Originally Posted by elreyhorus
Sandforce has improved since their 1200-series days. However, the more recent 2281-based drives only offer at best mid level performance compared to newer Marvell-based drives.Sandforce still has some major caveats due in part to the compression tricks that the Sandforce controller performs
Overprovisioning - they require the most amount of overprovisioning (on average)
Poor compressed data performance - they don't handle compressed files as well as Marvell drives
Mediocre steady state performance
Does not play nice with TRIM - Marvell drives benefit from TRIM far more than Sandforce (as in TRIM restores more performance for Marvell drives)The proof
Seriously, don't by Sandforce when there are better controllers available.
Not that I love SandForce compared to any other controller, but as usual the details matter, particularly in this case.
The SF-2281 controller is seemingly unchanged from its inception. There are apparently a few versions, Intel uses the "VB1" chips. As with any SSD, its firmware makes or breaks its performance and stability, no more so than with the SF-2281.
Oddly IMO, it took Intel to truly fix this controller's firmware. The early SF-2281 firmware (use on the Vertex 3 SSD) is what destroyed the reputation of OCZ, along with a few other holes they shot in their feet.
The majority of companies using the SF-2281 controller use pure stock firmware from SandForce. SanDisk is one of the few that don't, and their Extreme SSDs (not Extreme II, they use a different controller) also didn't suffer from firmware issues that plagued so many other manufactures. They must have found the same bugs that Intel did, independently.
Once SandForce finally got their own act together, they then released the (in)famous 5.0 firmware. It seemed great at first, with increased performance. Then the shocker, TRIM was not working correctly with the 5.0.0 firmware (this is a story in itself.) Again that was straight stock SandForce firmware, as used in the two examples above, Corsair and Kingston. The Intel SF-2281 firmware was not based on the 5.0 firmware. SanDisk did use it in their Extreme drives, but not the earliest versions IIRC.
The question to SandForce about the 5.0 firmware is, how in the world do you miss the malfunctioning TRIM? It's just stunning, I don't have words for it.
That so many SandForce based SSDs were released with this faulty firmware demonstrates another reality, those "SSD manufactures" are using the "reference" circuit card and firmware, just like some video cards. Did those companies test this firmware? Did any of them catch the issue that one PC review website did apparently the first time they tried the new firmware in a SF-2281 controller? No they didn't, because they don't test the firmware much, if at all IMO.
The "fix" was to go back to much of the (finally) working fine version 3 firmware, and include it in the "5.0.3" firmware. TRIM of course was then working as it had in the pre-5.0.0 firmware.A very important point:
It is incorrect to say that TRIM does not work on the SandForce controllers, it does, but does not only with the 5.0.0 firmware
Also, it's not true that TRIM was completely broken in the 5.0.0 firmware, IMO. I tested this myself with a SanDisk Extreme SSD, with the bad 5.0.0 firmware. I ran an AS SSD benchmark on an empty SanDisk Ex and saved the result. I had some of the files AS SSD uses for some of its testing, one GB files of non-compressible data. I copied that file, and folders of 10 of those files to that SSD. I stopped when I could no longer successfully copy even a single 1GB file to the SSD, it would fail, a colored bar appeared above the SSD in Windows Explorer, and the free space in Properties was a few MB.
I then formatted the SSD in Windows Disk Management, which sends TRIM commands to the SSD, since a format deletes every file on the SSD. I let the drive sit only a few minutes, and ran AS SSD on it. If TRIM was broken, at least the write performance would suffer, the usual issue when TRIM is not available or working.
I was surprised to see the AS SSD result after the format was identical to the test on the empty drive, before I completely filled it.
It has also been noted that SandForce based SSDs do not react to TRIM in the same way as other SSDs do. That is also noted in the articles in the links above. (That article is Anandtech's reaction to not catching the TRIM issue in the 5.0.0 firmware, their spin on that is, it's always been this way, as we have pointed out. That is partially true, but then they came up with some oddball results never heard about again.) What is strange IMO is in other SSD articles about TRIM and SSD firmware garbage collection, it is noted that GC algorithms vary from aggressive to slow, with advantages and disadvantages to each. Why does that not apply in this case?
I also can't understand why they have never noticed that when formatting a SSD with a SandForce controller, it takes much longer to complete than others, as in approaching 10 times longer. Or when running the Intel SSD Toolbox Optimizer on a 520, etc, it is much slower and takes a while for the SSD to recover (Samsung does it right by showing the countdown with its optimizer.)
I would finally point out, that in PC Mark Vantage testing done on SSDs at various levels of being filled, the 50% level being considered most similar to most consumer usage, what do we find? What SSD scores the highest in that state? Intel 520.