OCZ Vertex 2 Pro, Sandforce Powered SSD Preview

0 thumbs up

To look at the new Vertex 2 Pro SSD from OCZ Technologies, is like looking at most standard SSDs on the market these days.  Though Intel used a thinner metal casing with plastic stiffener brackets on its line of product, this OCZ Vertex 2 Pro has a straight-forward metal case like 99% of the SSDs on the market today, other than Intel's.

 


OCZ Vertex 2 Pro Series 100GB SSD - click for full res

The real secret sauce of the Vertex 2 Pro can be seen under the hood. This SSD is comprised of 64Gb Micron NAND chips that operate at 166 MT/s.  There are 16 of these chips on-board, 8 each on the top and bottom sides of the PCB.  Other than the small 8-pin configuration EEPROM and one tiny 6-pin logic device, the Sandforce 1500 controller is the only other piece of pure CMOS silicon on board.  The rest of the bill of materials consists of passive or electromechanical components (connectors etc).  What really sets the Vertex 2 Pro apart from many current SSDs on the market is its Sandforce 1500 NAND Flash controller.  Let's drop down for a look at this relatively new technology for SSDs.


Sandforce 1500 Series SSD Processor Block Diagram

That Sandforce 1500 series controller is a rather complex device and though we won't dive too deeply into the architecture, in consideration of your eyes possibly glossing over, we will point out a couple of key characteristics of Sandforce technology that sets it apart from other SSD NAND Flash controllers on the market.  As you can see, this is a complex system on a chip of sorts that has a number of functional blocks beyond its Flash and SATA interfaces and processor core.  Some of the microengines here that merit discussion are the DuraWrite and RAISE error correction blocks. 

DuraWrite and AES 128 Encryption Technology -
"Write amplifcation" is a term that is used to describe what a typical SSD has to go through in order to write data to a drive. In general, the SSD has to read a 128K block of data into system memory, modify the content that needs to be written to that block and then write that entire 128K block back to the drive.  This scenario holds true even for something as small as 4KB of data.  This read-modify-write scenario takes its toll on the write endurance of the average SSD reducing its functional life dramatically.  Typical write amplification for legacy SSDs has been in the neighborhood of up to 20X, which is a ton of wasted write operations and the reason why elaborate "garbage collection" algorithms have been developed to clean up data on SSDs after use so that they maintain performance. 

 
Image courtesy:  Sandforce, Inc.

Intel claims their controller offers a 1.1X write amplification factor, while Sandforce claims their DuraWrite technology brings that down to .5X or less than half that of Intel.  This also negates the need for on-board DRAM cache for a Sandforce-based SSD, because caching write data in order to alleviate amplification is no longer required.  Finally, Sandforce claims that write endurance of an SSD based on their controller will have SLC-like endurance of up to 5 years, once again due to their proprietary technology that minimizes write amplification dramatically.  We should also note that Sandforce-based SSDs like OCZ's Vertex 2 Pro that we're previewing for you here today, also support the TRIM functionality offered in Windows 7.

Finally, you'll also note that there is an encryption engine block detailed in the block diagram above.  Essentially, data is encrypted and decrypted on the fly, making it more secure from hackers, versus the simple password protection techniques on other SSDs.  This engine, Sandforce reports, also completely offloads the host processor of any workload related to its security algorithm and we're sure is combined with some sort of compression algorithm (like Deflate for example) which also keeps data transfer rates up.


RAISE Technology -
RAISE is an acroym that Sandforce coined that stands for "Redundant Array of Silicon Elements".  Essentially, the technology, in combination with Sandforce error detection and correction (ECC) algorithms, combines RAID-like redundancy without the requirement of writing data twice to the drive.  The company doesn't explain how this is achieved but they do go on to claim that "RAISE technology reduces the probability of a single unrecoverable read error by 100 times to 0.001%. Applying that same formula, the failure rate of the SSD drops from 12.0% to a mere 0.13%, nearly 100 times lower. 

In short, though it's a bit of a mystery how they achieve this, once again with these proprietary technologies from Sandforce, MLC-based NAND Flash SSD reliability is increased by orders of magnitude supposedly.  We'll touch on this again later...

Article Index:

0
+ -

This sector is moving so fast, there seems to be new which impacts it almost daily in one way or another. For a data center these things are almost a god send. For a normal user as well as a hardware junkie, performance gamer etc. they are good editions, and getting better all the time.

I personally have not taken the plunge and gotten one (I also don't have the cash to get a performance model), but would very much like to. Having one of these for an OS and for reactive software such as games, with one of these 1-2TB beauties for a storage drive, seems to be the best setup. However just using these type of drives in multiple and or raid configs is also awesome.

Maybe I will have the money to get one some time. I think it will mostly be one of these OCZ units, they seem to be the fastest as well as most successful (in operational hardware, and of course transfer ratios now as well), names in this market to me.

I wonder when then new ultra low Nm specifications will start affecting this market as well.

0
+ -

Totally agree rapid1. I do believe with the shrinkage that Intel/Micron just achieved will help out the SSD market and give us larger drives for less dollar.

I also have not made the jump to SSD but I am excited when I can. The speed these things offer will take all PC use to all new levels.

I found this from another site that might help some people pick out drives and give some info on SSD.

http://www.tweaktown.com/guides/3116/tweaktown_s_solid_state_drive_optimization_guide/index.html

0
+ -

I have to say that after I installed my SSD Raid in my PC (ran it for a few days before), the change was dramatic in several areas.  I mean, my PC is rather crazy to begin with, but it is all the little things that the SSD does that makes it worth it.

However, since I did drop a lot of money into my drives already it will be a while before I think of upgrading.

0
+ -

What seems to be the best strategy with SSD's is to just wait for the market to calm down a bit. Like Marco said, they have a strangely large selection of drives, which makes me think they are still figuring it out. I think a safe standpoint is to wait a couple months before moving in on one of these drives.

0
+ -

To be honest... i'm a bit dissappointed in the IOPS on this. It didn't even come close to the 19,000 they claim. In some cases it was only slightly better than the Vertex drive. I'm hoping it is just some bugs they need to work out. I wouldn't say this put them "squarely in the lead" as OCZ claims. It has it's benefits and drawbacks compaired to the Intel drives.

Still... that being said. It's a pretty solid drive. Hopefully the price is reasonable.

0
+ -

Acarzt, what are your thoughts on Iometer in general? We're not all that thrilled with it in terms of its relevance to real world performance. Look at virtually all the other tests we ran and the Vertex 2 Pro is stronger, especially versus what Iometer shows. We're considering dropping it (Iometer) all together actually... thoughts?

0
+ -

Well, as far as getting an accurate count of IOPS... i'm not sure how true the numbers really are. But I do think it is a good gauge of how well the drive will perform in the real world.

It's really hard to say for sure if that test would really even be valid or worth using anyway because in the real world i think it would be quite difficult to find any program and usage that would actually use such an large amount of IOPS.

Considering HDDs have IOPS in the hundreds vs SSDs have IOPS in the Thousands... You would really only see the difference in a Server environment. I don't think applications are really tapping into the IO potential of SSDs. The proof of this is in the fact that load times in games don't really change. I think if games took more advantage or this they could get better load times.

I'm no programmer... but if I were to take a shot at how a game loads... they seam to read segments peice by peice... like textures, models, etc. in a certain order. If they were to load these things parrallel, or all at once and really take advantage of the IO power I think they could tap into more of the bandwidth available on these SSDs. But they are more geared towards HDDs and their lower IO.

With everything going multi-this and multi-that though, I think it's only a matter of time before SSDs can have that kind of effect. But in current usage I don't think any users really tap into the IO power they have. A server environment is where they will really shine where you have hundreds of users accessing the same data.

Then again my understanding of IO could be all wrong here :-P

0
+ -

Great article Dave!

I was very impressed with the CrystalDiskMark benchmarks. I have to agree with rapid1, while I'm loving the performance increases in the SSD market, the price and size/density progression has been a little slow.

Especially interesting was the section of Durawrite. I'm a little confused by this. I assumed 1x would be limit for write amplification so Intel's 1.1x seemed pretty well optimized. With the 0.5x amplification, are we saying that 10GB of information will only occupy half as much on the actual SSD? Are we talking about some sort of data compression?

0
+ -

Essentially Write amplification writes the data at a larger size to write it faster. For example 4kb of data would be written to an entire block as 128kb. This is an amlification of 32x. Datawritten / Data Size = Aplification.... 128/4 = 32. So then the reverse would be true Amplification * Data Size = Data written. So for Intel this would be 1.1 * 4k = 4.4k. So .5 * 4k = 2k...... Which shouldn't be possible... So there is a big question of how, and this is not really explained. 1 theory I have is that the data is held in cache until 2 blocks can be filled in a single pass instead of needing to do multiple passes.

This would explain why it performs so well on CrystalDiskMark at 4k and why you see the sea-saw effect in sisoft.

BUT, if this IS true then it would make the potential for data corruption higher due to a power failure because data will be held in cache slightly longer and this may be why response times also went higher(poorer) in IOMeter when more writes were involved.

Login or Register to Comment
Post a Comment
Username:   Password: