|
So I just did the read test on a two day old Windows 7 Pro x64 installation on a brand new Samsung 840 EVO 250 GB drive. Keep in mind I'm using a motherboard that only supports a SATA 2 connection so it's not going to be nearly as fast as a newer SATA 3 board. 64 KB block size  8 MB block size  These results look pretty good to me. Seems pretty steady overall and the bottleneck is definitely the SATA controller, not the drive. So yeah, no real problems with my drive as far as these tests show. Then again I'm not exactly an expert on this stuff. For all I really know the test could be telling me my computer is about to catch fire. GreatGreen fucked around with this message at 02:01 on Sep 16, 2014 |
#
?
Sep 16, 2014 01:42
|
|
|
|
| # ? Apr 24, 2024 04:47 |
|
|
GreatGreen posted:From what I was reading on overclockers, the problem with the EVOs are specifically limited to the 128/250 GB models. They have found that a secure erase, meaning using a specialized app to totally zero out the drive, will restore performance to brand-new-drive levels. Technically, a Secure Erase on an 840 EVO is an encryption key erasure followed by a drive-wide TRIM, not a zero-fill. But it does eventually leave the drive filled with zeroes. You just can't run, say, DBAN and get the same effect even though in both cases the drive ends up with zeroes all over. In general, Secure Erase on TRIM-enabled drives doesn't do a write pass except as part of post-TRIM garbage collection. If a Secure Erase returns the drive to like-new performance, then we're definitely looking at something related to the controller rather than the NAND. That indirection table hypothesis is looking good.
|
|
#
?
Sep 16, 2014 02:08
|
|
|
GreatGreen posted:From what I was reading on overclockers, the problem with the EVOs are specifically limited to the 128/250 GB models. They have found that a secure erase, meaning using a specialized app to totally zero out the drive, will restore performance to brand-new-drive levels. Well, I'm glad I bought a 1tb drive if this is true. Nice to have a bit of extra piece of mind.
|
|
#
?
Sep 16, 2014 04:04
|
|
|
How long do we have to wait before 4-lane PCIe SSDs become normal?
|
|
#
?
Sep 16, 2014 04:22
|
|
|
Normal? Years and years and years. Commonly available? I'd bet on Skylake, which will up the number of PCIe lanes off the CPU to 20 - 16 for a GPU, and 4 for whatever (place your bets on an SSD vs. a Thunderbolt link).
|
|
#
?
Sep 16, 2014 04:38
|
|
|
GreatGreen posted:From what I was reading on overclockers, the problem with the EVOs are specifically limited to the 128/250 GB models. They have found that a secure erase, meaning using a specialized app to totally zero out the drive, will restore performance to brand-new-drive levels. I have the 500GB EVO, and my read performance takes a nasty hit on certain spots. Under 20 MB/sec, apparently (if the below chart is correct).  (My Desktop system uses SATA II, so max speeds are under 300MB/sec.) I got the drive in February of this year. Just under 7 months of use. Yeah, wiping the drive might return the performance, but only temporarily. It will just slow down again. My drive was wiped, and these "slow spots" showed up after.
|
|
#
?
Sep 16, 2014 05:44
|
|
|
Heh, glad my Samsungs are 830s. (Although I did purchase a stack of around 80 Evos for work machines, yay!) Samsung needs to answer to these problems soon. We need some professional analysis of multiple drives..
|
|
#
?
Sep 16, 2014 09:28
|
|
|
I'm sure it's just a firmware issue. Some software bug. If it is a hardware issue, then gently caress Samsung. I've now purchased dozens of SSDs, and I only use SandForce drives at work, but recently started recommending cheaper 840 EVO drives to people for their personal systems, ONLY because of the Samsung love in *this* thread.
|
|
#
?
Sep 16, 2014 13:39
|
|
|
I wouldn't worry too much about the issue. Firmware fix will resolve it. And its not just the reviews in this thread. Have you seen the reviews on Amazon? The evo is the highest, most rated SSD.
|
|
#
?
Sep 16, 2014 16:51
|
|
|
r0ck0 posted:I wouldn't worry too much about the issue. Firmware fix will resolve it.
|
|
#
?
Sep 16, 2014 16:54
|
|
|
r0ck0 posted:I wouldn't worry too much about the issue. Firmware fix will resolve it. HalloKitty posted:Heh, glad my Samsungs are 830s. (Although I did purchase a stack of around 80 Evos for work machines, yay!) Are the 840Pros affected by this at all? Have a couple of them running at home and not sure if I should be concerned or not. edit: thanks. VVV future ghost fucked around with this message at 17:23 on Sep 16, 2014 |
|
#
?
Sep 16, 2014 17:12
|
|
|
cisco privilege posted:Are the 840Pros affected by this at all? Have a couple of them running at home and not sure if I should be concerned or not. The 840 Pros have been tested and so far they do not exhibit any of the performance drops shown by the 840 EVO series.
|
|
#
?
Sep 16, 2014 17:19
|
|
|
Here's a dismal benchmark from a 240GB Crucial M500 in my work PC (Dell i5-something or another)
|
|
#
?
Sep 16, 2014 21:33
|
|
|
Magician says I've written 200TB in 35 days. How can I track what's causing it? Edit: It thinks I wrote 30gb since I made this post. Something is broken. Nothing stands out in Resource Monitor. Aphrodite fucked around with this message at 01:44 on Sep 17, 2014 |
|
#
?
Sep 17, 2014 01:19
|
|
|
Most SSD controllers rely on firmware running on one or more embedded ARM cores to do nearly everything related to wear leveling, mapping tables, data placement, etc. Only relatively low level functions need to be hardware assisted - things like BCH error correction, compression if it's Sandforce, encryption, and DMA engines. I know someone who worked on a team which built a 100% hardware enterprise SSD controller. This did not work out well for them. It's really really hard to fix problems in complex state machine logic when it's literally a shitload of interacting hardware state machines. And if you commit a bug to silicon... Uh oh. (They weren't completely insane, the first generation product was FPGA based rather than trying to tape out a bug free ASIC, but it was still a lot harder for them than if they'd done a more conventional HW plus SW system design.) What I'm saying is that if you're afraid of firmware do not use SSDs (or HDDs, or anything really)
|
|
#
?
Sep 17, 2014 02:20
|
|
|
Welp the issue cropped up on a Finnish forum too. Lots of guys posting their 840evo graphs with lovely 1/20 of advertised speed. My parents needed a budget pc and I bought a 250GB evo for them. I wouldn't touch them even with a long stick myself, but pro's were too expensive and evo was recommended everywhere. Hopefully they fix the issue in a way which does not cause data loss. Otherwise I'll just rma the drive next month and ask to exchange it to Intel or 850pro or something, I'll pay the extra.
|
|
#
?
Sep 17, 2014 17:06
|
|
|
The 850 Pro uses the same controller as the 840 EVO, so it may have the same problem.
|
|
#
?
Sep 17, 2014 17:55
|
|
|
Sandisk Extreme II vs. Intel 730 vs. Samsung 840 pro @ ~500GB look like they're all comparable, around $300 840 EVO clearly wins the $250 price point, and 850 Pro wins the $350 price point. Am I wrong in thinking that the Extreme II is the better of those three, though? So, like, SSD hierarchy would be: Samsung 840 EVO -> Sandisk Extreme II -> Samsung 850 Pro at each performance/price step?
|
|
#
?
Sep 17, 2014 17:57
|
|
|
This EVO nonsense is holding up my start on a new PC, I can't be arsed waiting 2 months just for Samsung to say "yeah there might be a problem lol". Is there any reason beyond price to choose the 840 EVO over something like the Intel 730 or the Crucial M500? Tha_Joker_GAmer fucked around with this message at 22:04 on Sep 17, 2014 |
|
#
?
Sep 17, 2014 21:40
|
|
|
Just go with the Intel. Their drives aren't always the fastest but they're rock solid and well supported. Lots of people still rocking X-25Ms from 2009 to this day. The original post should probably be redone to recommend the Intel 530 for laptops and the Intel 730 for desktops. It's a shame that Samsung seems to be going the OCZ route of releasing overhyped garbage made out of trash. I really thought they were better than that. Unfortunately it seems to be trumped up benchmark numbers and cheap prices that sell drives and not reliability and consistency. Dick Fagballzson fucked around with this message at 22:25 on Sep 17, 2014 |
|
#
?
Sep 17, 2014 22:05
|
|
|
Liu posted:This EVO nonsense is holding up my start on a new PC, I can't be arsed waiting 2 months just for Samsung to say "yeah there might be a problem". Speed. The 840 EVO is faster than both the 730 and M500 in consumer workloads. You can trust Samsung, current issues notwithstanding, more than Crucial to deliver an okay, bug-free product. Even if you do trust Crucial, the M500 is a pathetic unit compared to even their newer budget MX100, and the M550 is not much better than the MX100. The Intel's 7/530 are decent units. Of course, both the 730 and MX100 come with Power-Loss protection if that is your thing. We're not waiting on Samsung. We're just waiting on a major tech site like AnandTech to cover it to see if there is a real problem with the 840 EVO. Samsung has a good track record (basically by avoiding any major firmware issues at all on their client drives). The issue seem to be able to be resolved by a quick defrag (I know, I know) or secure erasing the drive, so there's that. Also, Alereon, it appears that the non-Pro/EVO drives are also experiencing this issue, and like the Evo, it is solved by defragmenting their drives. Well, that is what people are reporting in the Linus forums and OCN (and possibly reddit). fake edit: Dick Fagballzson posted:It's a shame that Samsung seems to be going the OCZ route of releasing overhyped garbage made out of trash. I really thought they were better than that. Unfortunately it seems to be trumped up benchmark numbers and cheap prices that sell drives and not reliability and consistency. I don't think you know what're you talking about. Garbage? There is a loving bug in one or two of their SSD models. poo poo happens. How about you quit it with your confirmation bias rear end and not pretend that a bug that appeared in otherwise good products means that they're poo poo. It's dumb, and it makes you look dumb. GrizzlyCow fucked around with this message at 22:34 on Sep 17, 2014 |
|
#
?
Sep 17, 2014 22:29
|
|
|
I'm up another 700gigs since that post yesterday. I see something off in Performance Monitor now though. Razer Synapse is writing 30 megabytes a second to disk. So unfortunately it looks like this isn't a broken counter. So probably don't buy Razer products.
|
|
#
?
Sep 17, 2014 22:35
|
|
|
GrizzlyCow posted:Also, Alereon, it appears that the non-Pro/EVO drives are also experiencing this issue, and like the Evo, it is solved by defragmenting their drives. Well, that is what people are reporting in the Linus forums and OCN (and possibly reddit). So it's all the drives that use TLC NAND then? You mean the stuff I said was cheap crap? It's not confirmation bias. It's me being smarter and having better instincts than the majority of the people in this thread. This is a case of you get what you pay for. And you guys apparently learned nothing from OCZ a few years ago, which everyone was also buying and recommending for awhile there. (USER WAS PUT ON PROBATION FOR THIS POST)
|
|
#
?
Sep 17, 2014 22:46
|
|
|
Dick Fagballzson posted:So it's all the drives that use TLC NAND then? No.
|
|
#
?
Sep 17, 2014 23:03
|
|
|
WhiskeyJuvenile posted:Sandisk Extreme II vs. Intel 730 vs. Samsung 840 pro @ ~500GB look like they're all comparable, around $300 E: To be clear, I am not considering the potential Samsung bugs in what I said above, because I don't feel we have actionable information yet. My general opinion is that since we have found out that decisions may have been made based on bad information, the most critical thing to do is to be doubly sure we are gathering correct information before changing our decisions. For example, the worst case scenario to me is that we change to recommend some new drive instead of the Samsung 840 Evo and find that it has the same issue. Alereon fucked around with this message at 23:59 on Sep 17, 2014 |
|
#
?
Sep 17, 2014 23:41
|
|
|
GreatGreen posted:The 840 Pros have been tested and so far they do not exhibit any of the performance drops shown by the 840 EVO series.  
|
|
#
?
Sep 17, 2014 23:41
|
|
|
Bob Morales posted:Here's a dismal benchmark from a 240GB Crucial M500 in my work PC (Dell i5-something or another) That's the same overall pattern as my 840 EVO mSATA:  SATA 3 Gbps interface. Two days ago, this looked the same, but the valleys were not as bad. Checking with defraggler, those dips correspond with LBAs where data is stored, with the spike at the 90 GB mark being the page file. I know this is a sample size of two models, but it looks like the 840 EVO and M500 SSDs have a pattern of being slow on areas where there is old data. I wonder what other drives show. Here's the standard pattern for a hard drive:  I don't know what that big spike is, but it's standard for the drive to be slower on average as it reaches the inner tracks. Here's a GIS image for an Intel X25-M:  Doesn't look like an M500, an 840 EVO, or a hard drive. It's just flat. I don't know if that's new or not. AnandTech's post-trim consistency checks:  840 EVO  Sandisk Extreme II (TLC + SLC cache) Anyone got an in-use SSD of a different model they can run HD Tach against to get a "used" benchmark? I did the Long Bench, so use that for apples-to-apples. I'll run it against my Crucial M4 when I get home.
|
|
#
?
Sep 17, 2014 23:44
|
|
|
Stock mSATA drive in an XPS12 for a fun new pattern.
|
|
#
?
Sep 18, 2014 00:02
|
|
|
Aphrodite posted:I'm up another 700gigs since that post yesterday. I see something off in Performance Monitor now though. Razer Synapse is writing 30 megabytes a second to disk. So unfortunately it looks like this isn't a broken counter. I had a similar issue for a while, but that was Catalyst Control Center fighting with f.lux and it was only about 5 gigs per day.
|
|
#
?
Sep 18, 2014 00:13
|
|
|
Patriot Wildfire drive. Sata 3, AHCI/Trim enabled, etc. Also relatively full (just realized).
|
|
#
?
Sep 18, 2014 00:16
|
|
|
Here's a Corsair Force GT CSSD-F240GBGT-BK R And here's my Evo 840 in the same system   e: Hm, apparently my Corsair drive is four full firmware revisions out of date. Better fix that I guess. flakeloaf fucked around with this message at 01:33 on Sep 18, 2014 |
|
#
?
Sep 18, 2014 01:27
|
|
|
Crucial M4 on SATA2 AHCI that was installed and has been running since July 2012: PNY XLR8 (non-pro) on SATA3 on a junky AMD E350 motherboard that was installed a couple months ago and has only 19 hours of power on time:  In the latter one I don't know if the dips are the junky ssd or the junky PC but it's never super slow. I've got an Intel 530 and Samsung 840 Pro as well but they don't see any significant dips.
|
|
#
?
Sep 18, 2014 01:52
|
|
|
Here's my Crucial M4 256 GB: The peak speeds in the back third seem to correspond with free space. I need a better LBA mapper than Defraggler. I haven't even put this stuff into an Excel spreadsheet, but I'm thinking that slowdowns on old data are not a Samsung-exclusive phenomenon. Certainly the 840 EVO results we've seen are the junkiest results so far, but it's a difference of degree rather than kind. I'm not sure if I'll do this, but if I were to investigate this further, I'd dump all these results into Excel and see what kind of flash was being used, what controllers are showing this effect, and (if possible) try to quantify the crappiness. E: even on gut feelings it's hard to generalize yet - the M4 above me was flat and mine is not. TRIM is definitely working on mine, TRIMcheck-verified. Factory Factory fucked around with this message at 03:44 on Sep 18, 2014 |
|
#
?
Sep 18, 2014 03:41
|
|
|
I think I win   (HD Tach shows a similar pattern: http://i.imgur.com/BwDoqg8.png) This is in a system that hasn't seen too much use recently. Someone said the 1TB drives weren't affected? Welp.
|
|
#
?
Sep 18, 2014 03:42
|
|
|
My data for the data gods. Here's my SanDisk Extreme 240GB, in use as OS/Apps since June 2013.  A barely used WD Blue of the same vintage for funsies. 
Peanut3141 fucked around with this message at 03:46 on Sep 18, 2014 |
|
#
?
Sep 18, 2014 03:43
|
|
|
Cawd Rud posted:I think I win  I have an inkling of a hypothesis: that what we're seeing is data retention related. Charge leaks from NAND cells slowly but surely, making reads harder. Many controllers have multiple levels of error correction: at least fast and slow, and sometimes heroic. As charge leaks, heavier error correction is needed on reads. The 840 EVO is the worst offender so far because TLC NAND is less tolerant than MLC of this kind of natural degredation, and it also has the slowest read-erase-program times when a slow read is wear-leveled. If this is true, then... *sigh* yes, the TLC NAND would be part of the problem. This does not explain why I am getting this behavior on a drive less than a week old, however. That's a ludicrously short time for data degradation in a system that's been powered every day. Plus the EVO is also equipped with a DSP that makes it much more tolerant of TLC's wandering voltage levels. So the indirection table hypothesis and other possibilities are still definitely in the running. Factory Factory fucked around with this message at 03:57 on Sep 18, 2014 |
|
#
?
Sep 18, 2014 03:55
|
|
|
Peanut3141 posted:My data for the data gods. Almost forgot about the 250GB EVO in my wife's computer as OS/Apps since March 2014.  I used to envy her the faster OS drive, not so much at the moment.
|
|
#
?
Sep 18, 2014 04:00
|
|
|
Here's my 180GB Intel 530: It's been in use since March as my OS drive.
|
|
#
?
Sep 18, 2014 04:08
|
|
|
I tweeted a couple of these at AnandTech's SSD guy. Especially that  -y 1 TB EVO reading at single-digit MB/s that Cawd Rud posted.
|
|
#
?
Sep 18, 2014 04:12
|
|
|
|
| # ? Apr 24, 2024 04:47 |
|
|
So with 200tb written in a month to my SSD, how much of it's projected life did I lose?Factory Factory posted:
They are pretty certain it's a flaw in wear levelling logic.
|
|
#
?
Sep 18, 2014 04:24
|
|