OCZ Vertex 3 Random Read

[Scytus]

Hi! new to the forums

I've made a new thread as in the other ones, the talk of company problems overpowers that of what I'm actually interested in, the V3 drives themselves. I've been interested in the (120GB) Vertex 3 for a long while as I've begun building myself a new system. After reading Anandtech's review here (the only current review I know of, of the V3) I was quite disappointed by the fact that the random read for this drive @ 6Gbps was hardly any different than if it were 3Gbps, and its quite significantly slower than the oldtech X-25M.

I'm not quite knowledgeable in the in's and out's of SSD's, I'm just a consumer that researches and buys. Now from what I know, and from some have been saying, isn't the random read the most important aspect of the drive? How fast the OS responds? The 240GB model did not have such an issue here, but the 120GB model definitely did.

Just looking to get some input, thanks C:

 

[groberts101]

If 4k reads were the most important metric than surely older drives with higher metrics in this area would be much more responsive and faster for an OS volume than a Sandforce. Latency is key and is the primary factor for the increased snap which all moving from an Indilinx controller(just an example) see at boot or in the GUI. Also consider how well these drives review/rate in performance comparisons to others with notably higher 4k reads.

Then there's all this single q-depth measurement going on now. Which seems silly to me since this controller will just dig deeper if they start to stack up a bit and actually has some of its greatest strengths in that aspect alone.

Not saying it's not an important factor to consider, but simply that is is only one portion of the equation.
​

 

[Edrick]

Not saying it's not an important factor to consider, but simply that is is only one portion of the equation.
​

True. How the drive is being used is also a consideration. A boot drive will have some different "important factor to consider" than a game drive or a 'working' video drive for example.

Personally I put a lot of value in the random 4K reads for my needs (hence my decision to purchase a C300). But not everyone's needs are the same as mine.

 

[Scytus]

Well I've been told that most enthusiasts claim that the random read of and SSD is one of, if not the most important metric there is for a snappy system. Is this true? would the year old C300 (for example) be a lot more responsive with the OS?

 

[groberts101]

"a lot more responsive" would be stretching it as the V2/V3 is very responsive. At this level your already at state of the art.

And I would say that latency is king.

 

[Emulex]

yeah 99% of the population - the non AT elite - will never know the difference between a X25-V and Vertex 3 (other than free disk space). this is truth. but it stands to make sense since your read/write ratio is so high.

 

[Old Hippie]

I was quite disappointed by the fact that the random read for this drive @ 6Gbps was hardly any different than if it were 3Gbps

Let's start with this.

The interface speed means nothing if the drive can't take advantage of it.

You're not going to see any increase in speed by going from SATA3Gb/s to SATA6Gb/s unless the drive is capable of faster speeds than SATA3Gb/s allows.

 

[Scytus]

Let's start with this.

The interface speed means nothing if the drive can't take advantage of it.

You're not going to see any increase in speed by going from SATA3Gb/s to SATA6Gb/s unless the drive is capable of faster speeds than SATA3Gb/s allows.

Could you comment on AS-SSD access time?

The 240GB Vertex 3 has an access time of about 0.050ms, as does the Crucial M4. The 120GB Vertex 3 however has an access time of near 4x that, shown here:
http://www.ocztechnologyforum.com/forum/attachment.php?attachmentid=19243&d=1302391628

Also there seems to be a drop in concern for random reads in all drives, as Crucial (especially), Intel, and OCZ have dropped their random reads around 40-50, when previously the C300, for example, was topping around 100.

They seem more concerned with sequential speeds..which makes no sense.

 

[Old Hippie]

Could you comment on AS-SSD access time?

I could but it wouldn't mean anything because I don't pay much attention to it. LOL!

I've had 10 different SSDs mostly in RAID0 set-ups and right now I'm getting kinda burnt-out with following the "current" line of thinking on what makes an SSD "fast".

ATM I'm running 2 Crucial 256GB C300s in RAID0 and they do pretty well in all areas.

 

[groberts101]

Here you go Scytus. Courtesy of one of the greatest out there right now via PM.

Computurd wrote. "The best description of QD I heard from an intel fellow at IDF...
he explained it like a cup of water..

if you have a cup with a hole in the bottom that is small the water will drain slowly. if you have a cup with a big hole, it will drain faster. the amount that builds up waiting to get out is like QD. the bigger the 'hole' the faster it goes. the speed of the storage solution is much the same.

Yes too much emphasis on 4k. I've been screaming that from the rooftops for years...but it falls on deaf ears. It is the overall performance of the drive that matters. And writing. If they really knew their stuff they would be comparing 4k write, esp mid QD, more than anything. Writing is THE inherent weakness of SSD's of any flavor. If you are *actively* writing, your read drops significantly. Simple test is to load iometer, and do a 80% read and 20% write test....watch your numbers go through the floor! The writing "clogs" up the onboard controller of the SSD, thus screwing you over on read speed. That is the thing with the newer generation of SSD. Faster WRITES! In real world usage that everyone has such a hard-on for, the SSD that writes faster, wins. It is the mixed read/write performance that either makes you, or breaks you.

Latency isnt being measured correctly on the V3 with the AS SSD test. AS SSD measures latency with 512b size file. That is incorrect. Latency should be measured with 4k *or* the base transaction size.

So...I need to see iometer 4k@qd1 for latency. then also, just for kicks, since page size is now 8k, I would like to see 8k@qd1 for latency.
Since 8k is base transaction layer....we should be able to see the TRUE latency there.

I need to test one to be honest, LOL.

4k is an important metric, but TOTALLY blown out of proportion. The people who are using it as a measuring stick, know not what they do....the true usage of it is to check latency. BUT what about the latency @qd2 and @qd3 so on so forth....if you have a device that at qd1 is average, but at qd2 it is twice as good as other drives...well that is the thing! You have to have a comparison of the latency across the different qd's.

Base page size of V3 is 8k, correct? So why in the world are they testing the latency @4kQD1?? It needs to be tested @ 8kQD1 since that is your 'base' transaction size.

But...even if you were measuring the random at any size of file...the QD1 is NOT the most relevant metric. A comparison of qd1 to qd8 would be the important measurement."

The last statement pretty much sums up why Sandforce is so strong due to better IOPS than all others on the writes. And if data starts to stack up?.. the controller just clears the path with much stronger performance at deeper que's.

I'd say just pick a top rated drive and enjoy it as all this hair splitting is just wasted time that could be used for actually enjoying a fast SSD in the present. The difference at this level will NOT be night and day. More like.. dusk to dawn.

 

[Old Hippie]

Yes too much emphasis on 4k. I've been screaming that from the rooftops for years...but it falls on deaf ears.

After 30 years in a steel mill my ears aren't the best but my reading skills are exceptional and I don't recall any rooftop screaming.

ANY drive is going to falter when reading/writing at the same time and it's no big deal.

BTW, Computurd seems to be well versed but to call him "one of the greatest out there" ? LOL!

 

[groberts101]

LOL. I'll try to speak up from now on. Yeah, I hear ya. A lot gets lost in translation and reading between the lines. Maybe I should have added "testers" in there, eh?

My extensive/dedicated SSD testing has taught me one thing so far.. the proof's in the pudding. IMHO, there are far too many experts that are just parroting what they've read a hundred times without much firsthand experience.

Not sayin' you're one of em'.. but I know for fact that Comp isn't either. Nuff said.

 

[=Wendy=]

The pendulum is swinging towards sequential performance for desktop SSDs in the real world. 4K bandwidth has been more than enough from the days of the Intel G1 and Indilinx based SSDs regarding desktop use.

It's no error that SSDs with excellent sequential performance, such as the Intel 510, Crucial M4, Corsair P3, Plextor PX-256M2S, and of course the Vertex 3, are topping the "real world" tests in reviews.

 

[Old Hippie]

My extensive/dedicated SSD testing has taught me one thing so far.. the proof's in the pudding. IMHO, there are far too many experts that are just parroting what they've read a hundred times without much firsthand experience.

And I've had 10 SSDs starting with an old GSkill/Intel G1's/Intel G2's/Vertex2s to the newer 256GB C300 Crucials ALL in RAID0 and single drive modes.

Your point is very well taken but it's obvious I've had a much different experience than you.

PSSS...So who's parroting what?

 

[groberts101]

LOL. I'm old enough to remember all my mothers hippie friends and I think I'm startin' to like.. dig you.. maannn. Power to the people! Funny stuff.