Correction: OCZ Vertex 3 Random Read Performance Data
by Anand Lal Shimpi on March 2, 2011 9:37 PM EST- Posted in
- Site Updates
- Corrections
A huge thanks goes out to AnandTech reader Andrei and some very attentive Xtreme Systems forum members. I just got an email pointing me to this thread where one particular number from our OCZ Vertex 3 and Intel SSD 510 articles was called into question. The problem? The 4KB random read numbers for the Vertex 3 were supiciously high. The reality? They were incorrect.
I was just alerted to the error and quickly powered up the SSD testbed to recreate the test. It looks like the original numbers were either run at a queue depth of 32 or accidentally copied from one of the runs of 4KB random write tests. Either way the number was incorrect and has been fixed in all affected articles.
The updated numbers don't change our conclusions. The Vertex 3 is still the fastest next-generation SSD we've tested thus far and it still maintains a random read performance advantage over the Intel SSD 510.
The integrity of our test data is something we all take very seriously here. Errors like these do you a disservice and hurt the reputation I've worked so hard over the past 14 years to build. I do hope this oversight hasn't negatively impacted your opinion of AnandTech - we aren't perfect, but we strive to be. I do apologize to all of you for the error and I will be restructuring how I run and record my Iometer tests to avoid this particular issue from cropping up again.
My sincere thanks goes out to Andrei and the XS folks who helped track down the error and inform us of its existence.
Facebook
Twitter
RSS
49 Comments
View All Comments
One_Hertz - Wednesday, March 2, 2011 - link
Thanks for the quick fix (:LeTiger - Thursday, March 3, 2011 - link
This is precisely the reason that I have been coming to Anand first, time and time again for my tech related news and reviews. - Thank you again for the transparency and honesty you bring to your review process.LyCannon - Thursday, March 3, 2011 - link
Anand-It's because you admit your errors, fix the data, and apologize that will keep us coming back for more and keep us trusting your tests and your opinion!
Calin - Thursday, March 3, 2011 - link
Just what I wanted to say - to err is common, and acceptable - but to accept those errors and let everyone see them (instead of silently update them, or add an update to the article that maybe nobody will see), that's why I like to visit the site.Bigu - Thursday, March 3, 2011 - link
And the fact that tech enthusiasts are spending time to point out odd numbers from Anand's review is a good sign that there are real supporters to Anandtech and its reputation.Anandtech is the only site I trust in terms of SSD reviews.
TheRealAnvil - Thursday, March 3, 2011 - link
"And the fact that tech enthusiasts are spending time to point out odd numbers from Anand's review is a good sign that there are real supporters to Anandtech and its reputation."Exactly!
Keep up the good work.
Frallan - Thursday, March 3, 2011 - link
We know you are human stop trying to act flawless :0)
It is not in making the errors you stand out from the crowd it is in the way you own up to them you place yourselfs at peak of credibility in the Internet IT community.
Keep up the good work
BR
/F
Calabros - Thursday, March 3, 2011 - link
your wealth, is your readers !iwod - Thursday, March 3, 2011 - link
I think someone started the propaganda that Random Read Write equals the Holy Grail of SSD Performance. And we start losing sight of actual Real World Performance that SSD give us. I really hope Anand could bump out an article about it.I started by asking why Toshiba SSD perform better then Sandforce in number of test.
( Those who read some of my comments will known i have been shouting about it for a long time ) And i had received no response. Of coz everyone looking at those silly Toshiba Random Write Number didn't pay much attention anyway.
So with Intel 510 SSD, Intel Wins on Most of the Write Test and lose out a lot on Read test, which resulted in Intel losing overall. However if we take a closer look;
Random write, Using Non compressible data Sandforce is faster by roughly 380% on QD3 to 490% on QD32. That is the minimal wins for Sandforce, if any of those data were compressible you will get 420% to 660% faster. These are HUGE numbers.
On Seq Write, if they were all compressible data you are 12% faster on Sandforce, however if the data are non Compressible Intel actually wins by 22%.
In today's general notion, everyone would have thought the relatively small win for Intel SSD during WORST case scenario, would have been trashed by Sandfoce due to the Huge win in Random Write and 12% win on Compressible Data.
It turns out, with Anand High Workload trace, Intel manage to beat Sandfoce, by a 7% difference. In Light workload, Intel still wins by 2%.
So yea, Random Write doesn't really matter when we pass 50MB/s given todays "HIGH End" work load. Which i think Anand didn't emphasis enough. We could give it 20 - 50% headroom, but that is still only 75MB/s. It also seems like Intel has known this all along and artificially make the limit with firmware.
With Random Read, Sandforce is 55% faster then Intel.
The king of Random Read is actually a C300 which is based on similar chipset as Intel 510, 20% faster compare to Sandforce,
On Seq Read Sandforce does a marginal 5.5% win compare to Intel.
And it should be noted king of Random Read C300 does very bad here and is 50% slower then Sandforce.
On Heavy Workload resulted in 24.5% win for Sandforce ( 50% of its advantage in Random Read ) and 52% win for Light Workload ( Nearly 100% of its advantage in Random Read )
For C300, it is 60% !! Slower then Sandforce in Heavy Workload and 56% for Light Workload.
The truth is Seq performance matters. And it matters a lot. But not many review got to this conclusion. The World of SSD has its focus on Random Read Write. And it is Sandforce the benefits most.
neotiger - Thursday, March 3, 2011 - link
"Real world" performance according to whom? Just because Anand created a benchmark doesn't mean there's anything "real world" about it.I know this website is mostly populated by gamers who care mostly about how fast their games run. But in reality the vast vast majority of the world don't play games other than browser-based ones like Farmville. They don't care about 3D graphics and FPS. And they certainly aren't going to pay $250 for a SSD so they could boot their computer 30 seconds faster or rip a movie a minute faster.
Most people who pay the huge premium for SSD do so because SSD gives them something HHD simply cannot: IOPS. I could spend thousands or tens of thousands of dollars on a disk array that would give me a maybe 5,000 IOPS, or I could spend $250 on a SSD that gives me 50,000 IOPS. For the real world uses of SSD like database and VM, random IO is king and will always be king.