Anandtech investigates Intel's SSD performance

"Robert, the main reason why it's good to defrag a mechanical hard drive is because it takes significantly more time for the head to read fragmented files (especially if they're scattered in non-sequential order) than it takes to read non-fragmented files."

Right, but that's far from the only reason. Another reason is that it takes significantly more time for the drive to write out a file if the drive's free space is fragmented. Flash drives are notoriously slow at random writes.

OCZ Vertex uses Indilinx Barefoot controller (Indilinx is a new company). All the JMicron controlled SSD's have the same problem: the controller can deliver high sequential throughput (i.e., while copying one large file), but BADLY stutters, even worse than traditional HDD's in random write performance (like writing many small 4 KiB files). When buying an SSD, the most important thing to look for is the random write performance, since small random writes keep happening in the background when you use your system. Sequential write performance is not that important, since you're not going to store many big files in the SSD's anyways - since they're so expensive, most people will only use it as a system drive and not a data drive. Intel drives offer around 80x to 100x random write performance of JMicron drives. But Indilinx controller is much better than JMicron. But Intel still offers 4x the random write performance of Indilinx. Considering that Indilinx is half the price of Intel's drives, that is not bad performance at all, especially from a new company.

I read this article last night and the issue comes down to cell management. Base flash logic says you can write 4kB increments but only delete 512kB increments. So if you need to write 8kB located in two different areas of the drive it may involve relocating 1024KB of data since you have to move the other data on the 512kB page you want to keep.

That causes stuttering when, say, you got a web page that has lots of small gif images as instead of writing ~80kB your SSD acts like it's writing 8MB.

The Intel SSDs have an intel-designed controller that is better at dealing with the page management issues while keeping performance because in part they have extra non-partitioned cells to allow those deletes/rewrites to happen durring lulls and because the controller has onboard cache that doesn't require tapping system memory over the SATA link to do the rewrite/move. Intell SSDs do degrade but not the same huge way the SSDs with JMicron controllers did and the degraded intel units still perfomed much faster than hard drives.

The OCZ Vertex uses a different controller (samsung I believe) that can also handle the write issue but loses some of the peak performance. It degraded a little bit but also outperformed hard drives when degraded.

The Intel drives are the FASTEST of the lot. The others still sell JMicron crap (except OCZ). Look at the random writes (the most important thing for an SSD) test at this page and you'll know:
http://www.anandtech.com/storage/showdoc.aspx?i=3535&p=3

According to Some tests, SSD Break Down upon first useage. Remember. these Units could take hammer blow & still operate, So Its NOT Surprising that some minimal output continues after theFRY.org. Maybe 3 Volt SSD or Contollers for 1 Volt SSD or Move SSD unto CPU, by changing SSD interal address to match cpu, it probably does that automatically, anyway, You Be Engineer & figure out seperation of Volatile from Non Volatile memory in 1 volt system W/ CPU & Wowie, Mult-millions: Bucko. STeWie drashek

Ken, I 'defrag' my ram every time I reboot my computer. It gets a partial defrag every time I close a program. Now, ahem, what was your point? :P

Pointless benchmark

"Anand of Anandtech has had a crack at it and he has managed to replicate the results which means that Intel must have been able to do so too."

That's the reason why I don't touch Anandtech. :\

Like Anand said in his article, he could reproduce the "used SSD" slowdown, but it was with a barrage of benchmarks. Some slowdown will most likely occur with normal use as well, but I honestly don't think it's going to be a problem.

Even with my crappy OCZ Core V2 60gb, I'm happy after using it since Nov 08..

of course.. it would be nice with a Vertex or X25-M ;)

Robert, the main reason why it's good to defrag a mechanical hard drive is because it takes significantly more time for the head to read fragmented files (especially if they're scattered in non-sequential order) than it takes to read non-fragmented files.

SSDs "shouldn't" have this disadvantage, since the technology they're based on, RAM, is quite good at "random access" data. How often do you defragment your RAM? That gets written and rewritten far more often during normal usage than the hard drive.

In any case, I'm not surprised. SSD is a new technology. Its benefits are limited mostly to server environments still.

I understand the article as, in normal use of desktop, NO degradation occurs as the condition only appears in specific benchmarks under specific set up runs, and no one who use the disk and computers ever do that.

if you don't defrag normal disks, they slow down significantly, so I guess that proves normal use of mechanical disks slows them down.

No one put such up as a headline?
Use the disks normally and they slow down unless you defrag them?

Anandtech actually did another article about this a while back. It's not only Intel's drives that suffer this problem, it's actually to do with the general operational patterns of SSDs.

http://www.anandtech.com/storage/showdoc.aspx?i=3531

1.) Of course a company like Intel is going to lie about something like that unless it its obvious.
2.)Are you saying Anand is somehow complicit in hiding the fact, saying it was "buried" deep in the article?