OCZ Discovers Bug in Onyx SSD, Updated: We're in the Clear
by Anand Lal Shimpi on May 5, 2010 1:19 PM ESTA week ago I got a shipment of a bunch of new SSDs including OCZ's long awaited Onyx. This is based on the Indilinx Amigos controller, which is basically a cost reduced version of the Barefoot controller with only half the channels. A 32GB Onyx should sell for around $99.
The first drive I received showed CRC errors during a Windows install. I told OCZ to stop shipment two Fridays ago. OCZ responded by saying that they wouldn't stop shipments after only one bad drive. Their mistake.
In my experience, I never get a "bad drive", it's either DOA or has a firmware/controller bug. Three days later OCZ issued a recall on the drive and got in touch with its customers directly to make sure they were taken care of.
OCZ had the bug fixed within a week and sent me a new drive (as well as posted an updated firmware on their site). I've been testing the fixed Onyx and while it doesn't exhibit the same issues as the first drive, I am seeing an unexpected CRC error in one particular test that shouldn't be there. I just got off the phone with OCZ and they are going to be pulling back all drives until the problem is squashed.
Building SSDs is a challenging business. I stressed to OCZ that this could have been avoided if they just sent out samples a week before shipping to etailers. I have to at least hand it to OCZ for acting quickly to pull the drives, especially after today's warning.
Update: OCZ's earlier firmware update appears to have squashed the bug completely. Its engineers and I have been working to reproduce the more recent CRC error and it doesn't appear to be an issue with the drive itself, rather something limited to my test platform. I was concerned about the possibility of another data corruption bug given the issue I had with the first drive and arrived at the wrong conclusion. Using the latest firmware the OCZ Onyx has completed almost all of my tests thus far without issue. I will keep you all updated on my experiences with the drive.
17 Comments
View All Comments
tejas84 - Wednesday, May 5, 2010 - link
I cannot believe you are comparing SSD cost per gigabyte to a mechanical HDD. Winchester mechanical drives have been proving themselves for decades. Yes they are a lot slower than SSD's but they are bulletproof when it comes to data.Reliable data storage is non negotiable even for an OS, I will sit back and watch you early adopters have a load of stress while I reap the rewards of a stable reliable cheap SSD a few years from now!
It remains to be seen whether the wear levelling on SSD's will stop data loss or corruption. I admire the controller tech especially Sandforce SF-1200 but all of you early adopters cannot say that any OEM's will be selling these overpriced SSD's to the masses.
The great unwashed buy mainstream stuff in case you did not know that!
GullLars - Wednesday, May 5, 2010 - link
"I cannot believe you are comparing SSD cost per gigabyte to a mechanical HDD. Winchester mechanical drives have been proving themselves for decades. Yes they are a lot slower than SSD's but they are bulletproof when it comes to data."He was NOT comparing cost pr GB to HDDs, he was comparing Intels SSDs "bang for buck" to other SSDs, and it's true they are in the top in this respect.
And HDDs being bulletproof when it comes to data, that's a fallacy. I've seen a lot of harddrives fail over the years, especially in laptops. Most HDD failures result in total loss of data, or very costly recovery. HDDs also have a 3-5% anual falure rate, if you discard DOA, infant death, and FW buggs or flashes gone wrong, SSDs seem to be <1%.
"Reliable data storage is non negotiable even for an OS, I will sit back and watch you early adopters have a load of stress while I reap the rewards of a stable reliable cheap SSD a few years from now!"
Reliability of storage is totally negotiable. If you don't store user data on the SSD, and have a spare HDD (or small dedicated partition on a HDD in use) to substitute in case of the SSD failing, downtime and data loss can be virtually negleced in end-user enviroments. Downtime from re-imaging is normally 15-60 minutes.
"It remains to be seen whether the wear levelling on SSD's will stop data loss or corruption."
No it doesn't. SSDs have been around since the 70's, and flash SSDs since the 90's. They have mainly been used in the enterprise, so you likely wouldn't know. TMS, Stec, BitMicro, Curtis and Adtron all have great track records in the enterprise.
It remains to be seen wether low-end consumer-SSDs have sufficient CRC/ECC strenght and methods of countering silent errors, especially with cheap MLC.
This is the reason why i bought Mtron Pro (enterprise version SLC with a full 5-year warranty for use in servers) to RAID back in 2008, and gave roughly $1000 for 64GB to use for OS and apps. A short-stroked partition on my 300GB Velociraptor is slow in comparison, the difference comparable to an old 5200rpm 2,5" disk compared to the velociraptor. When i went back to OS on the velociraptor temporarily after 1 year on the SSDs just to test some settings when i installed W7, i cought myself wondering WT*'s wrong with my machine?
Belard - Wednesday, May 5, 2010 - link
Yeah, why?Oh, of course. The un-suspecting non-educated consumer are looking at the $80~99 price tag on these drives, thinking that ALL SSDs are the same.
OCZ has SO many variants on their drives, its hard to keep up.... what, 8 different lines I think. Intel has 2.
So for $100, a person gets 32GB of a reduced performance version of an already not-so great SSD.
$210 gets you an 80GB drive that'll blow it out of the water with performance and security.
Intel, who I don't generally like as a company - still makes the fastest (overall) drive on the market - as RANDOM performance is more important than sequential/burst mode. The year+ old X25-G2 is still near the top and has the best reliability!
I can't wait till all BRANDS of SSDs can reach the quality of intel's X25. Because so far, Intel is STILL the best to get and generally cost less than the others too, go figure.
Chloiber - Wednesday, May 5, 2010 - link
Well, to be honest, the Indilinx Controller is even older than the X25-M G2 ;)But Anand was right when he said "Intel just conroed the ssd-market"...they did... ;)
B3an - Thursday, May 6, 2010 - link
Intels X25-M's are not the fastest anymore. The C300's and other Sandforce drivers are now faster.Intel also had firmware issues too, nothing serious, but still.
I've had two X25'M's in RAOD0 for months, now i have two Crucial C300's, and they ARE faster in random writes, sequential writes and reads. All areas. You just have to look at the Anand review to see this. Try reading up on stuff next time so you dont look like such a fool.
Roy2001 - Wednesday, May 5, 2010 - link
I have used 80GB 2G Intel SSD for more than 6 months and it is good. My laptop just got another one and improvement is huge!Jeff7181 - Wednesday, May 5, 2010 - link
I used to like OCZ, not because of their products, but because of their support. It was always easy to get in touch with someone that speaks English well and made an honest effort to work with their customers to rectify a problem they had. Their lack of concern for their customers seems to be increasing lately. They've lost one customer for good, I have a feeling if they keep going down this road they'll lose more.