Sandforce 2000

Y

Yoda4561

Gawd
Joined
Jul 2, 2008
Messages
896
Anandtech just released the info.

http://www.anandtech.com/show/3971/...troller-sf2000-capable-of-500mbs-and-60k-iops


All on paper right now but don't see any reason they won't be able to pull off those performance numbers. The article indicates that 25nm flash won't be in the early shipping products, so prices will probably be higher than Intel's G3 per gigabyte. The article also hinted to actual benchmark numbers on real hardware in a few weeks, should be interesting to see where they're at with it.
 
if they can actually pull that off, would even using one on a 3.0gbps controller be a huge bottleneck?
 
There is a big reason to expect that Sandforce will NOT achieve those claimed specifications: current Sandforce SSDs do NOT achieve anywhere close to their claimed performance.

The most careful SSD reviews these days are coming from bit-tech.net. They use AS-SSD to test sequential write speed for incompressible data, and also they fill the drive up with data, delete it, run TRIM, and then test the drive again.

http://www.bit-tech.net/hardware/storage/2010/09/03/ocz-revodrive-review-120gb/3

Check out the lighter-colored bars on the sequential write speeds. Those are the speeds after writing a lot of data to the drives and then TRIM. Note that the Intel X25-M 160GB gets 99 MB/s sequential write even after being heavily used. It is spec'ed at 100 MB/s sequential write. Just as Intel specified, so their SSD performs.

Next, look at the Sandforce drives lighter-bar sequential write, for example, the OCZ Vertex 2E 120GB. This is a drive that is spec'ed at 275 MB/s sequential write. But when someone actually measures the speed with realistic data, after the drive has been used, it only manages a pathetic 83 MB/s sequential write. That is only 30.2% of the spec'ed value, and is even lower than Intel's 99 MB/s !

Or look at the Revodrive, which is two SF drives in parallel with a RAID controller. It is spec'ed at 490 MB/s sequential write, which looks quite similar to what Sandforce is claiming for the forthcoming SF2000 series. But what is the actual, real world sequential write for the Revodrive? bit-tech.net measured it, and it is a pathetic 139 MB/s. A single Crucial C300 256GB drive achieves 190 MB/s !

Bottom line is that none of Sandforce's specifications can be believed.
 
How much real world data is really incompressible though?? That's one of those things that's never been really clear to me. Benchmarks using only non-compressible data don't represent real world scenarios I think, except perhaps for archival storage with zips, jpgs, and large compressed video files which all aught to be on platter drives anyways. So for example how much stuff in the average steam/game installation folder is compressible/non-compressible????
 
I have to disagree with you also, John. A fully random data set shows Sandforce drives in the worst light possible, but real data is often compressible. I'm not surprised a Sandforce drive is relatively slow in such a scenario thanks to all the work it is trying to do to reduce write amplification. I don't think it's deceptive advertising for Sandforce to claim speeds based on typical or potential performance.

Bit-tech does a few things right, but for SSDs I'd much rather read a review from Anand. I'm not even going to start on their abysmal fan testing method.
 
In this context, 'incompressible' means 'incompressible to the Sandforce controller'.

Note that the Sandforce controller must have a total power dissipation of a few Watts at most. It is not very powerful. To achieve 500 MB/s write speed, it would have to run its compression algorithm at 500 MB/s. A compression algorithm running at 500 MB/s on a, say, 2W processor, is going to be crude. Probably can not do much more than compress running strings of identical bits.

AS-SSD also has a copy test, which duplicates folders full of real data files -- a folder with large ISO files, a game folder with a mix of large and small files, and a program folder with a lot of small files. tweaktown shows a comparison of SSDs on this test, and the Sandforce SSDs do not perform anywhere near their specifications when given this realistic data:

http://www.tweaktown.com/reviews/3565/adata_s596_turbo_128gb_solid_state_drive/index10.html

Bottom line is that most realistic data is going to look incompressible to the Sandforce controller.
 
Last edited:
If nothing else, I think this preview makes me wonder what will happen next. If we're already approaching the boundary of SATA 6gbps (600MB/s after encoding), will future SSDs switch to native PCI-E? Will it even matter? I wonder what the current "bottlenecks" are when it comes to consumer I/O for apps and games.
 
Kaldskryke said:
If nothing else, I think this preview makes me wonder what will happen next. If we're already approaching the boundary of SATA 6gbps (600MB/s after encoding), will future SSDs switch to native PCI-E? Will it even matter? I wonder what the current "bottlenecks" are when it comes to consumer I/O for apps and games.
Click to expand...

I don't think greater max throughput will matter much, just a matter of actually getting a real 500-600 MB/s across the board in various situations. That should be more than enough in the consumer space for the next 10 years. Somewhere around that point sata6Gb will become a bottleneck for highly multicore cpu/gpu computing again probably. Full duplexing may have some nice benefits in heavy multitasking situations but since there aren't any storage products to test there's no way to find out until it gets here. That's in the cosumer space mind you, enterprise will probably keep pushing the bar up since they can use as much as they can get.
 
Kaldskryke said:
Do you have anything to back up that claim?
Click to expand...

Sure, have a look at the tweaktown link in my previous post.

Also, here is another copy test. Note that the SF drives only achieve write speeds of about 130 MB/s on copy tests for folders full of medium or large files:

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

That 130 MB/s is about what bit-tech measured for incompressible data with AS-SSD on a fresh drive. But the Sandforce write speed goes down to about 83 MB/s on a used drive, even after TRIM.

But really, common sense should tell you that a 2W processor cannot achieve much compression when it has to have a throughput of 270 MB/s, let alone the 500 MB/s they are claiming for the 2000 series.
 
Last edited:
God, this is strange....
i feel dizzy, like the room is spinning, i can barely focus enough to type...
cant believe im typing this...

John is right. totally.

there are serious differences between the real and 'perceived' speed of the SF drives. we must wait to see this drive develop...

these performance problems of 'steady state performance' (thats what its called guys) is what intel is trying to get rid of. the thing about these new intels that most jokers dont get is that intel has stated that they are more concerned about the steady state performance. with the G3 it WILL NOT WAVER. it will be the same, from first write, to last one. they are looking for a total solid line of performance. no variability. all you have to do is watch last years IDF presentations to see that is their goal. hell they say it~!
 
Last edited:
Computurd said:
God, this is strange....
i feel dizzy, like the room is spinning, i can barely focus enough to type...
cant believe im typing this...

John is right. totally.

there are serious differences between the real and 'perceived' speed of the SF drives. we must wait to see this drive develop...

these performance problems of 'steady state performance' (thats what its called guys) is what intel is trying to get rid of. the thing about these new intels that most jokers dont get is that intel has stated that they are more concerned about the steady state performance. with the G3 it WILL NOT WAVER. it will be the same, from first write, to last one. they are looking for a total solid line of performance. no variability. all you have to do is watch last years IDF presentations to see that is their goal. hell they say it~!
Click to expand...

Speaking of IDF, did you see the two slide sets on endurance at the latest IDF?

Apparently the Intel SSDs now allow you to measure actual workload endurance demands via SMART.

Also a lot of interesting data on game load times, etc.
 
there are serious differences between the real and 'perceived' speed of the SF drives.
Click to expand...
IMHO this has been true with all the SSDs I've owned. The Sandforce units do use the compression "trick" to advertise great speeds but the only place the "trick" really shines is Rar/unRARing files. In those situations it outpaces competitors by a mile.

The Anand article points to other improvements in the SF controller (DQS signal) and even says...
Let’s talk about reality for a second. SandForce quotes standard iometer numbers, which are usually quite optimistic for SandForce's controllers. I’d expect real world performance to be a bit below these figures but not by a lot for many workloads.
Click to expand...

Even if the new SF drives do 1/2 of the quoted numbers at steady state it'll be a screamer.
 
Old Hippie said:
Even if the new SF drives do 1/2 of the quoted numbers at steady state it'll be a screamer.
Click to expand...

Unfortunately, half may be optimistic. The current batch of Sandforce drives can deteriorate to about 30% of claimed write speed on data its controller cannot compress, after the SSD has been used a bit. And as you say, the Sandforce controller cannot compress most data.

Bottom line is that you can be looking at 83 MB/s sequential write for drives which Sandforce claimed would have 275 MB/s sequential write.

http://www.bit-tech.net/hardware/storage/2010/09/03/ocz-revodrive-review-120gb/3
 
Last edited:
Quality of the NAND should not be underestimated as this is the SSD-manufacturers choice and not Sandsforces one...
 
The flash memory used in Sandforce-based SSDs has nothing to do with the deterioration in performance I mentioned. It occurs in all Sandforce-based SSDs, regardless of flash memory make.
 
Just out of curiosity, do you have some kind of personal grudge against sandforce? ;)
 
sweloop64 said:
Just out of curiosity, do you have some kind of personal grudge against sandforce? ;)
Click to expand...

I do not know anyone at Sandforce personally. Nor, in fact, do I have any personal association with any SSD manufacturers or resellers.

Are you insinuating that I have posted something about Sandforce that is incorrect? Because if so, I'd like to know about it. As far as I know, everything I have posted about the performance of Sandforce SSDs has been supported by reasonable evidence.
 
Unfortunately, half may be optimistic.
Click to expand...
It may or it may not.

Your guess is as good as mine and I'm going to guess 140 MB/s sequential write for the new drives.

Soooo...let the guessing games begin! :D
 
One of the reasons I have been posting a lot about Sandforce's write performance degradation issue is that I hope that if the issue gets enough attention, that Sandforce may put some resources into fixing it. So, I will decline to guess an exact number on the worst case write speed for the new SF2000 series, and instead hope that SF fixes it so it is no longer 30% of their claimed performance.
 
Plus one. these numbers he is quoting have been around awhile.

@aaron-i will watch them tonight, sounds interesting!
 
Computurd said:
Plus one. these numbers he is quoting have been around awhile.

@aaron-i will watch them tonight, sounds interesting!
Click to expand...

Also probably worth checking out STOS002 which covers the new integrated SAS/RAID functionality coming out next year.
 
that one i did watch, that and run time performance of alist games. the next gen raid is going to be ultra powerful, and if it werent for the last major hurdle of system crashes/data integrity, i would swear off raid cards when it comes about.
i am wondering how long it will take for them to put a integrated supercap onto a chipset, thus giving some power protection for onboard raid sets?
the last part of that where they talk about OS crashing and losing raid sets troubles me as well, that is the reason i initially went to hardware raid, when pushing some severe clocks, bsod are a way of life, and the cards give me that edge to push higher....
there are some exciting things coming, and the increased performance of the onboard is going to be just unbelievable. for 24/7 systems at stock clocks i think this is going to really start infringing on some hardware raid setups..
 
You must log in or register to reply here.
Back
Top