SandForce SF-2000 Enable SSDs with 500 MB/s

[DooRules]

looking very nice...http://news.softpedia.com/news/SandForce-SF-2000-Enable-SSDs-with-500-MB-s-Speeds-159807.shtml

http://www.techpowerup.com/132368/SandForce-Unveils-SF-2000-Series-SSD-Processors-Enabling-Up-to-500-MB-s-over-SATA-6G.html

[SteveRo]

Mr DooRules - nice find - very interesting!

[flamenko]

Had a two hour interview posted below two help the thread out if it is ok with the site. If not, please advise and I will remove the post. I always have the Analyst Brief available.

[SteveRo]

and another - http://www.anandtech.com/show/3971/sandforce-announces-nextgen-ssd-controller-sf2000-capable-of-500mbs-and-60k-iops

[Gilgamesh]

Next week first demostration:
http://www.legitreviews.com/news/9162/

http://legitreviews.com/images/news/2010/owc_mercury.jpg

OWC’s 8x and 16x PCIe SSD cards will be available with up to eight SSDs for capacities up to 3.2TB, offer up to 4GB/s sequential Reads/Writes and up to 480K IOPS, provide 28% over provisioning for data reliability, and double data encryption for security assurance. Exact pricing and model configurations for the OWC Mercury Extreme Pro Enterprise PCIe Solid State Drive (SSD) cards will be announced in early 2011.

Definitely, I can't buy it :(

[alfaunits]

I'm estimating prices at one Lamburghini a piece :D

[Peakr]

I'm estimating prices at one Lamburghini a piece :D

Think I'd take the Lamburghini!

[mak1skav]

Think I'd take the Lamburghini!

Call it Lamborghini and you will have a deal. :p:

[Computurd]

well one caveat here....the sandforce drives arent exactly known for their great steady state performance, especially with uncompressable data. lets think of that internal chip on these new drives, pulling what, two watts? at around two watts how sophisticated can its compression really be? rudimentary at best! especially considering going at 500mb/s!! something seems amiss here....if the power of the compression will be very limited, how does it handle already/uncomperessable data? that becomes more important..
i think its steady state performance will be shat. there are tests done at bit-tech that show current SF drives performing at 30 percent of write capacity with normal data once they achieve steady state performance.
already compressed data writing along at 80 mb/s sound good? NOPE. who cares how fast it is out of the box. what about in two months? there is a chink in the SF armor.....
so if these issues plague these new drives as well, SF has a problem.
Intels stated goal has been to improve steady state performance above all others
who wants to bet that the intels will beat these handily in real world usage?

[Praz]

there are tests done at bit-tech that show current SF drives performing at 30 percent of write capacity with normal data once they achieve steady state performance.
That would be a drive in a deliberate, forced, fully degraded state not steady state.

[Ao1]

That would be a drive in a deliberate, forced, fully degraded state not steady state.

Hi Praz,

The second largest complaint that continually crops up on the OCZ forum is (percieved) poor write perfromance.

This appears to be due to a combination of issues:

• Duraclass implements performance throttling
• TRIM commands are not implemented in the same way as other SSD’s
• Garbage collection is a lot slower requiring extended periods of idle time to enable the drive to recover.
• Static data rotation is occurring, which presumably slows things down in certain scenarios.

According to Tony:

"Here is the thing, if you hammer the driver with say enough writes that the drive would under normal use/see in 7 days within a few hrs, the drive will slow down for 7 days, maybe longer. It does this to protect the nand life. So your guys seeing a 50% drop may actually see 30% which is the normal drop, then a further 20% because at some stage they have hammered the drive and then not realised its going to take 5 days or longer for the speed to creep back up. Also remember this write quantity slowdown is further impacted by how you use the drive after you have hammered it."

http://www.ocztechnologyforum.com/forum/showthread.php?73537-Windows-Write-Cache-Flushing-discussion.&p=528034&viewfull=1#post528034

All of the above implies that SF based drives are not good for continual heavy use, uncompressible data or raid 0 set ups. In addition SF drives are unique in that sequential read speeds are also subject to degradation depending on what data is being read.

Despite this the performance can still be really good, but the problem is that the marketing performance figures do not stack up in most scenarios.

In the copy test below it can be seen that Intel drives are close to manufacturers’ specs. SF drives on the other hand are significantly below manufacturers’ specs.

I appreciate that SF drives are unique in the SSD market and for that reason it only makes apprising them even harder when SSD performance is already very hard to quantify.

Personally I think OCZ would get a lot more credit if more realistic marketing performance figures where used, that consisted of a performance range rather that “up to” figures. Complaints about “poor” performance would then be limited to other vendors selling SF drives and in reality everyone knows they are essentially the same product anyway.

I’d also like to suggest that OCZ should try and focus on improving reliability and support for existing drives rather than bringing out new drives every other week, which don't really deliver anything new and (I assume) only distracts limited resources away from existing drive development.

Please take all this as an attempt to be constructive.

http://img829.imageshack.us/img829/6762/26547810.jpg

http://www.behardware.com/articles/794-11/ssd-2010-report-15-models-compared.html

[-Masterchief-]

The PCI-E SSD, OCZ Revodrive (2xSF 1200 in Raid0) are very nice and cheap too !

[alfaunits]

OCZ may have issues, but we have to give them credit - they are essentially the only company trying to do something more in the SSD market than just stay in..
That said, they seem to continuously have issues with every SSD they tape out and they have more choices than the rest of the market put together :D

[Computurd]

the revo drive seems to be the worst of the 'offenders' of poor write performance.

dont get me wrong, i love OCZ, and they have been good to me :)
i still have ocz vertex gen1 going strong.
do not take it as a knock at OCZ< they are an awesome company with the best support imo.
I am referring to the sandforce controller as a whole, not OCZ, these are inherent issues with the SF controller. OCZ is just a reseller, one of many who incorporate SF into their drives...
and in response to Praz..not to argue, i respect you and your work, i know you work tirelessly...


That would be a drive in a deliberate, forced, fully degraded state not steady state.


To simulate a protracted heavy workload we then connected the drives to a secondary system running without TRIM support and copied the entire 100GB contents of the c: drive over to the SSD. These files include operating system files, multiple game installs, MP3s and larger video files – the typical contents of a modern hard disk. Once the write to the SSD was completed these files were then deleted and the process repeated ten times, resulting in a total write of over 500GB to our SSDs.

We then move the SSD back to the test system (filled from their last write), and clear the drive with a standard windows delete command, followed by an emptying of the recycle bin to ensure the TRIM command had been triggered. The drive is then left for an hour to ensure the TRIM command and any garbage collection algorithms had been completed before being retested using our new suite of benchmarks.

that is the exact method they used, and the results are not very good...at all! considerign especially that it is up against single devices. in raid they would crush the revo..

even with the sandforce, and them taking so long to recover...the whole point of the SSD is to not wait. with SF there seems to be a whole lotta waiting going on...

[Johnny87au]

How fast do you think the g3's will be compare to these new SF drives??? Technology is moving to fast hard for me to keep up now a days!

[SteveRo]

If we can believe the sales literature SF-2k should be fastest but the proof will be in the benching

[alfaunits]

The question as with current SF is - what are the numbers without compression ;)

IMO, I feel SSDs move too slow. The rest of the market as well, though I've reached a point where a faster CPU/GPU/MEM won't make much of a difference for me, but the SSDs would.

[Praz]

Ao1 and Computurd,

I agree and disagree with both of you if that makes any sense. Stated performance figures are an issue. The manufacturers quote numbers for a drive in the best possible light. On the other hand results such as AS SSD after multiple runs show a worst case scenario. For what I would consider normal 24/7 use the actual performance is a somewhat middle ground between these two.

I believe Tony has stated he has been asking for other performance numbers to be included in the drives' specifications. I can appreciate what this change would mean from both sides of the fence. As a user I would expect specifications to be a realistic representation of what I would see in normal day-to-day use. Currently this is not the case. As a business owner though I would have to look at the mindset of the buying public and the competition. Unlike the two of you and others here at XS the general buying public has no knowledge of the workings of SSDs. The drives have reached a point to where the major purchasers are now mainstream users instead of enthusiasts. As such, advertised price, capacity and speed is the determining factors when purchasing the drives. With all the drive manufacturers and Sandforce themselves using benchmarks that cast the drives in the best possible light it would be financial suicide for one manufacturer to deviate from this and advertise lower specifications.

I remember when the Core series of drives were released. I raised concerns of stating drive capacities which included the overprovisioned NAND. I'm sure it had nothing to do with me but OCZ did revise the stated capacities to not include the overprovisioning. This had to have impacted sales. I read many times on forums where users went with a different brand drive because that drive was larger then the OCZ. Still today some manufacturers are advertising drive capacities including the overprovisioned NAND. No doubt OCZ will get the advertised specifications in line with what can be reasonably expected but it's not an easy fix when bucking the entire industry.




To simulate a protracted heavy workload we then connected the drives to a secondary system running without TRIM support and copied the entire 100GB contents of the c: drive over to the SSD. These files include operating system files, multiple game installs, MP3s and larger video files – the typical contents of a modern hard disk. Once the write to the SSD was completed these files were then deleted and the process repeated ten times, resulting in a total write of over 500GB to our SSDs.

We then move the SSD back to the test system (filled from their last write), and clear the drive with a standard windows delete command, followed by an emptying of the recycle bin to ensure the TRIM command had been triggered. The drive is then left for an hour to ensure the TRIM command and any garbage collection algorithms had been completed before being retested using our new suite of benchmarks.


The above quote in bold is what I meant by "deliberate, forced, fully degraded state". I don't think there is a single person that would argue the above method of preparation is indicative of normal use. To me this type of testing methodology only serves to illustrate two things. That Trim does not function the same on Sandforce based drives as it does with other controllers and that the reviewer does not understand how the Sandforce controller works.

Duraclass is responsible for everything that happens with the drive. Through the use of Trim the controller is aware of discarded data but that data is not immediately acted upon. The drives does slow down once in a used state compared to fresh but the slowdown is nowhere near as severe as this type of testing wold have one believe. Even in a slowed state the drive is still the fastest overall of any drive on a SATA II interface.

Sandforce has change the game as to how Trim is acted upon. Personally I don't think they will be alone in this area much longer. With the coming die shrink the aggressive cleaning that we have become accustomed to almost has to change. Otherwise the available P/E cycles will be gone through in no time. Unless manufacturers do something such as lowering spec'd data retention to some ridiculously low amount such as three months.

[Ao1]

Storage is interesting because it is the only dynamic product in a PC and anything dynamic will have variable performance. (Dynamic as in it’s always in a continually variable state).

Reviewers obviously like to find chinks in a product, but if you can only find them via an unrealistic test methodology (for non-enterprise applications) then I don’t judge it as a fail. I agree that this particular test falls in to that category as it is based on a completely unrealistic usage pattern and as such it is therefore a useless “real life” performance indictor.

The other problem is that reviewers need to do something extreme to try and find something that differentiates between different SSD products, because in normal use there is little if any difference between good SSD’s and if there is a slight difference it is traded off somewhere else.

That said the reality is that the great performance of a SF drive is being detracted by unrealistic OTB performance specs.

I have always found that it is always better to under promise and over deliver as opposed to over promising and under delivering. You get more respect in the long term.

I really think the focus (in the non-enterprise market) is going to shift from marketing hype about “up to” performance to reliability, durability, capacity and price.

It certainly appears to be the way Intel are going with the G3. They haven’t really increased head line performance but they have significantly upped durability and I suspect that sustained performance will also be improved.

As an aside I didn’t follow storage when HDD first came out but I could imagine the headlines if it came out today as a new technology.
• WTF…… these things scatter data all over the drive causing performance to drop like a stone.
• WTF…… I have to spend 8 hours defragmenting the drive and even then performance still sucks compared to new.
• WTF......my PC used to boot in half an hour and now it takes an hour
• WTF…..my PC now sounds like a cement mixer.

Edit: Back on SSD the other big problem with reviews is timeframe. They all seem to want to get reviews out the door the day they get the drives. Hammering the drives in a short period induces a false used state condition that is unlikely to occur in real life, but it’s the only option they have in the time frame.

[Johnny87au]

jesus 2 much reading!!

[ripken204]

jesus 2 much reading!!

sparknotes: SSDs are fast, SandForce is very fast and beats Intel SSDs.
pretty much sums up this thread hehe