Vertex 4 Reviews and Thoughts

[Highendtoys]

That is not accurate according to OCZ. We were told the 512GB gets 1GB, 256GB gets 512MB and 128GB gets 256MB.

[Nizzen]

http://techreport.com/articles.x/22736/1

Not the best numbers for OCZ vertex 4. Some comment from OCZ Ryder?

Yes I am confused. Some numbers are very good for Vertex 4. Asssd etc a d other tests is just bad. Like seq read at low QD.

[johnw]

other tests is just bad. Like seq read at low QD.

That is the main problem right there. If you remember, Crucial came out with a FW update (0009) a while ago that increased sequential read speeds. Also, Plextor FW 1.08 for the M2P increased sequential read speeds (the M3P already had that tweak when it came out). The Intel 510 was tuned for good sequential reads from the start.

All of the other Marvell controller SSDs seem to have eventually tweaked the FW to have good sequential read performance at low QDs and low block sizes.

But the Vertex 4 was rushed, so I guess they did not tweak for high sequential read speeds. Or maybe OCZ's people do not know how to, or even intentionally decided not to because there might have been a tradeoff with the high QD IOPS. I'd gladly trade a bit of high QD IOPS for better low QD read performance. But OCZ has always gone for hype over substance. Maybe they just like touting the high IOPS numbers at high QD.

[Christopher]

That is not accurate according to OCZ. We were told the 512GB gets 1GB, 256GB gets 512MB and 128GB gets 256MB.

All of the initial units of all sizes ship with 1GB of dram. It's not all being used in the 128gb V4, but it dos have it. The Plextors are using 512mb for all sizes, but the 128gb only has half enabled in the FW.

[Morgoth]

The Toggle Mode NAND version (max iops again maybe?) should be interesting~!

[onethreehill]

Vertex 4 256GB is better than Octane 256GB, no doubt.

5 years warranty is always good.

Waiting for street prices after 1/2 weeks.

In stock @newegg

[Royteane]

Quick question,
Vertex 4 or M3 Pro?

[canthearu]

M3 Pro would be my answer. At least until the Vertex 4 has had it's firmware tuneup and establishes a reasonable history of reliability.

[johnw]

M3 Pro would be my answer. At least until the Vertex 4 has had it's firmware tuneup and establishes a reasonable history of reliability.

I also think M3Pro is the one to buy, even if the Vertex 4 were from a company other than OCZ (I will never buy an SSD from OCZ).

By the time the Vertex 4 firmware is fixed to give decent low QD sequential read performance (if it ever is) and the V4 has been shown to be free from major problems for several months, I bet Plextor releases a new SSD model using the Marvell 88SS9187 controller, and that would be a better choice than the Vertex 4.

Today: Plextor M3P

Future: whatever Plextor releases after the M3P

[Ao1]

+1. Being a guinea pig is hard to justify on an untried product from a company with a track record of putting out half-baked products, unless the drive is faster than anything else on the market. There are faster drives out there and even faster drives will soon be available.

Strange how the "industry standard" ATTO benechmark no longer seem to be in favour



http://www.ocztechnologyforum.com/fo...l=1#post722565
http://www.xtremesystems.org/forums/...=1#post5081320

[User9498]

+1. Being a guinea pig is hard to justify on an untried product from a company with a track record of putting out half-baked products, unless the drive is faster than anything else on the market. There are faster drives out there and even faster drives will soon be available.

Strange how the "industry standard" ATTO benechmark no longer seem to be in favour

Funny how AS SSD and incompressible data suddenly has become important to OCZ, the V4 is certainly tuned for it. I guess they also have permission to talk down Sandforce and the V3.

Except for the Vantage numbers OCZ does have an interesting drive, competition is always a good thing. I wouldn't mind seeing the Plextor M3P real world pitted against the V4, similar to what Tony did below.

Vertex3 vs Vertex4 Load Time Comparison: WOW
http://www.youtube.com/watch?v=YsGx6ttaou4

Vertex3 vs Vertex4 Load Time Comparison: StarCraft 2
http://www.youtube.com/watch?v=OkD3yQODMSw

[canthearu]

+1. Being a guinea pig is hard to justify on an untried product from a company with a track record of putting out half-baked products, unless the drive is faster than anything else on the market. There are faster drives out there and even faster drives will soon be available.

Yep, there is plenty of evidence that the V4 is half baked still.

If only they could release a product without glaring flaws that doesn't need a firmware update right out the gate, it would show a lot of people they were serious about fixing their image.

[Ao1]

Funny how AS SSD and incompressible data suddenly has become important to OCZ, the V4 is certainly tuned for it. I guess they also have permission to talk down Sandforce and the V3.

Except for the Vantage numbers OCZ does have an interesting drive, competition is always a good thing. I wouldn't mind seeing the Plextor M3P real world pitted against the V4, similar to what Tony did below.

Vertex3 vs Vertex4 Load Time Comparison: WOW
http://www.youtube.com/watch?v=YsGx6ttaou4

Vertex3 vs Vertex4 Load Time Comparison: StarCraft 2
http://www.youtube.com/watch?v=OkD3yQODMSw

The Tech Report review did not show any advantages for game loading. I suspect that the game loading comparison on the OCZ forum was achieved with pre-fetched reads via the on-board DDR3 RAM. There is nothing in the benchmarks results from the review sites that would explain why the V4 is “twice as fast” at loading games compared to the V3. (Quote from Tony’s post linked below). If anything benchmarks indicate that game loading would be slower and this appears to be what Tech Report found. Something else is at play......

Tech Report real-world applications:

http://techreport.com/articles.x/22736/10

The V4 does not show any sign of being faster at boot up or with game loads. In fact it is coming out slower than the M4 & 520 & 830.

Guru3D real-world applications

http://www.guru3d.com/article/ocz-vertex-4-ssd-review/8

AvP load time is equal to the 520, 320 and P256 Pro.

It’s a real shame that none of the reviews to date have experimented with the pre-fetch capability, when it appears that it plays a significant role.

According to Tech Report:

“Shortly after we received our Vertex 4 sample, OCZ warned us that benchmarks run on unpartitioned drives might not show drive's performance in the best light. The Vertex 4 has a prefetch mechanism that only works on partitioned drives, but we couldn't pry more details from the company about what's going on under the hood. We do know that this prefetch functionality isn't part of the Octane. For what it's worth, running HD Tune on a formatted partition failed to produce higher read speeds on the Vertex 4”.

It’s hard to understand what Tony is saying here:

“Games tested were World Of Warcraft and Starcraft2, a simple game saved game load was captured to a seperate SSD drive so we could see the difference, then the captures were edited side by side so there is no mistake which is faster”.

The game was pre-loaded and then the reload time was compared? Does that mean that the game was pre-loaded on the onboard DDR3 RAM and then run straight from the DDR3 RAM? Nothing wrong with cache and if this is what the V4 is doing, It’s an interesting development for SSD’s, but its also disingenuous, as you will only see advantages in certain scenarios. In other scenarios you might well see a disadvantage.

It would be great if anyone with a V4 could try playing with the on board DDR3 cache to see how the pre-fetch works.

[User9498]

It would be great if anyone with a V4 could try playing with the on board DDR3 cache to see how the pre-fetch works.

That would be a great experiment! The video almost gets one thinking that perhaps synthetics don't tell the entire story. However, compression hocus-pocus has simply been replaced with pre-fetch, and perhaps being presented in a disingenuous way. Thanks for setting things straight.

I've now caught myself up on the reviews, including Wendy's, where she compares the 512GB V4 with a field of much smaller drive$. Of course, it's the first link on OCZ's review page.

The V4 still looks impressive but the new Marvell appears more like the one to own.

[WhiteFireDragon]

Geez a vertex 4 already? I think the hype for vertex 3 just died and now 4 came out. So vertex 5 in a few months?

[canthearu]

Is there a utility from OCZ used to pre-fetch into the drive's DDR3?

Seems a bit strange because if you start a game, exit it and start it up again, all the data won't be loaded from the drive's cache ... because it is already sitting in the windows page cache.

If OCZ is measuring the second start of a game on the Vertex 4 vs the first start of a game on the vertex 3, then that is very wrong, but would explain the dramatic difference. The second game start would run mostly from the windows page cache, not even accessing the SSD.

[Computurd]

in my testing with consecutive game level loadings, the reloads are way faster, due to caching not only in the OS, but also on the GPU. if you monitor the free GDDR space on your GPU using AIDA 64 as you load the level, you can watch it fill up. hell, i watch it on my keyboard screen (gotta love logitecch).

however, once the level is exited, the GPU stays loaded with data. Reload goes BAM!!! most of the wait is for the data to get into GDDR.

too many intangibles for apples to apples with the consecutive game loadings. you have to exit the game, then reload the game to clear the GDDR.

[Ao1]

Benchmarks do not indicate performance levels that would result in a game loading that much faster, so something else is at play. Either OCZ flat out cheated or the on-board cache had something to do with it.

I noticed that the V4 specs seem to be based on different benchmark specifications when compared to the V3. It seems strange that the max V4 IOP spec is achieved with 512 bytes RR.

Why would 512 bytes RR be so much faster than 4K RR?

Why are details of the LBA range provided for the V3 specs but not for the V4?

Why does performance drop on an un-partitioned drive?

Vertex 3
Maximum Sequential Speeds are determined using ATTO
Small file I/O performance is measured using Iometer2008, Queue Depth 32, 4KB Aligned; Logical Block Address (LBA) range: 85% of total drive capacity
Small file I/O performance is measured using Iometer2008, Queue Depth 32, 4KB Aligned; Logical Block Address (LBA) range: 8GB
Incompressible data performance is measured using AS-SSD; 64 thread for 4K specs

Vertex 4
Maximum sequential speeds are determined using ATTO
Small file I/O performance is measured using Iometer 2010 (1.1.0 rc1)
Maximum I/O performance is measured using Iometer 2010, 512 bytes Random Read

[Nizzen]

Ao1:
You can ask OCZ on the OCZ forums.
I have spammed to much there already :p
http://www.ocztechnologyforum.com/fo...d=1#post722642

[canthearu]

Lol, thats like trying to convince christians that god doesn't exist ... while at a church. And probably with the same nasty consquences.

BTW, the latency figures of the IBM card were perfectly fine .... 0.2ms latency is about right for sandforce drives in steady state.

[Wendy]

I've now caught myself up on the reviews, including Wendy's, where she compares the 512GB V4 with a field of much smaller drive$. Of course, it's the first link on OCZ's review page.

Reviewers are always sent the fastest drives available. For the OCZ Vertex 4 and Octane, that happens to be the 512GB versions.
For SandForce SF-2281 based SSDs 240GB are the fastest, OCZ sent the 240GB Vertex 3, and Intel sent the 240GB 520, the 480GB versions are a good deal slower.
For Marvell, again they sent the fastest, the 256GB Crucial M4, and Plextor sent the 256GB PX-256M3.

So reviewers are comparing the fastest against the fastest, regardless of capacity. It's not exactly comparing apples to apples, but that's how it goes.

[johnw]

So reviewers are comparing the fastest against the fastest, regardless of capacity. It's not exactly comparing apples to apples, but that's how it goes.

Not to single you out, but that is how it goes because reviewers value saving a few bucks over testing the models that most people buy.

It would not cost very much for reviewers to buy the SSDs themselves, and then for the models that they don't want to keep, sell them on a FS forum. It should be possible to recoup almost all of the cost, since some people would actually like to buy an SSD that has already been thoroughly tested and found not to have any factory defects. If the warranty is not transferrable, the reviewer could offer to handle any warranty returns themselves, which would have the added bonus of letting the reviewer see various long term failure modes that are never seen with review units that are only in hands for a week.

[Ao1]

Wendy could you do an IOmeter 300 second run on an un-partitioned drive with 512 bytes (100%) random read and 4K (100%) random read at QD1 using a test file that fills the drive and then do the same with a partitioned drive?

Below is what I get on drives that are all in a steady state using the above test procedure on partitioned drives.



I estimated latency using this:

Queue depth*(1/latency in ms) = IOPs

Edit, to be really cheeky could you also run an ASU benchmark with a test file size of 32GB? (On both a partitioned drive and an un-partitioned drive).

[Wendy]

Wendy could you do an IOmeter 300 second run on an un-partitioned drive with 512 bytes (100%) random read and 4K (100%) random read at QD1 using a test file that fills the drive and then do the same with a partitioned drive?

Below is what I get on drives that are all in a steady state using the above test procedure on partitioned drives.



I estimated latency using this:

Queue depth*(1/latency in ms) = IOPs

Edit, to be really cheeky could you also run an ASU benchmark with a test file size of 32GB? (On both a partitioned drive and an un-partitioned drive).

I can't at the moment as the Vertex 4 is on the Analyzer running a stability soak test, and will be there for another 50+ hours. I also don't own the analyzer, so I don't have the drive at home at the moment.

@John
Regarding buying smaller SSDs to test. It may surprise you to learn that most reviewers are unpaid for reviewing. I'm one of them, and on a school teachers pay I certainly can't afford to lay out that kind of cash.

[johnw]

http://www.anandtech.com/show/5741/o...-based-silicon

I received a tip yesterday with proof that OCZ's Everest controller was actually a Marvell 88SS9174 controller (the same controller used in Crucial's m4, Intel's SSD 510, etc...) with a custom Indilinx firmware. After a bit of digging, it turns out that this is indeed the case. Although OCZ is working on non-Marvell based solutions, the Everest 1 (Octane) and indeed the Everest 2 (Vertex 4) are both based on Marvell hardware.