koitsu MVM join:2002-07-16 Mountain View, CA Humax BGW320-500
3 recommendations |
koitsu
MVM
2014-Oct-21 7:42 pm
Samsung 840 EVO SSD firmware update (read performance bugfix)Figure I'd take the time to make folks aware: » www.computerworld.com/ar ··· wns.html» www.anandtech.com/show/8 ··· ance-bug» beta.slashdot.org/story/208795The firmware update, as of this writing, comes in the form of a Windows application that both updates the drive firmware as well as moves series of NAND pages around on the drive. Look for Samsung SSD 840 EVO Performance Restoration Software on the below page: » www.samsung.com/global/b ··· 11000279I've used this on my own Samsung 840 EVO drive (250GB, used for OS), doing read benchmarks before and after. Since it's my OS drive and I don't tend to use it for many writes, I noticed that the read performance near the start of the drive (e.g. files that were written at the time of OS installation and probably not used much since) was quite bad -- maybe 25-35MBytes/sec tops. After the software finished running, everything is back up at 375MBytes/sec range. Sorry I didn't take screenshots. :/ I should note that my (non-EVO) Samsung 840 SSD did not require an update, nor was it adjusted in any way -- more importantly, it does not (in my experience) appear to suffer from the bug that the EVO series does/did, but I also tend to use that drive much more heavily (for both reads and writes) than my OS drive, so take what I say here in this paragraph with a bit of salt. I've a FreeBSD system with a Samsung 840 EVO (250GB) drive to update as well, once Samsung releases a bootable ISO. |
|
Oregonian Premium Member join:2000-12-21 West Linn, OR |
Oregonian
Premium Member
2014-Oct-21 10:36 pm
Thanks for the info. |
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
to koitsu
Cool, thanks. I was going to ask a couple of questions but they were answered in the Anand article lol. |
|
|
to koitsu
Thanks for the info, koitsu I've run the update on two systems - W7 and Windows Server 2012 - mostly "just in case" as the drives were 4 months old. No ill effects though. One thing that seemed strange was that the first step: Updating firmware seemed too be fast compared to the usual SSD's firmware updates. Probably not a full firmware update? |
|
lxAsTrOxl Premium Member join:2000-10-13 Chicago, IL |
to koitsu
This update is just for the 840 EVO not the 840 PRO correct? Silly question but I have been seeing mixed responses on other forums. |
|
|
Ghastlyone Premium Member join:2009-01-07 Nashville, TN |
said by lxAsTrOxl:This update is just for the 840 EVO not the 840 PRO correct? Silly question but I have been seeing mixed responses on other forums. Just the 840 EVO models. |
|
koitsu MVM join:2002-07-16 Mountain View, CA Humax BGW320-500
|
koitsu
MVM
2014-Oct-22 3:52 pm
What Ghastlyone said is correct. At this time it's believed the issue only affects EVO models, and the software mentioned can only be run on EVO models (PRO and others will not appear in the list if you were to run the software). What people speculate on forums etc. is purely speculative at this point. I'm sure Samsung is doing an internal review to determine if there are any other models impacted, but as of this writing the only models this applies to are EVO. |
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
JoelC707
Premium Member
2014-Oct-22 4:08 pm
Doesn't the regular 840 also use TLC NAND? If so, what I've read about this says that this was caused by the calibration algorithm used on TLC based drives. Assuming that to be true, and assuming they use the same algorithm as the EVO, I would expect an update for it soon as well. The 840 Pro uses MLC and I would assume does not use the same algorithm. |
|
koitsu MVM join:2002-07-16 Mountain View, CA Humax BGW320-500
|
koitsu
MVM
2014-Oct-22 4:27 pm
I wonder the same thing (my 2nd drive is a standard 840, but it doesn't show this kind of problem). I'm inclined to believe (as of this writing) that Samsung may have introduced a bug/quirk in the EVO firmware which did not exist in the non-EVO version, regardless of the fact that both drives are TLC. Meaning: prior to EVO's release-to-market, some programmer somewhere within the company decided to change some part of the algorithm (which was working fine on the 840) without thinking his/her changes through. Samsung doesn't exactly have a very good track record with their overall engineering with NAND products (I knew this in advance of purchasing their drives anyway): » beta.slashdot.org/commen ··· 48198863 |
|
pnjunctionTeksavvy Extreme Premium Member join:2008-01-24 Toronto, ON |
said by koitsu:Samsung doesn't exactly have a very good track record with their overall engineering with NAND products (I knew this in advance of purchasing their drives anyway): »beta.slashdot.org/commen ··· 48198863 Interesting, my original Galaxy S I9000m (along with thousands of others) failed due to bad flash and required warranty repair at about 6 months old...and that isn't even in that laundry list of past issues. |
|
|
to koitsu
Would this performance slowdown issue affect a Linux O.S. installed on a 250 GB Samsung 840 EVO SSD? |
|
koitsu MVM join:2002-07-16 Mountain View, CA Humax BGW320-500
|
koitsu
MVM
2014-Oct-22 10:17 pm
It affects any "long-standing data" (e.g. any data written to LBAs on the drive and after that haven't been marked as rewritable or erasable), regardless of the OS used. The referenced articles should be fairly clear about "what" gets affected. The Windows-based application very likely moves lots of data around, and to do so safely while Windows is running, would have to mark areas of the drive as preallocated via the filesystem, then move data around, rinse lather repeat, thus tickling wear levelling algorithms to store data in alternate NAND pages. The "Performance Optimisation" feature of Samsung Magician 4 already effectively does this (only necessary for use on XP and Vista due to lack of TRIM), in the same manner as how Intel's optimizer does. I'll be interested to see how they deal with this under non-Windows OSes. My gut feeling is that it'll be a bootable ISO that boots into FreeDOS, does a firmware update, and proceeds to read a series of LBAs into memory + then write them back to the same LBAs as before (and pray nothing goes wrong during that time), thus tickling wear levelling as mentioned above. I don't see how else they'd be able to do it through standard ATA CDBs. But there's probably other stuff going on under the hood that's proprietary as well, but heck if I know, I don't work at Samsung. |
|
trparky Premium Member join:2000-05-24 Cleveland, OH ·AT&T U-Verse
|
to koitsu
said by koitsu: I'm inclined to believe (as of this writing) that Samsung may have introduced a bug/quirk in the EVO firmware which did not exist in the non-EVO version, regardless of the fact that both drives are TLC. I'm thinking that it has more to do with the fact that the EVO uses a smaller silicon die process and that the algorithm didn't take that into effect properly. |
|
trparky
1 recommendation |
to koitsu
said by koitsu:The Windows-based application very likely moves lots of data around, and to do so safely while Windows is running, would have to mark areas of the drive as preallocated via the filesystem, then move data around, rinse lather repeat, thus tickling wear levelling algorithms to store data in alternate NAND pages. While I was running the Samsung Performance Restoration Tool on my system I noticed some interesting things while it worked. 1. The HDD activity LED on the front of my computer's desktop case wasn't blinking during the operation. 2. Windows 8.1's Task Manager didn't recognize that anything was being done to the SSD during the operation. I would've thought that my HDD activity LED would have been blinking like a Christmas tree and that the Windows 8.1 Task Manager would have been showing something during the operation but it wasn't. Something must have been going on so my thinking that the moving of the data in the SSD to restore performance was probably being done by the internal SSD NAND flash controller chip as versus the program itself. The program was just instructing the on-board SSD NAND flash controller chip to do whatever magic it needed to be done via proprietary ATA commands. |
|
koitsu MVM join:2002-07-16 Mountain View, CA Humax BGW320-500
|
koitsu
MVM
2014-Oct-23 11:19 am
Good piece of information, trparky (my case doesn't have HDD activity LEDs). Your theory is quite sound in that regard; the drive itself may be moving data around or doing some internal operations w/out the host controller involved. |
|
koitsu 2 edits
1 recommendation |
koitsu
MVM
2014-Oct-29 4:50 am
While running | Finished! |
Wanted to follow up with my experience updating another SSD, this time from Samsung's native bootable ISO (which boots some form of DOS) since it's been released. I should note the filesystems on the SSD were UFS/FFS (i.e. FreeBSD). Please be sure to read (not skim) Samsung's PDF instructions first!Things went smoothly and without a hitch. I didn't have to change controller modes (i.e. keeping AHCI enabled was perfectly fine). The bootloader appears to use ISOLINUX and loads some form of DOS. The firmware upgraded very quickly (others cited concerns but it seems to be part for the course). The "performance restoration" phase took about 7 minutes (you can see the timestamps in the screenshot), and happened in 2 phases. Periods/dots printed in the 2nd phase were slower and took a while to get started, but eventually did. Just be patient. The system kicks you back to a DOS prompt when finished. I recommend powering off the system, letting things sit for about 15 full seconds, then powering back on, solely to ensure the drive operates off the new firmware. No data loss or issues encountered post-upgrade. I did, however, notice that SMART tests (e.g. short/selective/etc.) no longer print a valid power-on-hours value in the self-test log when finished -- the firmware instead chooses to store 0. Don't ask me why that is, but developer gut feeling says it's probably the result of some in-flux development code that got included in the release to the world (I can only imagine the chaos over at Samsung when this issue came to light). This shouldn't impact 98% of users, but for those of us that do drive analysis and use SMART tests a lot it's something to take note of. I took a couple pictures with my mobile phone as things were upgrading, so people could see what it looked like. |
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
JoelC707
Premium Member
2014-Nov-17 6:46 pm
Apologies if this is in any of the links provided but is there a way to just update the firmware without running the performance refresh? I just bought an 840 EVO 250GB and I'd like to go ahead and update the firmware but also since it's a new drive, it doesn't exactly need to run the performance refresh. |
|
trparky Premium Member join:2000-05-24 Cleveland, OH |
trparky
Premium Member
2014-Nov-17 7:09 pm
From what I know, no... there's currently no way to just update the firmware. |
|
koitsu MVM join:2002-07-16 Mountain View, CA Humax BGW320-500
|
koitsu
MVM
2014-Nov-17 8:26 pm
No, the "performance refresh" is part of the procedure. In fact, I'm under the impression it's what re-established the performance. It acts like a partial drive TRIM, from what I can discern. My guess is that it moves (reads then writes) NAND pages in such a way that guarantees optimal read conditions (i.e. not having to be re-read repeatedly, which is the bug in question). Thankfully the procedure does not take very long. |
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
JoelC707
Premium Member
2014-Nov-17 9:17 pm
Hmm, OK then. Sounds like the easiest way to do it then will be to just install Windows on it and let it run the program as is instead of booting the ISO and updating it first. Thanks! |
|
norwegian Premium Member join:2005-02-15 Outback |
to koitsu
Couldn't you use the bootable media for Mac/Linux for a pre-windows enviroment so you do not need windows installed?
|
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
JoelC707
Premium Member
2014-Nov-19 6:35 pm
Yeah, you could certainly do that. |
|
norwegian Premium Member join:2005-02-15 Outback 1 edit |
So in theory, as I've not tried because my update was run within the windows enviroment (Edit: Win 8.1 pro x64) and I don't have a new SSD to try with; you could fix/update the SSD before anything is applied to the drive? That is: If you wanted to run this tool on a new HDD before setting partitions and operating systems, it will work fine?
On another note, has anyone found or heard of the update process not actually working correctly, or more specifically corrupted data on the SSD? |
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
JoelC707
Premium Member
2014-Nov-20 7:26 am
I simply installed Windows on the drive before updating it but yeah it should be that simple. The performance refresh will likely still do something but it may go quicker. |
|
JoelC707 |
to trparky
said by trparky:The HDD activity LED on the front of my computer's desktop case wasn't blinking during the operation. Windows 8.1's Task Manager didn't recognize that anything was being done to the SSD during the operation. They seem to have made a new version of the software (the Windows one is now labeled v1.1 while the bootable ISO is v1.0). I just ran the v1.1 Windows utility on my system and saw the exact opposite, the drive activity light on my case was on solid and Windows 8's task manager showed considerable periods of 100% disk activity. The utility has also changed slightly, it now does three passes of optimization instead of two (I don't recall seeing if the Windows utility did two or three originally but I do see the ISO does two so I'm assuming the Windows utility did two originally). I've attached screenshots of task manager while running and the utility after running to show the three passes. |
|
JoelC707 |
to koitsu
said by koitsu:I did, however, notice that SMART tests (e.g. short/selective/etc.) no longer print a valid power-on-hours value in the self-test log when finished -- the firmware instead chooses to store 0. I wondered about this when I noticed the version number change and now see the firmware number has changed. Mine is currently reporting 44 for power-on hours, which would be about right since I installed it about two day ago. Well, I say the firmware has changed but after closer inspection I see that's probably a "before" shot as it didn't reboot after the update (and the number in your pic is the current firmware for the 840 EVO as released in December 2013). This is of course pulled from their software and not a 3rd party software so take it with a grain of salt as it may not have updated after the firmware update. |
|
koitsu MVM join:2002-07-16 Mountain View, CA Humax BGW320-500
|
koitsu
MVM
2014-Nov-21 6:40 pm
said by JoelC707:said by koitsu:I did, however, notice that SMART tests (e.g. short/selective/etc.) no longer print a valid power-on-hours value in the self-test log when finished -- the firmware instead chooses to store 0. I wondered about this when I noticed the version number change and now see the firmware number has changed. Mine is currently reporting 44 for power-on hours, ... You've misunderstood what I was referring to. I was not talking about SMART attributes, I was talking about SMART tests. Proof of what I was referring to: root@icarus:~ # smartctl -l selftest /dev/ada0
smartctl 6.3 2014-07-26 r3976 [FreeBSD 9.3-STABLE amd64] (local build)
Copyright (C) 2002-14, Bruce Allen, Christian Franke, www.smartmontools.org
=== START OF READ SMART DATA SECTION ===
SMART Self-test log structure revision number 1
Num Test_Description Status Remaining LifeTime(hours) LBA_of_first_error
# 1 Short offline Completed without error 00% 0 -
# 2 Selective offline Completed without error 00% 0 -
# 3 Short offline Completed without error 00% 0 -
# 4 Short offline Completed without error 00% 3235 -
# 5 Short offline Completed without error 00% 3208 -
# 6 Short offline Completed without error 00% 3035 -
# 7 Short offline Completed without error 00% 3021 -
# 8 Short offline Completed without error 00% 3017 -
# 9 Short offline Completed without error 00% 3017 -
All tests done under the updated firmware have a hour count of 0. I haven't seen any problem with SMART attributes, only tests. So like I said, someone at Samsung either left some debugging code in place, or some other anomaly; possibly someone restructured the SMART test sector area to use a different format between firmware revisions. I have no idea. All I can do is speculate, but my gut feeling is that the bugfixed firmware was released in massive haste given the nature of the problem, and developers had not re-stablised their SMART testing code. |
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
JoelC707
Premium Member
2014-Nov-21 7:36 pm
Ahhh yeah I misread that. Was the version of the utility you ran 1.0 or 1.1? |
|
koitsu MVM join:2002-07-16 Mountain View, CA |
koitsu
MVM
2014-Nov-21 7:40 pm
1.0 was used. 1.1 was not out yet. According to the PDF, however, 1.1 does not do anything different relating to the actual bugfix (i.e. the firmwares are identical) -- it has something to do with the Windows software. |
|
JoelC707 Premium Member join:2002-07-09 Lanett, AL |
JoelC707
Premium Member
2014-Nov-21 7:41 pm
Ahh OK so the new version isn't different firmware (it would be kinda silly for there to be different ones). I was hoping maybe they had fixed that oddball "feature" lol |
|