Sandforce's flaky SSD SMART attributes

Elixer

Lifer
May 7, 2002
10,371
762
126
Health Status : Good (94 %)
05 247 247 __3 00000000000000 Retired Block Count
E7 _94 _94 _10 00000000000000 SSD Life Left
RBC's value is 100 - (100 * RBC/MRB). RBC = Retired Block Count and MRB = Minimum Required Blocks, according to SandForce.
If that formula is accurate, then how the heck do I have 247% ?

This is on a new, out of the box SSD with 0 data written to it.

I have found that any maker of Sandforce products have odd SMART readings on new SSDs, from forums like OCZ & Corsair & Gskil & Mushkin & ...

This guy has 12 new sandforce SSDs that all report different things.
http://gskill.us/forum/showpost.php?p=26068&postcount=1

Anandatech should show the SMART attributes for all SSDs they get in, and see what the deal is.

Doing some searching, it seems that Sandforce SSDs have lots of issues with SMART data, so the question is, why can't Sandforce fix the issue(s) in the way it writes SMART data ?
 
Last edited:

Mark R

Diamond Member
Oct 9, 1999
8,513
16
81
The only standard is that a failure should be reported when the 'health' value drops below the 'threshold' (in this case 3). Your drive appears to be compliant with that standard.

SMART is otherwise completely unstandardised. Manufacturers are free to report pretty much what they want. Some OEMs may modify the firmware to report their own specific values, or modify the values to be comparable to their previous products. E.g. big business/enterprise OEMs will often modify the firmware of their drives to improve compatibility with their RAID cards/server software.

Even the actual meaning of the values and their 'code numbers' are completely unstandardised. For example, there is no standard that says that SMART field '05' means 'retired block count' or 'reallocated sector count'. However, a lot of manufacturers choose to use 05 for this.
 
Last edited:

bulanula

Member
Apr 20, 2011
76
0
0
IMHO I would stay away from sandforce at all costs. Just get Intel if you want most reliable. Get Vertex 3 if you want speed that only lasts a couple of months before it dies etc.
 

Mark R

Diamond Member
Oct 9, 1999
8,513
16
81
This guy has 12 new sandforce SSDs that all report different things.
http://gskill.us/forum/showpost.php?p=26068&postcount=1

Anandatech should show the SMART attributes for all SSDs they get in, and see what the deal is.

Doing some searching, it seems that Sandforce SSDs have lots of issues with SMART data, so the question is, why can't Sandforce fix the issue(s) in the way it writes SMART data ?

Misread this part of your question.

The 12 drives all have different SMART values because they all have different health status. The way the values are presented appear entirely consistent between all the drives. Flash is not perfect. You always get some 'dead blocks' in much the same way as you get dead pixels on LCD monitors. Some drives came from the factory with a more dead blocks than others, so the score is adjusted. This is really just laziness on the drive OEM, because they didn't zero the SMART scores after burning in the drives. Instead, the results of the manufacturer's burn-in test are there for all to see.

As to how the sandforce RBC formula - it was probably for a different model of SSD controller. Your SSD has a different controller which calculates the score in a different way.
 
Last edited:

Elixer

Lifer
May 7, 2002
10,371
762
126
Misread this part of your question.

The 12 drives all have different SMART values because they all have different health status. The way the values are presented appear entirely consistent between all the drives. Flash is not perfect. You always get some 'dead blocks' in much the same way as you get dead pixels on LCD monitors. Some drives came from the factory with a more dead blocks than others, so the score is adjusted. This is really just laziness on the drive OEM, because they didn't zero the SMART scores after burning in the drives. Instead, the results of the manufacturer's burn-in test are there for all to see.

As to how the sandforce RBC formula - it was probably for a different model of SSD controller. Your SSD has a different controller which calculates the score in a different way.

Yeah, I understand that NAND isn't perfect, but there must be some threshold that the OEMs test against before they make a 80GB SSD into a 60GB SSD because it has too many defects...

If it states the drives health is at 85-99%, just how many defects did it find/detect, and is a 85-99% health rating acceptable for a new SSD ?
I also wonder where the 'bad block' list is stored at, and if there is any way to read that information from the controller ?

True, you can't go by rating alone, since my last SSD showed 100%, yet it failed without any of the SMART parameters increasing or decreasing.

Do the enterprise class SSDs handle SMART data, differently in any way? I suspect that intel does wipe the SMART data from all their SSDs, before they ship them out, but I am not sure if that is ethical or not. If the health status before was 92%, and they wipe it, the controller still has a list of bad blocks stored someplace I assume, and it will never reflect that 'already found bad block' back to the SMART parameter that corresponds to those bad blocks.
 

jiffylube1024

Diamond Member
Feb 17, 2002
7,430
0
71
Drive health goes up and down due to TRIM, IIRC (and possibly active garbage collection as well). I had both an OCZ Vertex 1 and Agility 2 (Indilinx and Sandforce, respectively) report low SMART health values - something like 90% for the vertex and 67% for the Agility 2. After running them in my machine for a few weeks, where I do minimal copying of large files to the drives and have about 30-40% of the drives unused, the health climbed up into the 90's for both.
 

Mark R

Diamond Member
Oct 9, 1999
8,513
16
81
Do the enterprise class SSDs handle SMART data, differently in any way? I suspect that intel does wipe the SMART data from all their SSDs, before they ship them out, but I am not sure if that is ethical or not. If the health status before was 92%, and they wipe it, the controller still has a list of bad blocks stored someplace I assume, and it will never reflect that 'already found bad block' back to the SMART parameter that corresponds to those bad blocks.

The normal way is that the drives are built with more spares than would ever been needed for the normal life. This allows manufacturing defects to be 'hidden' without affecting the specified life.

Your drive may have unusually few defects, and you would get a SMART warning when approaching end-of-life. However, you may get a bit of extra life beyond the recommended replacement time if your drive has unusually good flash.
 

groberts101

Golden Member
Mar 17, 2011
1,390
0
0
most(if not all) drives can go right up to 0% life left an start counting all over again. Should be called "dumb data".

And with how many Petabytes worth of data these drives can take.. it's hardly worth getting caught up on some arbitrary numbers generated by the controllers internal smart counters/algorithms.

Like I've always said.. if you don't like the lifespan shown by smart data?.. just destructive flash the firmware to reset it to 100% again. Problem solved and life goes on without all the paranoia.