OCZ's Vertex 3: Second-Generation SandForce For The Masses

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.


The 3Gb/s drivers were always tested in the 6Gb/s ports. The only time we used the 3Gb/s ports were to force the 6Gb/s drives into a 3Gb/s configuration.
 
I appreciate the MTBF discussion, and the math is not lost on me. Like Phate, the burning question for me is how long can these drives realistically be expected to last under typical usage scenarios?

I have seen SSD products with >10% failure within the first month on Newegg's forums. Even though that is not conclusive, and possibly skewed, it is troubling, given that a failed drive can represent substantial loss of time getting back up and running, and possibly permanent loss of data/work/play. As far as I am concerned, the jury is still out on reliability and longevity of SSDs.
 


Well it is important to separate wear from reliability. You could have a drive that doesn't fail but uses up all its PE cycles (that is why write amp is so important). In the opposite scenario, the drive could completely fail because the controller or NAND device had a fab error. When we talk about MTBF, we are talking about reliability. This would be the equivalent of a HDD suddenly start suffering from motor and head errors (that dreaded click click).

I'll try and look into reliability and see if we can get some concrete answers.
 
[citation][nom]boletus[/nom]I appreciate the MTBF discussion, and the math is not lost on me. Like Phate, the burning question for me is how long can these drives realistically be expected to last under typical usage scenarios? I have seen SSD products with >10% failure within the first month on Newegg's forums. Even though that is not conclusive, and possibly skewed, it is troubling, given that a failed drive can represent substantial loss of time getting back up and running, and possibly permanent loss of data/work/play. As far as I am concerned, the jury is still out on reliability and longevity of SSDs.[/citation]

Wondering which SSD's you speak of. I have had an OCZ core 2 fail on me within 3 months and I have another both bought black friday when they came out for 99 bux each. (30GB) one I beat to hell, half the paint is missing used it for USB storage medium, 40% and laptop hard-drive about 30% of the time since then, and a desktop hard drive (primary) for about 10% of the time. another 20% spent banging around the center console of my 02 WRX wagon in the Texas hot sun, (I define the car due to how much jostling a bit of spirited driving in an AWD car can create) I just took it out of the console a week ago and through it in a Core 2 duo 2GB RAM Sony Vaio laptop with windows 7. Woke that sucker up. and it hasn't had 1 hiccup.

What I'm trying to say here is, if it's going to fail I would think it will relatively quickly (within 4 months or so up to a year) if it was not a "lemon" it may last 20-100 years.
Out of the 5 I bought that day only one failed, although the other stories are not quite as colorful.
 


That has actually been the convention, to assume that there is an "infant mortality effect." According to the studies (see the links in the article), we actually know that doesn't occur with HDDs. The limited data that I've see on SSD (all NDA so I can't release it quite yet) also suggests that there is no infant mortality effect.
 
[citation][nom]bto[/nom]Wondering which SSD's you speak of. [/citation]
For example:
http://www.newegg.com/Product/Product.aspx?Item=N82E16820227461 ,
which is not atypical.

[citation][nom]acku[/nom]Well it is important to separate wear from reliability. You could have a drive that doesn't fail but uses up all its PE cycles (that is why write amp is so important). In the opposite scenario, the drive could completely fail because the controller or NAND device had a fab error. When we talk about MTBF, we are talking about reliability. This would be the equivalent of a HDD suddenly start suffering from motor and head errors (that dreaded click click).I'll try and look into reliability and see if we can get some concrete answers.[/citation]

Of course both are important, but reliability is my biggest concern at this time. Thanks.
 


I'm glad you raised the issue. I've long wondered this myself. Most people assume that because there are no moving parts, the failure rate is lower. However, this is just an assumption without any hard data. We know that mechanical doesn't mean less reliable -- i.e. first gen hybrid cars.
 


Out of about 30 SSD's used in various deployments only 1 died, So I obviously don't have enough data to compile something of the scale that you can look at. In all honesty I think your right, but this infant mortality that we see may ghosting due to power fluctuations/surges, and also static discharges during installation or during upgrades. in fact It died when I re-implemented it so it is possible that may be a big factor, these are more susceptible to discharge than a regular HDD; Which I find DOA quite often, much more so than SSD's in general.

Static discharge can happen without you noticing many times, just because it doesn't jump, or the fact you couldn't feel it just means it's under 1000 volts and hadn't built up to a noticeable discharge. I find this same rate in RAM and flash drives.
 


Well I can't speak off SSDs, but we know that for HDDs those ghosting effects had no impact. In the paper by Bianca and the subsequent Google paper they looked at thousands of drivers deployed at big data centers in huge clusters like the High Performance Computing Division at Los Alamos National Lab, Pittsburgh Supercomputing Center, and the Lawrence Livermore National Lab. Google's paper was based on thousands of drives in its own lab. The drives they analyzed were deployed once then just monitored. Secondary variables (ghosting as you refer to it) were accounted for in the statistical analysis in both studies. Since we are talking about national labs and huge data centers, I'm ready to discount static discharges and other factors that may be an issue within the home.
 
Yes, if it was a professional lab then, you can rule it out I'm sure google has them all strapped in to minimize such issues as they would become apparent rather quickly. Even in a small computer shop years ago I set up mats and straps wired to an isolated ground. As an Administrator, I keep one hand on the case (or elbow) during the installation and during any touching of the inside.

All that said, what is the issue based on your experience? Honestly I think they are 3-4 times more robust than a standard HDD about 20 times more robust than Maxtors (had to throw that one in there). Experience or opinion what is "your" verdict?
 
Why does the Vertex 2 beat the Vertex 3 when both are in a 3Gb/sec configuration? Specifcally in application loading using PCMark, the Vertex 2 wiped the floor with the Vertex 3. But then when the Vertex 3 is put into 6Gb/sec it gets an amazing score. I would think that if the Vertex 3 can get a great score while connected at 6Gb/sec it would max the bandwidth when connected at 3Gb/sec, but it is not the case. Any ideas why?
 


It's usually that they are tuned to a certain throughput, backwards compatibility sometimes can cause this.
 


The current price for that drive is on the Egg is $219. Egg had a Gorilla Deal for 20% off of all SSD's (-$44 = $175 + shipping = $179). OCZ has a $40 rebate going (gets me to $139).
 


I try to be agnostic. I get the sense that SSDs (in terms of reliability) have caught everyone a bit off guard because there is no data to draw conclusions from. We are still in the first/second generation, so there has hardly been enough time to collect relevant data. Everyone is basically playing the game of "SSDs should...." I'm interested in the raw numbers.
 
I understand your point, honestly, if it's built with good components and the build process is proven (which is what we are both watching) then the only other factor is NAND wear which in some implementations can be mathematically proven.

As they get better at building them, early adopters sometimes get the brunt of the issues. I'd like to think we are past that now that windows 7 plays nice with them among other maturing reasons; I still have an
Apple II E that the kids play with. Hell, it works great! The components used to make this were built to last. When you rule out mechanical failures, things are then based off of component quality. (Not that it's anything like it was when the Apple II E was made LOL)!
 


bto hit the nail on the head. But to add to his explanation, there are so many variables we can't actually draw a definitive conclusion beyond saying using a Vertex 2 in x works better than Vertex 3 in y. There are so many cases where you can back a SSD into behaving a certain way.

We secure erase our drives before every test to limit the affects of garbage collection, but there are other issues specific to changing ports. Remember, the 6Gb/s and 3Gb/s ports have different controllers. I'm not even sure that the ports use the same driver AHCI driver. It is possible that the Vertex 3 was simply optimized to the 6Gb/s. Second PCMark, as we explained in our results, has a wide variance when we move to the C300, Vertex 3, and Vertex 3 Pro, and I suspect this has to do with the insanely high throughput compared to older drives. Furthermore, when PCMark sees a number that it deems odd, it just tosses it out the door rather than use it. How the host controllers prioritize requests with the CPU, all add additional variables to the the question of why.
 
A novel could be written as to why something benchmarks differently than it's "supposed to"

I've seen a weak power supply affect a benchmark negatively (not overclocked)

I've seen an overclocked cpu positively affect an SSD. Actually my results to this sparked OCZ to make a Vertex "Turbo"
When I found you could literally overclock your SSD via the port clock speed (around 8-10% increase anything else just wouldn't work). The SSD gets it's clock from the port
 


I'll second that. However, technically, this is what OCZ considers the "near mainstream version" There should be an Agility 3 out at some point, but like the Vertex 3, it will be priced higher the the 3Gb/s version (Agility 2).
 


Ha. That novel would be a definite snooze. There would be so many footnotes; I'm sure we would clear an entire forest for enough paper.
 
To expand on my previous comment. I think you got the size of the Pro confused with the regular drive. The Pro is the one available in 100GB.
 
Status
Not open for further replies.