Vertex 3 real world performance

[JohnnyLucky]

Brand new review:

http://www.anandtech.com/show/4341/ocz-vertex-3-max-iops-patriot-wildfire-ssds-reviewed

Real World performance on page 3 of the review is pretty much what I've been saying for several months.

 

[groberts101]

Johnny.. regardless of constantly lumping everyone you tell that to into one "average user" catagory?.. the Heavy Storage bench results show HUGE time saving's with a Max IOPS drive over a similar capacity non-IOPS version. Even the light usage storage bench shows larger time saved with the max IOPS over the regular version.

Personally speaking.. I could care less about the numbers in benchmarks as the disk busy/project completion time is my overall measure in real ife. Is one of the best things I like about Anands current testing methods. Time is the best measure of these drives and that test suite pretty well sums it up. My usage models the heavy suite much moreso than the light suite.

So what you are saying here would be the exact same thing as saying something like.. "most users will never feel the difference in speed when going from a 60GB drive to a 240GB drive". Certianly NOT the real world result if you do more than just navigate your OS's GUI, open the occasional word doc, or surf the web. And God forbid that you ever write to your drive?.. it becomes even more noticable.

 

[JohnnyLucky]

I think there is a misunderstanding.

The point on page 3 of the new AnandTech review I linked to indicates that real world performance does not always match synthetic benchmark results. This is the first time I have seem more than a sentence or two addressing the subject. Usually I only find bits and pieces of information.

It appears as if a lot (but certainly not all) ssd's form a very tight performance group with only minor differences in real world performance. The synthetic benchmarks are designed in such a way that they greatly exaggerate those minor differences. In that respect the synthetic benchmarks can be somewhat misleading. The problem is compounded by differences in motherboards, chipsets, datalinks, and software applications used.

 

[groberts101]

Yeah..I feel ya... and I know you have been seeking info on this question for a while now. My point was just that "real world" is just sooo subjective these days. Many systems are capable of larger amounts of mutitasking than ever before and more are taking advantage of it simply because.. they can. The hardware really has come a long ways now. Biggest issue IMHO is the software limitations(multi-threaded operations) and even the file systems themselves have inherent weaknesses. I personally wish NTFS would just go away.

So,.. I would have to agree for the most part that most would never know the difference between one fast 6G SSD to the next.

The thing that I found most interesting though, and actually quite hipocritical, was that the dialogue you speak of in that article said that same thing?.. BUT.. then went on to show literally hundreds of seconds for "disk busy" time saved(page 6 I think?) with the heavy workloads. Then the jump between the 120's to the 240's was quite amazing as well. That's SURELY something you can see, feel, and touch in the way of the overall user experience, right?

And more importantly?.. the time saved is why I bought these drives. NOT for benchmark numbers as once you've seen a couple hundred speed tests?.. it's all just numbers on the screen with no "real" perspective for actual usage. That's why I really like Anand's newest heavy suite testing for real time saved regardless of benchmarks.

 

[JohnnyLucky]

The disk busy time on page 5 has me perplexed because of the paragraphs that explain the heavy workload tests up above the charts. Here's all the stuff the heavy workload test is supposed to represent:

a lot of:
downloading
application installing
editing in Photoshop
HTML editing in Dreamweaver
web browsing
game playing/level loading (Starcraft II & WoW)
virus scanning
email downloading
use Visual Studio 2008 to build Chromium

I guess I am not a heavy user because I do not do a lot of things simultaneously. When I am creating an 8x10 head and shoulders glamour portrait about the only other thing I do is listen to my favorite Internet radio station while the anti-virus is doing its thing hidden in the background. I would have preferred to find out how long it took each pc/ssd combination to create a finished portrait from a raw image.

The last two days I worked on a database of 80+ platinum and gold certified power supplies for a web page. My ssd was bottlenecked by my own typing and mousing skills, an Internet connection that isn't ideal, and jumbled search engine results. The fastest ssd would not have helped.

 

[adampower]

I agree on both counts. First, time matters. Second, SSD to HDD real time savings is huge whereas good ssd to great ssd savings time is little.

Everyone must ask. What is my time worth? If it is worth thousands per hour. You need the best! If $50/hour and so many hours per year saved... etc. If you are a 'regular user' perhaps a 'regular ssd' will be adequate. Intel 510/320? On my wish list for next build.

 

[JohnnyLucky]

adampower - For the past 3 months Anand has been recommending the Intel 510 until OCZ and/or SandForce solve issues. The 510 is the ssd that disappointed gamers and enthusiasts because of the non-Intel controller and synthetic benchmarks.

In the meantime, the Patriot Wildfire that Anand reviewed the other day has a SandForce SF-2281 controller and brand new firmware. Preliminary results look promising. It may be a very serious contender if it does not develop any issues.

 

[tecmo34]

Brand new review:

http://www.anandtech.com/show/4341/ocz-vertex-3-max-iops-patriot-wildfire-ssds-reviewed

Real World performance on page 3 of the review is pretty much what I've been saying for several months.

I would expect to see Tom's taking this approach soon when reviewing SSD's and looking "Real World" performance...

 

[jaquith]

Hey!

Do you know how many Apps {updaters} are firing-up/quitting/waiting at start-up through up to 5 minutes?!

Real World is great, just test it right...

RE:

 

[JohnnyLucky]

That's one of Anand's old charts he included in the review to demonstrate the problems of a tight performance group. I've seen similar charts for loading programs in other reviews. In one of the newer charts the Intel 510 was only 1/10th of a second behind the OCZ Vertex 3. It is one of the reasons why other tests have been used to demonstrate performance differences. The synthetic benchmarks exaggerate the differences in a tight performance group.

 

[jaquith]

The chart wasn't my point, the Immediately After Boot is my point, you really need to wait 5 minutes before benching anything. Different update Apps start up 'randomly' to further the skew the tests.

 

[groberts101]

I see the many sides here as I'm far from narrow minded about thses things. The point that I would make is that:

yes.. those are very heavy workloads used to differentiate between drives. But the simple fact is that they obviously do just that. They do show possible gains from faster write speeds and higher small file IOPS. I would also ask the question.. what if you "could" do so much more at any one time?

I do multitask heavily(with many of those above mentioned apps) simply because?.. I can. Faster SSD(or RAIDS) makes it possible without one little bit of slowdown being perceivable to me even when copying larger incompressible data sets to my array(which is tough on Sandforce drives).. Try that with a "typical" little 40GB SSD and see if you don't want more, eh? lol

And if that newer Patriot Wildfire drive looks promising?.. well.. it's not any different than the Max IOPS version it seems. If the Max IOPS version is not needed or gain from its use can't be noticed for a lighter workload?.. why even consider it?

Surely you don't think Patriot will NOT be using similar revised firmware code along with the newest OROM's/drivers which other mfgrs have had to wait for as well? They're just coming along at the tail-end of the Sandforce issues and will surely make good use of known fixes right from the get-go. They may have even been holding up production/release until it's all sorted by others, for that matter. Which of course consumers will appreciate.

 

[jaquith]

If you want to see how your core {SSD/CPU/RAM} system really runs try one of my dinky SQL/PHP runs with 20+ files, 2.5GB of raw data records {no image data} and flatten a complex multi-relational data output; complex calcs, random + sequential data, etc. It's much harder than rendering on your system, and makes you realize just 'how slow' your system really is

The real daily output is comprised from 100's of GB of text.

 

[tecmo34]

I would expect to see Tom's taking this approach soon when reviewing SSD's and looking "Real World" performance...

Here is an example of what I was talking about

http://www.tomshardware.com/reviews/revodrive-3-x2,2967-6.html

 

[jaquith]

IPEAK is just one of many benchmarks like PCMark, etc and no more 'Real World' than many of the others. As mentioned above, I look at the IOPS and published Read/Writes amongst other attributes including cost. The Read/Writes published by most OEM Mfg's use ONLY the fastest sequential Read/Write times on the most 'disabled' OS enviornment and generally on the most expensive SATA3 Card they can find, etc.

Testing and Reviews are only 'good' if substantiated, consistent environment, and reasonably obtainable.

Example:
SSD {1~10}
Reviewer 1 + MOBO A + CPU B + RAM C
Reviewer 2 + MOBO C + CPU A + RAM A
Reviewer 3 + MOBO E + CPU D + RAM B + LSI G

This is what happens to many sites and ALL of the tests on the SAME HDD/SSD are different. In the Reviewer 3 instance adding a $900 LSI SATA3 Card.

So my conclusion is there's never going to be the 'ideal' test(s) and never anything approaching 'real world.' I've stated this many times before.

 

[groberts101]

Have to agree with some of that but I never believe much in advertising anymore these days anyways. Proofs in the performance as it relates to your own personal usage when comparing to another hard drive.. HDD or SSD. Proofs in the pudding, so to speak.

 

[JohnnyLucky]

The chart wasn't my point, the Immediately After Boot is my point, you really need to wait 5 minutes before benching anything. Different update Apps start up 'randomly' to further the skew the tests.

Interesting comment. Once the Windows 7 Desktop appears on my monitor screen I don 't wait 5 minutes before launching a software application. I don't think anyone else does either. Would waiting 5 minutes to launch programs be like moving away from a real world scenario?

At the other extreme we have users who run their computers 24/7 so waiting 5 minutes is unnecessary for the 24/7 group.

I think I would rather see benchmarks for rendering and compiling processes like they have over on the enterprise side of the market.

 

[jaquith]

tecmo34 knows already, I own an IDX/REO/GIS & Tax database company and process several states of IDX/County/State data meaning we have all data on every property in all those states, and even with a decent sized server room the daily runs are barely completed. Enterprise SQL again strains the CPUs/RAM/SAS to it's max and IMO a good indication of what you've got under the hood. An i7 980X or i7-2600K would take weeks to what's done in hours.

In the post above, I was referring to the Terminal PC's i7 930/SSD used to test and code SQL/PHP. My utility electric/telco bills are obscene enough, I don't waste $ leaving 'PCs' on.

On weekends I tinker and hobby build nice rigs, and post in the MOBO section but like tecmo34 getting bored.

That said, after startup launch your Resource Monitor and leave it open and watch, most of the background 'stuff' settles down after 5 minutes -- yep even with an extreme rig + SSD. If you're looking at 5.1 vs 5.2 heck the 5.2 may have been faster if not for some 'Google updater', Microsoft time, Flash, JAVA, ping / polling, etc you name it costing 0.2 secs.

 

[tecmo34]

I think both of you (JohnnyLucky and Jaquith) are coming to the same conclusion by different paths... Basically, SSD or hard drive testing in general doesn't truly reflect "Real World Usage" to a degree. One of the issues (IMO) is the testing typically is done with a freshly formatted drive, as a secondary drive. It is not based on a drive with an OS & Apps installed on the drive itself. This was the one difficulty I had with my article, in the testing methods were not "designed" to test the OS drive (set to test secondary & learned the hard way when I wiped a drive of my games). You can see by how my results didn't always match the "benchmarks" but reflected more of a real world setup.

As for being bored... I'm with you Jaquith, which is the one reason I try to move from section to section and avoid being in just one (as much as I can). I burnt myself out after a year of posting in the Systems forum more times in a day than I care to remember. Now, I only stop by for a quick post or a "best answer" I'm like Bobby Fisher of the System section :lol: (well... not really but I am a fan of the movie though).

 

[jaquith]

After going from MOBO Gold -> Bronze, 'the' PHP bug, with enough BA's to be Double-Platinum the only reason I stuck in the MOBO section is because you have to know ALL of the sections well. Maybe I'm the Forest Gump to the MOBO section?!

Getting back to the OCZ Vertex 3, I've never run across, yet, the BSOD issues, but I constantly run across people not understanding their SATA port limitations, Interface vs speed, and particularly those on non-Intel SATA3 ports e.g. Marvell 9128 {x1 PCIe lane} which craps-out 360-370 MB/s.

Real World, subjectively you cannot 'sense' ±0.1 second or ±10 MB/s on a 300+ MB/s SSD - IMO 9/10 people cannot sense ±1 second or much less ±30 MB/s. In gaming, none of this really amounts to anything -- don't get me wrong fast 'is' good. People who run Apps that rely on fast random and sequential R/W cannot have it fast enough and hope for 'RAM Drive' speeds someday soon in their 'SSD' or whatever it'll be called e.g. 'RSD' and without any concern of limited write cycles. I've been looking at Enterprise SSDs what I do would tear them to pieces. OCZ Intrepid -> http://www.oczenterprise.com/ssd-solutions/ocz-intrepid.html

All I know is over that past year and a half the speeds are geometrically getting faster

 

[Badelhas]

Guys, based on your experience, will I notice any real world advantage from going from a OCZ Vertex 1 to a OCZ Vertex 3? I do heavy gaming, watch movies and surf the web, thats all. I should say that I have a recent board with sata 3, the Asus P8Z68-V Pro and a i5 2500K.
Thanks for your input

Andre

 

[jaquith]

^No, barely noticeable game load times. Movies, especially 'uncompressible' formats (MPEG-2/4, H.264, etc) are actually slower on an SSD; the 'trick' to ultra-fast SSD performance is compression/expansion of the files

 

[groberts101]

Guys, based on your experience, will I notice any real world advantage from going from a OCZ Vertex 1 to a OCZ Vertex 3? I do heavy gaming, watch movies and surf the web, thats all. I should say that I have a recent board with sata 3, the Asus P8Z68-V Pro and a i5 2500K.
Thanks for your input

Andre

While not earth shattering.. it's still worth while upgrade(especially on that 6G mobo). Sandforce controllers have lower latency and MUCH stronger multitasking performance compared to any Indilinx based drive. This I know from experience through testing and owning many versions so far.

 

[tecmo34]

@groberts... I thought OCZ's new Octane (based on a newer Indilinx) had lower latency than their Sandforce drives?

 

[groberts101]

that's supposed to be true.. but the previous poster asked about the comparison of a Vertex 1 compared to a Vertex 3.

The Octane will be using faster/larger DRAM and quite a bit more evolved Indy controller with 6G speeds and specially tweaked firmware on top of it all.