrelevant.
The test system is a solid one, it consists of the following core components:
- Intel Core i7 2600k CPU.
- Gigabyte P67a-UD4-B3 motherboard.
- MSI HD6970 GPU.
- Thermaltake Toughpower XT 1275w 80+ Platinum PSU.
- OCZ Vertex 3 MaxIOPS 240GB SSD used as an OS/boot drive for the empty tests.
I first ran each benchmark with the drive completely empty. I then installed Windows 7 Pro x64 and GPU drivers and booted off of it and ran each test (excluding PC05) again. In all photos the pink border is the empty drive result and the blue border is the result with Windows 7 on the drive.
PCMark05 XP Startup
For reference, the 240GB Vertex 3 MaxIOPS scored around a 240MB/s, the Agility 3 120GB scored around 300MB/s. Anything over 220MB/s is illegal for HWBot and is a sign that you’ve got a good drive for PC05 benching on your hands (there are other HDD tests you see, so you force the drive speed down for XP Startup and let it fly for the others).
400MB/s! That is incredible for a single drive! You can tell when you’re booting off it, too.
I know of no other single drive that can hit 400MB/s.
ATTO Disk Benchmark
ATTO is arguably the best SSD benchmark as it not only tests a wide variety of file sizes but also tests by sending multiple read or write requests at once, much like when your computer is booting or loading a large program. The default setting is a queue depth of 4. I’ll explain later on in this section why that is important.
With small file sizes the performance is in the same general area as the Vertex 3 and Agility 3, but as the file sizes get larger the RevoDrive3 really starts to shine. It eventually caps out at 1008MB/s read and 873MB/s write. Incredible numbers for a single drive. You can definitely see the multiple SF-2281 controllers in action here.
As usual the results with windows 7 installed on the drive are a bit slower, less so on this drive than most others however.
The peak (and large file size) numbers eat every SSD Funkykit has tested alive. The best drives so far have managed 550MB/s in read/write, nowhere close to the 850-1027MB/s the RevoDrive3 puts out. With very small file sizes it is a closer match, but as the file sizes get larger the RevoDrive3 pulls away and never looks back.
You can see the optimazation for a workstation environment here.
AS SSD Benchmark
AS SSD does not use a queue for reads and writes, this can have a dramatic effect on the perceived performance.
The lack of a queue clearly hurt the RevoDrive. These results are solid, but not dominating like the ATTO results. In AS SSD the RevoDrive3 performed better overall with windows 7 installed on it!
CrystalDiskMark
No queue here either, can you tell? Again the results are good, but not spectacular. SATA6 Sandforce based drives are faster in this benchmark. The design of the drive doesn’t deal especially well with single small files, it likes either a big batch of small files or larger files. On the plus side, that’s exactly what many workstation uses feed the drive.
ATTO Disk Benchmark Revisited
Having seen the varied results of the benchmarks I decided to test the queue depth setting in ATTO. With a queue depth of 1 performance dropped sharply on the RevoDrive3 and also dropped on the Vertex 3 MaxIOPS. When I set the queue depth to 10 the story was different! While the conventional SSDs I tested had similar or lower performance to a depth of 4, the RevoDrive3 laid down the following run:
Wow. Just wow. The peak results are quite similar to the default settings (though marginally higher peaks and the peak occurred sooner), but the low end and midrange is much higher. Given that the RevoDrive3 was already brutally fast, the fact that a deeper queue doubled its performance is truly spectacular!
My theory on what is going on here is thusly: The VCA features of the controller mark is as something designed for enterprise use. Stuff like massive databases, storage servers, workstations for disk heavy design use, that kind of thing. All of those uses will almost certainly give you very heavy disk usage and the heavier the disk usage the faster the RevoDrive3 goes! Internally it has a command/data cache as well as an ordering mechanism. Most SSDs deliver data in the same order it is asked for, much like older CPUs performed calculations in the order asked for. I think that the RevoDrive3’s controller has enough brainpower to sort through the incoming requests and change the order they are in so it can get more done at once. A flash chip can only deliver one datastream, so if the commands are re-ordered so that requests for data that don’t conflict with each other are run in parallel the whole operation will go far faster. This is what modern CPUs do with calculations as well.
The down side is that the system requires time to work, every access takes a little bit longer to complete. If there are enough requests coming in this delay is more than taken care of by the fact that multiple requests are being served at once. If only one request at a time is coming in however that delay hurts.
From a practical standpoint the end result is that the operations that take the most time to complete will go far faster, while the operations that are already extremely brief will take slightly longer. Personally I think that sounds fine to me, if my programs load a few seconds faster I am happy to deal with saving a picture taking a few thousands of a second longer.
The numbers in the ATTO test above are (barely, sometimes) achievable by running a pair of very fast SSDs in RAID0. Unfortunately that means no TRIM support, and the chances of loosing your data are exactly doubled compared to using a single drive. Plus the RAID controller introduces a similar delay every access cycle.
