Anand Vertex 3 Review

[One_Hertz]

Ok this graph really grinds my gears:

Image removed because Anand fixed the issue

Looking at all the other Vertex3 reviews, the ~190MB/s values for 4KB reads are actually the result from using QD32 NOT QD3 like Anand is claiming. Extrapolating from all the other reviews, the real QD3 numbers should be around half of the result he claims, if not less. By definition, the numbers from QD1 can not increase more than 3 times for QD3. Since QD1 is 3XMB/s or less, QD3 can not possibly be 190MB/s.

Does anyone here talk to the guy? Could you drop him an email? He is going to keep reusing this most likely false info (like he did in the Intel 510 review).

[RyderOCZ]

Most likely a typo in the header of his chart is all. He just left off the 2.

Nobody tests with QD=3 that I know of.

[zalbard]

You can email him yourself. Click on his name in the article for a mailto link.
And yeah, I think Anand's results are highly suspicious. And it's not just V3...

@ RyderOCZ: Anand does. He reviewed Intel 510 and tested it @ QD=3 as well.

[RyderOCZ]

Oh really.. interesting.

One-Hertz, I will test if for you right now if you like.

Only SATAIII board I have up and running is an MSI 890FXA-GD70 though.

[One_Hertz]

A quick iometer run would be awesome

[Anvil]

It looks suspicious.

Here's from my previous testing of some of those SSDs.
Based on the C300 the readings look like they are measured at QD3.
The C300 at QD32 ~ 220MB/s

[RyderOCZ]

Ok, I am apparently rusty with Iometer... where do you set the QD? {embarassed}!!

[Anvil]

Outstanding I/Os is what you are looking for.

edit:

Link to thread with iometer profiles

[alfaunits]

That MIGHT also be compressible data?

[One_Hertz]

That MIGHT also be compressible data?

This should not matter much for low QD 4kb random tests. The bottleneck in this case is the time the SSD takes to locate each piece of data which is requested (i.e. access time). Compression can not speed up this kind of latency. Compression can speed up how quickly the data is read once it is located.

[RyderOCZ]

This should show my config for the test and the results.

My system and drivers:



Iometer config:





Result:

[One_Hertz]

Thanks Ryder!

This proves Anand made a mistake in his testing...

[RyderOCZ]

Or he just put the QD32 results with the QD3 header

He did use P67 and I used SB850, but I can't see it making 100+ MB/s difference.

I just repeated the test with QD32 and got ~160MB/s average.

Note that my tests are not with a fresh SE'd drive, I have Windows on it (for another machine) so all the blocks are not fresh and clean.

[alfaunits]

Someone call the Shrimp, to make sure.
Frankly, I believe it is a case of a good controller (ICH) and a fresh drive with compressible data. If the data consists of typical IOMeter data (a repeat of 0123456789 over and over), the controller could very easily push that in a very small amount of space and need not read the actual cell content but minor auxiliary data that says the LBA is a copy of another LBA which was already read (hence in cache). 128MB cache or even 64MB would be enough to hold an 8GB IOMeter file in compressed form.
Either way the results would be pointless, but the test might be true.

[jmke]

Someone call the Shrimp,

you called?

[Anvil]

That mistake does make a huge difference to that chart.

The SB850 is actually quite alright at low QDs, not so sure it will make that much of a difference.

The MyCE review puts the V3 at ~95MB/s at QD4 and pcper shows that it really starts to pick up speed at QD > 4

[Brahmzy]

Wow - this changes the game a bit. This drive is not so head n shoulders over the 510 as we may have thought?

[F@32]

OK, I hope he reads his e-mails. These 4K numbers in Vertex 3/Pro and Intel 510 reviews are nuts...

[Computurd]

well i read vertex 3 reviews last night until my eyes bled. i mean i read like 9 of them front to back. i too noticed some inconsistencies in testing, and some sites just obviously do not know how to test.
the anand results did look skewed and that is a helluva call there OneHertz, nice catch.


the Anand result and the results of some other sites had me scratching my head.

but lets keep things in perspective here: from all of this reading that i did i came to one conclusion

this is one helluva drive. i think right now it is the clear leader (duh) and unless intel really tweaks its next series that is coming out, this here Vertex 3 might be the leader of the class for a good good while.

i wasnt reading reviews over and over without the intentions of picking up a few (or 8) of them very soon.

you might think that would saturate most raid cards, but ohhhhh...what i know....

now all we need is a lower capacity V3 to save me at least a little bit of money

@anvil: hey there is a 2008 verison of IOMETER, but i cant seem to find it anywhere. is this the same as the beta that you posted in the other thread?
ryder: can we please get a QD1 test?

[F@32]

ryder: can we please get a QD1 test?

By praz:

[johnw]

For some reason, the 4KB random read QD=1 numbers are hugely different between AS-SSD and and CDM on Vertex 3. I asked about that in the AS-SSD thread on OCZ forums, hoping the AS-SSD author might have some ideas, but no replies yet.

Here is an example of the discrepancy:

http://thessdreview.com/our-reviews/...-and-win7-wei/

The QD=1 results I have seen from IOMeter are closer to AS-SSD than to CDM. I think CDM's number may be misleading.

[Computurd]

yeah i would like to see iometer QD1.

btw F@32 is that jim carrey?

[F@32]

btw F@32 is that jim carrey?

Lol yeah, he was epic in the SNL. Especially Black Swan skit: http://www.hulu.com/watch/205721/sat...ive-black-swan

[onethreehill]

[Anandtech] Correction: OCZ Vertex 3 Random Read Performance Data

Anand read your email

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....

http://www.anandtech.com/show/4204/c...rformance-data

[One_Hertz]

Good to hear. Who actually ended up emailing Anand? I didn't...

And my 70mb/s prediction was almost spot on (: