Vertex 3 and +500mb/s SSDs where?

provoko

Gawd
Joined
Aug 6, 2004
Messages
656
As an Amazon Associate, HardForum may earn from qualifying purchases.
The Vertex 3 can write at 200MB/s not 500MB/s.

The 500MB/s+ scores are only achieved using compression and writing zeroes. If you write real data it would go at much lower rate. Be very careful about Sandforce SSDs as all the specs are for zero write only!

Same with Vertex 2 that lists as 285MB/s write, while some versions only can write at 35MB/s (25nm 60GB version).
 
The Vertex 3 can write at 200MB/s not 500MB/s.

The 500MB/s+ scores are only achieved using compression and writing zeroes. If you write real data it would go at much lower rate. Be very careful about Sandforce SSDs as all the specs are for zero write only!

Same with Vertex 2 that lists as 285MB/s write, while some versions only can write at 35MB/s (25nm 60GB version).

That's not true, real world benchmarks are out already which include: Playing movies, downloading movies, streaming, game loading, browsing, etc:
http://www.storagereview.com/ocz_vertex_3_review_240gb

The results are 450mb/s.
 
ocz_vertex3_240gb_2mb_sequentialtransfer_4k.png


These results are false, because IOmeter is not patched to use incompressible data instead of zeroes. By default, IOmeter writes zeroes just like ATTO does and as such you are only limited by the speed of which the SSD controller can compress data; not the speed of the NAND!

You should look for reviews from more reputable sources who have experience with benchmarking solid state devices. The real write speeds of Vertex 3 that i have seen are between 150MB/s and 250MB/s depending on the size of the SSD.
 
So Vertex 3 256 is basically the champ?

Where do we buy this damn thing is what I really want to know like OP.
 
Last edited:
Vertex 3 is based on the new third generation SSD controllers. You should compare Vertex 3 against Intel G3 and Marvell C400; both are not yet sold, but should be available within a few months.

So while Vertex 3 will be here rather soon, don't consider it the 'champ' so quickly; possibly Intel G3 is a very strong contender due to leaked specs of 160MB/s random write, which is a considerable improvement over current performance specs. Intel G3 also should have a SuperCapacitor, while for Vertex 3 only the Pro version has one. This protects against corruption and sudden death on an unluckily timed power failure.
 
How long am I gonna have to wait? Starting to put a computer together now.
 
What are you building? Sandy Bridge has the bug, Bulldozer isn't out.. it's time to wait pal! Buy a new computer around July. New third generation SSD will be readily available then, plus AMD bulldozer which should contend very well against Sandy Bridge.

If you don't want to wait that long, you can buy a simple Corsair F60 and call it a day, buy a faster and bigger SSD at a future time. I wouldn't buy an expensive SSD right now, with the third generation controllers around the corner.
 
So Vertex 3 256 is basically the champ?

For incompressible data sequential write speed, no, the 250GB Intel 510 is faster (313 MB/s).

http://www.anandtech.com/show/4202/the-intel-ssd-510-review/9

35684.png


Also, if you do decide to go with a Sandforce 2XXX series SSD, I would avoid OCZ. They have some shady business practices. Other more reputable companies, like G.Skill and Corsair, should release SSDs with similar peformance to the Vertex 3 in the next month or two.

For those who were confused about the sequential write speed of the Sandforce 2XXX drives (like Vertex 3), a good site to check out is bit-tech. You can see that the preview Vertex 3 they tested gets 284 MB/s sequential write with AS-SSD, but drops to 220 MB/s after the drive has been heavily used. That is a common issue with Sandforce SSDs, but the 2XXX series of Sandforce controllers is faster than the 1XXX series, so instead of dropping to well below 100 MB/s after heavy use (1XXX series), it drops into the low two-hundreds (2XXX series).

http://www.bit-tech.net/hardware/2011/02/24/ocz-vertex-3-ssd-preview/3
 
Last edited:
Really available or just listed? Initially i read that B3 corrected motherboards would be available end of March - begin April, so it's perfectlty plausible. I also heard motherboard manufacturers will be putting B3-stickers on the motherboard boxes to distinguish them from non-corrected B2 Cougar Point revisions.

Regardless, it does appear that both the new CPU platforms (Sandy Bridge + Bulldozer) as well as the new SSD offerings (Sandforce SF2000 family, Marvell C400, Intel G3) are available in a few months. Bulldozer probably will be a bit later around July as i've heard, but could be August too before the products are readily available; we know how this works. :)

Anyway, i'm excited about both new platforms (third generation SSDs and next-gen CPUs from Intel/AMD). Would be great if you could pair both of them, that's why perhaps it would be good to wait just a bit if you don't really need to build a computer right now.
 
I picked up a B3 mATX H67 board for a HTPC build from newegg last week, so they are available (it's humming right along now with an i3-2100). If you want a particular board you may have a more difficult time, and the availability on the new SB Xeons/boards is wierd - they were available, then weren't.
 
Stop thread jacking. Go to MOTHERBOARDS subforum and make your own 1155 mobo thread.

As for the link I submitted, sub.mesa, you failed to scroll to real world benchmarks.
GIoPV.png


Storagereview.com has a very good reputation.

But you know what, all of this babbling doesn't answer the question, when are these drives coming out?
 
Storagereview.com has a very good reputation.

It may have a good reputation among the ignorant, since they show lots of pretty charts and act like they know what they are talking about.

But their SSD reviews are TERRIBLE. They run IOMeter without the random data setting on Sandforce SSDs, and they even go so far as to test the SSD loaded power consumption under write with IOMeter writing repetitive data, then exclaim that the Sandforce drives have low power usage under write! That is just pathetic.

And their "real world" traces are hardly credible, without telling us exactly what is being written to disk, how many MBs, and how compressible it is. Judging from their other "tests", this one is not credible at all.
 
I raised this issue on the forums and it seems the guys are open to improvements, and will be changing their testing methods:
http://forums.storagereview.com/ind...ew-discussion/page__view__findpost__p__268065

The sad thing is, that it's hard to contradict users who see professional-looking benchmark graphs and conclude these are more trustworthy than technically correct advice on forums like these. Provoko, probably the best thing you can do is compare as many reviews as you can, and try to see common patterns which multiple reviews share. That would be much safer if you cannot judge the validity of the tests yourself.
 
Here's another thread where they say that they are "working on it". Hopefully they get it ready soon.

http://forums.storagereview.com/index.php/topic/29665-updated-power-benchmarks/#entry267641

I wish every SSD review would start with an AS-SSD screenshot of the drive. Not because AS-SSD is the be all and end all of benchmarks, but because it is so simple and widespread. If the values in AS-SSD look significantly different than others, then you have to think that the review is suspect since they probably have not configured things correctly, or have incompatible hardware. If the AS-SSD screenshot looks okay, then you can go on and read the rest of the review, with a bit of confidence that at least the SSD is configured reasonably and hardware is compatible.
 
FYI, the V3's were supposed to hit Monday, but OCZ said they have a short delay and will be available end of next week.
Also, that heavily used "slowdown" on the SF drives is by design.
It's called throttling. If the drive sees too much data being written in too short amount of time it will throttle to preserve wear.
These review sites didn't know this and run the benches and within minutes have invoked a throttle state that can take a while to reset.
CDM can write over 100Gb in 10m while it's benching - that's not normal usage.
Posted via Mobile Device
 
Also, that heavily used "slowdown" on the SF drives is by design.
It's called throttling. If the drive sees too much data being written in too short amount of time it will throttle to preserve wear.
These review sites didn't know this and run the benches and within minutes have invoked a throttle state that can take a while to reset.
CDM can write over 100Gb in 10m while it's benching - that's not normal usage.
Posted via Mobile Device

How does this help preserve the drives? If I am going to write 100Gb of data I am going to generate the same amount of writes (and same amount of wear) whether that happens in 10 minutes or 12 minutes (cause that's about the order of slowdown we are seeing)?

I'm just a bit dubious here.
 
It's all explained in a thread on OCZ's site in the SF section. No sense regurgitating it all here...
Posted via Mobile Device
 
How does this help preserve the drives? If I am going to write 100Gb of data I am going to generate the same amount of writes (and same amount of wear) whether that happens in 10 minutes or 12 minutes (cause that's about the order of slowdown we are seeing)?

I'm just a bit dubious here.

You should be dubious. It is just BS from OCZ and Sandforce, claiming that a problem with their implementation is actually a feature, when anyone with any common sense can see that it is a bug, or at best a known flaw that Sandforce cannot correct without major design changes.

If it were really a feature, then they would specify that their throttling feature will take effect if you write X amount of GBs in Y minutes or less, and how much the speed will be throttled. They would also offer an option to turn off this dubious feature. None of this is the case. It is obviously a flaw, not a feature.

It amazes me that there are people that actually fall for OCZ and Sandforce's BS, especially after all of the obfuscation from them in the past.
 
It's all explained in a thread on OCZ's site in the SF section. No sense regurgitating it all here...
Posted via Mobile Device

Could you direct me to the thread - I took a quick peak over there, but, well, I don't really want to surf their entire forum to find the technical information on this. Thanks!

john4200 said:
It is just BS from OCZ and Sandforce

Yeah, that's what I think also, but I'm still interested in what their technical explanation is :).

It's probably pretty true that in normal use you wouldn't see this type of random transfer (maybe sequential, though given the size of drives maybe only with imaging) - but that would still be filed under "flaw that happens in corner case" as opposed to "feature" in my book. And I haven't seen any information establishing what the thresholds for this "flaw/feature" are - it could very well be something that happens if you are using a SSD as a scratch drive for video processing, etc. - which is something the consumer should be made aware of.
 
It's all explained in a thread on OCZ's site in the SF section. No sense regurgitating it all here...
Posted via Mobile Device


Okay, I assume you mean:
This Post

Which has such usage guidelines as:
OCZ Sandforce Usage Guide said:
Specific to SF:
Restrict un-compressible data pattern based benchmarking:
AS_SSD = one full benchmark on fresh install or restore, there-in every 1-3 weeks Sequential ONLY.
CDM = one full benchmark on fresh install or restore, there-in every 1-3 weeks 50-100MiB settings, 2 rep, Random/QD ONLY

So I now have to be "Carefull" how I use my drive so I don't slow it down? If I'm going to use the drive as a high speed scratch for video I can only process 50Gb of video every 1-3 weeks?

If I update the drivers for my SATA controller I have to wait 3 weeks between back to back benchmarks to see if I got a performance increase?

And from the quoted thread, while they mention "protect the nand", they don't every explain any viable mechanism for "protecting the nand". Instead it looks like the issue is the firmware on the drive can't keep up with the large amount of sector usage generated with large amounts of uncompressible data (relevant since compressible data maps to a smaller size - though I would expect you would have the same issue if the amount of data you moved was multiplied by the compression factor).

Basically there seems to be a performance window on the amount of blocks that can be recycled, and the drive firmware/controller hits it's limit if you move a bunch of data? (A limit independent of interface raw write/read speeds).

So it looks to me like the slowdown is a real and valid thing, doesn't have anythign to do with "protecting the nand", and may or may not impact you depending on your usage limits.

Obligatory car analogy:
Like having an electric car with a 75 mile range, except not telling anyone what the range is and blaming the user when they go over it.....
 
Tired of waiting. I don't see a reason to wait for Bulldozer neither from a price nor performance standpoint especially considering you could be waiting another 6 months. Need 3rd generation SSD NOW.

I read everything you guys said and I'm think I'm going to get a Vertex 3 anyways. Just buying it for fast loads in games anyways and I don't want to have to worry about the Marvel micro-controller. Don't think the throttling will affect me and just a little concerned over OCZ quality control from another thread.
 
So it looks to me like the slowdown is a real and valid thing, doesn't have anythign to do with "protecting the nand", and may or may not impact you depending on your usage limits.

Welcome to the real world. :D

I've had the Vertex 2s and sold them in a week.

They're not bad drives. Just average drives that pale when compared to the best.

And they're great for those GB/$ buyers.
 
ARGH!!!!

I'm trying to help a friend put together a system and this is such a mess.

Does the intel gen 2 drives have the same "throttling" problems the vertex 2 did?
 
IMO right now the drive to get is the Intel 510. Micron C300 is also a solid performer. If I were buying one today it would be one of those two.
 
Ordered the 240GB! Holy shit I'm excited! :D

IMO right now the drive to get is the Intel 510. Micron C300 is also a solid performer. If I were buying one today it would be one of those two.

The Intel 510 has sequential performance, but that's about it. It's beaten by the Vertex 3 in every single area except for purely random writes (and even then, it's close). Since it doesn't even use an Intel controller, I wouldn't even count on it having the reliability of the G2. Furthermore, performance degradation with new gen SSDs is a no no.

IMO:

If you want a small OS drive (~60-80GB): Intel G3
If you want a large OS + games drive(~120-240GB): Vertex 3, and maybe the Corsair m4 if sequential performance isn't as important to you
If you want a moderately sized drive (~120GB) and cost is an issue: Intel G3, or discounted last gen drive (Crucial C300, or Vertex 2)
 
I was really set on getting the C400, but it won't be out until the 26th and the Micron OEM version sold out when I placed the order on SuperBiiz.
Amazon has a couple of weeks waiting period (Vertex 3), so I pulled the trigger on Newegg.

It'll be my first SSD.
Can't wait!
 
If it were really a feature, then they would specify that their throttling feature will take effect if you write X amount of GBs in Y minutes or less, and how much the speed will be throttled. They would also offer an option to turn off this dubious feature. None of this is the case. It is obviously a flaw, not a feature.
How would that be possible? The SF controller's ability to avoid write amplification is dependent on the premise that data is compressible. If you do garbage collection to re-arrange the pages after a bunch of incompressible data is written, you're still not avoiding the basic problem of too many writes to the flash.

SF only has a few kb of cache so unlike Crucial/Intel, it can't predict the kind of data that is to be written. This has both advantages and disadvantages. It's neither a bug nor a feature...just a limitation of this design, which seems unlikely to affect most people during normal use of the drive.
 
How would that be possible? The SF controller's ability to avoid write amplification is dependent on the premise that data is compressible. If you do garbage collection to re-arrange the pages after a bunch of incompressible data is written, you're still not avoiding the basic problem of too many writes to the flash.

SF only has a few kb of cache so unlike Crucial/Intel, it can't predict the kind of data that is to be written. This has both advantages and disadvantages. It's neither a bug nor a feature...just a limitation of this design, which seems unlikely to affect most people during normal use of the drive.

You make no sense. Most of what you ahve written is irrelevant to the slow-write flaw of the Sandforce SSDs.

As I wrote, it is obviously a flaw. It is also a bug, since the flaw is not documented. You write that the flaw is a limitation of Sandforce's design. While that may be true, it is nevertheless a flaw. A design flaw is still a flaw.

And it is absurd to say it is unlikely to affect most people. Anyone who runs CDM a time or two will find their SSD writes going much slower for days.
 
You make no sense. Most of what you ahve written is irrelevant to the slow-write flaw of the Sandforce SSDs.

As I wrote, it is obviously a flaw. It is also a bug, since the flaw is not documented. You write that the flaw is a limitation of Sandforce's design. While that may be true, it is nevertheless a flaw. A design flaw is still a flaw.

And it is absurd to say it is unlikely to affect most people. Anyone who runs CDM a time or two will find their SSD writes going much slower for days.

This is not the case with the SF2xxx controllers. (i.e. Vertex 3's)
While the throttle is still there supposedly, it takes much heavier writes to kick off any kind of throttle.

I've been trying and can't get 'em to slow down. This is what was found on OCZ forums as well.

Don't assume the new SF controllers act the same as the old - they don't and I've got first-hand experience with it which confirms what others have said as well.
 
This is not the case with the SF2xxx controllers. (i.e. Vertex 3's)
While the throttle is still there supposedly, it takes much heavier writes to kick off any kind of throttle.

I've been trying and can't get 'em to slow down. This is what was found on OCZ forums as well.

Don't assume the new SF controllers act the same as the old - they don't and I've got first-hand experience with it which confirms what others have said as well.

As usual, you don't know what you are talking about. I am not assuming anything, I have looked at actual measurements on the drives. As usual, bit-tech has good data on this -- they looked at a pre-release 240GB Vertex 3 some time ago.

http://www.bit-tech.net/hardware/2011/02/24/ocz-vertex-3-ssd-preview/3

Also, just today Anand looked at a 120GB release Vertex 3, and he had it slow down worst case to less than 40 MB/s sequential write, and in two other AS-SSD runs it slowed down to 100 MB/s and 130 MB/s.

http://www.anandtech.com/show/4256/the-ocz-vertex-3-review-120gb/13

It looks like the drop in incompressible data sequential write speed, after TRIM and some time for the GC to work, is from 280 MB/s to 220 MB/s for the 240GB pre-release Vertex 3, and from 160 MB/s to 130 MB/s for a 120GB release Vertex 3, although the speed can drop even lower for a short period of time, and the speed can drop a lot lower if you don't have TRIM.

Don't kid yourself. The Sandforce write speed flaw is definitely still there.
 
Last edited:
I thought the interesting part of the review had nothing to do with the actual drive's test.

The controllers' firmware isn't finalized.....
The shipping Vertex 3s are using RC firmware from SandForce, the MP label can't be assigned to anything that hasn't completely gone through SandForce's validation suite. However, SF assured me that there are no known issues that would preclude the Vertex 3 from being released today

Unless you're buying a Vertex 2.25 (80GB, 160GB, 200GB) and Vertex 2.34 (60GB, 120GB, 240GB) a regular Vertex2's NAND will be a crapshoot.....
These drives will only use IMFT NAND - Hynix is out. The idea is that you should expect all Vertex 2.25 drives to perform the same at the same capacity point, and all Vertex 2.34 drives will perform the same at the same capacity as well

And Anand has made a "Truth in SSDs when buying from OCZ" law to make their products more "transparent" for concerned comsumers......

The drive tested as I expected but SF & OCZ still aren't getting any more of my money....maybe in the future but not now.
 
Back
Top