Samsung’s Marketing guys probably pulled a marketing stunt of the year, with a Youtube video which probably cost them less than any of those fancy ads in papers. Guys’n'girls took 24 Samsung 220/200MB Read/Write SSD drives, put’em on an RAID controller and tested them using a Skulltrail system. While achieved speed of 2GB/s was awesome, it also struck me as a pretty low figure for a 24 drive setup. Each drive achieved only 85MB/s, far cry from what those drives are capable of.
Given that a RAID0 array with five Intel or OCZ SSDs gives out 1GB/s, why did Samsung need 19 more drives to only double the speed? 24 drives should achieve 5.2 GB/s read and 4.8GB/s write speed. The problem lies within our current PC architecture. The authors used two PCI Express x4 controllers that cannot give out more bandwidth than 2GB/s read and 2GB/s write – thus, from eight drives onwards, the speed of each individual drive only decreased, from 220MB/s to a lowly 85MB/s.
Secondly, using Skulltrail for seeking maximum bandwidth was never a good idea. Intel Core i7 or AMD’s Phenom II would do much better, without the limitations of archaic FSB architecture.
Guys were so excited that they even made a typo at the number of drives... 
If Samsung had access to PCIe Gen2 x16 controller (not certain that anyone makes it), then those 24 drives would run at its peak speed, 5.2/4.8 GB/s in R/W operations. This is not to say that the video wasn’t awesome, but bear in mind that those drives could perform way better once that the RAID controller industry catches up. On the other note, I managed to max out Intel’s ICH10R controller – it took six Raptors (no SSDs for me), but the top bandwidth figure out was 450 MB/s, roughly identical to limit which Nova found in his 4-drive SSD test a while ago. If we would put these Samsung hard drives onto ICH10R, you would only need two drives to max it out.
One thing is certain: for SSD technology, PCI Express Gen3 and SATA 3.0 can’t come soon enough.
P.S. Video was awesome guys, now, do the same thing on i7 just for kicks – see the difference in bandwidth between 5.5GB/s Skulltrail and 38.4 GB/s on i7 platform (DDR3-1600).
UPDATE – March 10, 2009 15:10 CET - I have received reports that the story is not appearing on the homepage – seems like an odd thing to happen, but this isn’t the first story that experienced “disappearing act” on this blog. One reason more for going with Custom CMS for upcoming site, BSN*. I apologise for any inconvienience this might have caused, but have no control over it.
Due to a filming delay, this was shot before i7 was out. If I was to do it again, it’d be on i7.
As for the quantity typo, we had 26 but kept 2 on standby for safety. No need, they were all reliable as heck!
-cr3, the purple haired nerd
By: cr3 on March 10, 2009
at 15:14
So, your standby was 2 hard drives and a fire extinguisher?
In any case, I understand you were maxed out on the PCIe bus, but for some reason, I think even AMD Phenom would be a better choice due to memory bandwidth. Memory bandwidth is just too low on Skulltrail – experienced that on my Skulltrail system with 5x Seagate SAS drives.
Time for FSB to die
BTW, excellent work – it was a brilliant idea and excellent execution. Did you get to keep the system?
PP
By: Theo Valich on March 10, 2009
at 15:38
Yawn. OCZ seems to think they can resell their crappy returned Core V2 drives by throwing them in a 2-slot box with a Highpoint 3520 and calling it the Z-drive. 8x 120GB unsold/returned V2s in a box with an 8ch controller and 256MB cache and sell it below street price of just the SSDs? (i.e. $400 controller for free?) Can we smell the desperation?
Corsair is going to kick their ass once they not only come out with “performance-optimized” SSDs (versus the current rebranded Samsungs they admit are “reliability optimized”) and a real PCIe SSD like Fusion-IO but not priced to the moon.
Oh, and as for controllers, 6GBps SAS + PCIe 2.0 coming right up: http://tweakers.net/nieuws/58824/areca-toont-6gbps-sas-raid-adapter.html
Interestingly, Areca has dumped Intel for Marvel as the CPU brand of choice. I wonder if this has anything to do with the semi-closed Intel/LSI firmware code, and the issues the latest gen Intel-based boards were having with Samsung Spinpoint F1s and Mtron SSDs? Think maybe Intel is waging a quiet little war against Samsung-based controllers so that SSD RAIDs don’t crush their own SSDs? Nah, they’d never resort to that, right?
By: Toby Hudon on March 11, 2009
at 14:39
hi. There. They. Broke. The. Test. Meter. Became. It. Was. Going. Too. Fast. They. Need. Too. Run. The. Bench. Marks. Up. Too. Say. 2. Terabytes. Per. Second. That’s. Right. A. 2. Tb/s. Scale
By: James. Braselton on October 28, 2009
at 05:38
Do you mind if I quote a couple of your posts as long as I provide credit and sources back to your website? My blog is in the very same niche as yours and my users would genuinely benefit from some of the information you provide here. Please let me know if this ok with you. Regards!
By: Elias Vina on September 6, 2011
at 22:07