It's too bad the second one I tried to endurance test died -- it was plugging along, full steam ahead, and one day it didn't survive a system power cycle. It powers on, and I was able to update the FW, but it is otherwise trashed.Originally Posted by ZOMGVTEK
Of course, like a moron, I shorted the first one out a few days in.
FYI, for the uninitiated: http://www.nextlevelhardware.com/storage/battleship/
Yeah, the drive will likely outlive the computer by a factor of 10
I don't know that the MTRONs are all that reliable. The NAND? Sure. The rest of it? Way too complicated.Yeah, the drive will likely outlive the computer by a factor of 10
I blame the massive assortment of components on the PCBs. There is just too much damn stuff needed to make the magic happen.
Last edited by Christopher; 03-21-2012 at 08:55 PM.
Your drive still survived 90TB of writes or so ... for a normal user that is decades of writes
But yeah, risk of sudden failure always exists with computer parts
Vertex Turbo 64GB Day 20
162346.92 GiB
83.92MB/s
2911 Average Erase Count
42 MWI
552 hours
0/0/0 Program/Erase/Read Failures
----------------------------------
Samsung 830 256GB Day 2
48706.84 GiB (in 48.00 hours)
282.58MB/s
232 Wear Leveling Count
93 MWI
123 hours
0/0 Program/Erase Failures
----------------------------------
Lol, about 25 days to blast through the drive's 650TiB life. It kinda embarrasses all the other drives tested here.
However, it certainly does look like your 830 is performing more normally then the previous 830 tested. 600TiB is my conservative estimate for the drive life of a 256gig drive with 3000 cycle NAND, and your 830 is going a bit beyond that.
I see Ao1 wants to run an octane or petrol drive here.
http://www.ocztechnologyforum.com/fo...nce-Technology
Christopher has beaten me to it as he has already ordered one for testing.
It will be an interesting test. Someone over at OCZ reported a MWI of 59 after 6 weeks use. Was that due to relocated sectors or P/E cycles? If it was the later then the drives don’t have LTT. If it was the former that is a new capability as drives tested so far seem to only respond to P/E cycle counts.
All will soon be revealed. Kudos to Christopher.
http://www.ocztechnologyforum.com/fo...y-dissapointed
Patriot Torqx-2 64GB - Day 35
Drive hours: 839
GiB written: 53,232.00 GiB (51.98 TiB)
Avg MB/s: 19.88 MB/s
Bad blocks: 2/0/83
Wear cycle counter: 0/3327/5053
Intel 520 60GB - Day 26
Drive hours: 640
Avg MB/s: 91.66 MB/s
Host GB written (F1): 194,095.94 GiB (189.55 TiB, 6211070 raw)
NAND writes (F9): 137,455 GiB (134.23 TiB)
Reallocated sectors (05): 0
Failure count (AB, AC): 0 program, 0 erase
Media Wearout Indicator (E9): 57
I believe I have solved one of life's deeper mysteries. Its too early to tell, but I am almost 100% certain I know why the first one did that.
Hmmmm..........
Hmmm.
You know, the Plextor has less-but-faster cache (256MB vs 512MB on the Octane). And faster NAND (166MT 32gbit 24nm Toshiba Toggle). Both have abysmal SMART data.
The drive the Octane really reminds me of is the Marvell-powered Intel 510, not the Marvellized Corsair Performance/Plextor Ms.
510 on the left, Octane on the right
Just sayin'.
Coincidence? I think not.
Perhaps more questions will be answered when I start testing of the Octane
Last edited by Christopher; 03-22-2012 at 01:06 PM.
The Marvell 9174 controller has been used by Micron (C400), Corsair (Performance Series 3) and Intel (510) Oh and one other vendor that prefers to remain anonymous.
Each vendor has used customised firmware, although there are close similarities in the SMART values if you look carefully. Although the controller is the same each vendor uses different hardware configurations. The Intel for example has only 128MB Hynix DDR3-1333 SDRAM. Different NAND geometries & specification etc.
Performance of the 510 is however very similar to the Octane. The Smart values for the Performance Series 3 are also quite similar to the Octane. Strange.
Considering the difference between the Octane and Intel 510, it's pretty much remarkable how they perform so similarly. I am basically certain that if the two drives were to mate, their offspring would be pretty messed up. Their family tree doesn't seem to fork so much.
The Performance 3 is very similar, but I do not have one. However, the P3 makes up for it's lack of speed with FW that just won't quit when pounded and absent TRIM.
The Octane is like a 510 with slightly better high QD random performance and 25nm flash. And worse SMART attributes.
Last edited by Christopher; 03-22-2012 at 02:24 PM.
Anybody has an idea why current SSD controllers are pretty bad at staying alive ?
Especially with the SF drives they cause BSOD, not detecting by BIOS or simply die. It seems that the controller is at risk of dying much more than the NAND.
Maybe this is because no controller uses heatsinks and they get quite hot and electromigration takes its toll or something ?
I disagree with your assertion that SSD controllers are bad at staying alive. Some people do have problems with SSDs, but most do not (me included). In particular, it seems the Crucial M4 and other drives based on the Marvell controller are showing a very good history of reliability. Same with all of Intel drives.
In testing here, all drives have exceeded their design write life except for one old vertex turbo. (which was a dud drive) Some have exceeded their design write life by a huge margin.
Just like hard drives though, some drives are just duds, some people see multiple failures.
As for SF drives, their problems were mostly fixed by the last firmware release. The reason the sandforce controller used to cause frequent bluescreens was that the firmware had a bug somewhere in its SATA power transition logic. Fixing this bug fixed most of the bluescreens people were getting.
The other main failure mode of sandforce drives is panic locking. This occurs when the metadata for the LBA and NAND allocation tables on the SSD becomes corrupt, or some other internal error occurs. This is why we see some pretty weird behaviour on the old indilinx barefoot drives, but very rarely on the sandforce drives, because the controller detects it and locks the SSD down. The controller typically isn't actually dead.
Have you actually installed an SSD yourself? Or just reading what is appearing on threads?
So the Samsung nearly gave me a aneurysm today... I spent some time watching it spin the odometer, and I noticed that the avg. was dropping precipitously. It did take a few hours, but it seemed to have finished doing whatever the hell it was doing to cause such a massive drop in speed.
I expect it was just doing some truly extensive housekeeping.
So, I thought I'd toss in some temp data for the first 40 hours.
UPDATE:
I was able to hit 72,400+GiB in 72.01 hours, even with the slowness today.
Last edited by Christopher; 03-23-2012 at 12:17 AM.
That Samsung is running really hot. No surprise due to the large DRAM cache and sustained workload. In normal use with an ambient room temp of 20 degrees Celsius I have recorded a max temp of 31 degrees Celsius, but the average is only 21.11 degrees.
I wonder if the temperature has anything to do with the read/ write speed dropping (?) The sensor must be there for a reason. Maybe it triggers a temporary throttle to allow the temperature to reduce? It would be worth getting air flow over the drive to cool it down.
That would be incest. The Octane is using a rebadged Marvel controller. Indilinx customised the firmware and in the process managed to turn a mature and reliable controller into a less reliable controller.
Last edited by Ao1; 03-23-2012 at 01:34 AM.
Airflow? It's hanging out the back of the case, in open air. I don't think temp is an issue as the max operating temp is like 75c. Plus, those are hourly readings. I put the short term temp chart up, didn't I? It shows a lot more variation during the loop cycle. There are 64 32gbit dies attached to a 3 core ARM9. I think it's earned the right to run a little toasty when under prolonged load.
No, I believe the drop in speed was due to some rather intensive wear leveling and housekeeping operations. I've been keeping a journal of sorts, and when I first noticed the issue I spend about 30 minutes watching it. Then, I stopped the testing and ran a bench. At that point, temp was down considerably, and it was no faster after the bench break and cool down. After resuming the test, speed increased after several loops. What changed? Time. And more writes.
I do fully expect that to happen again in another couple days, but the general trend should be upward in terms of avg and instant speed. And sure enough, despite the couple of hours it spent optimizing itself, it's now back up to a higher avg than yesterday.
The first 830, once it's writes straightened up, kept getting faster and faster and faster.
Reads were fine. No change.
That would be illegal in 87.93% of America, FYI. I don't want people to get the wrong idea about our procreation habits.
All I'll say on the matter is this: The Octane is curiously like the 510 and P3. The Vertex 4 is shaping up to be curiously like the Plextor M3P (Toggle NAND? I was expecting an Toggle-laden Octane, but...).
Last edited by Christopher; 03-23-2012 at 02:53 AM.
Where did you get that max operating temp from? I’ve been looking at the SMART values but I can’t see a threshold value.
It would make sense that the controller is throttled if the temp goes above a predefined limit. The 830 is not an enterprise product so it would not be reasonable to expect sustained writes for extended periods, which obviously puts a high load on the DRAM, which in turn elevated the temperature.
Currently I am finishing off a test on my 830, but when I have finished I will try heating it up with a hair dryer whilst it is under a full write load to see what happens.
I saw it in the Q and A section of the Samsung support site. It's not temp related, but I would very much like for you to put the 830 under the "hair dryer test"
It just ripped off a whole loop at 350MB/s and the drive is at 53c. It doesn't get much warmer than that, ever. The drive was surely made to operate inside of laptops where temps could get much higher just from external sources. But I doubt that even 75c temps would hurt the drive. I took it more as a general recommendation -- "Please don't operate the drive below freezing or above the boiling point of iron."
I'm going on record here: you can take the hair dryer to the 830, but it's not going to throttle. But you should definitely try it anyway.
From Samsung's FAQ
Samsung recommends that your drive does not go below 32 Fahrenheit or above 176 Fahrenheit (0 to 75 Celsius). If your drive goes below or above these temperatures, damage could occur to your drive and information your drive may be lost. [FAQs]
Last edited by Christopher; 03-23-2012 at 04:56 AM.
Gentlemen,
Perhaps this deserves another thread but I wonder if the course I am about to take would provide useful information to this thread.
I intend to buy a 120-128Gig SSD from a choice of m4, performance pro, intel 320/520, and here's why;
I run MJ-12 bot as an app from one of my rigs (linux) and as an XS team member. What is it? well;
Consider if you will the following process:
Download from various websites a file of average size ~30-40KB and save to disk, repeat this process until there are 10,000 files then GZip them into another file of maybe 1-4Mb (this process uses a ramdisk as a workspace) which is then written back to disk. once complete this compressed file is sent back to a web address. At this time there are no need for any of the files so the space becomes available.
A single instance of the above is described but in fact the app is running maybe 50-60 concurrently. The folder that contains all of the files is typically around 4.5 GB
The potential here for expanding the operation may mean that there will be up to 10 concurrent versions (currently running 6) of the app so at any one time the amount of disk assigned to doing this work could be in the order of 45GB
I see that there seem to be a shed load of writes to HDD (Last month I downloaded around 10TB of data) that would maybe benefit from having an ssd or two to speed up the whole process and I thought with such use, provided I can find something to monitor just how much, the info might be useful.
The down side of this is that I am not really well versed in the workings of ssd's so as I said .....Provided I can find a program to give me the information you guys need I could only provide the raw data.
2 points then. Any preference of drive? and Would my "real world" data be of any use?
Last edited by OldChap; 03-23-2012 at 07:08 AM.
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
How well do the files compress? If they compress well the 520 would be an excellent choice. In addition it has comprehensive SMART data that enables you calculate wear for a work load over time and it can differentiate between NAND and Host writes. If you go for a 520 it would be great to get “real life” data for the Sandforce Life Time Throttling thread, which also covers how Sandforce based drives compress data at the SSD level.
If you go that route I can give you a SMART logger that will track SMART movements.
LZMA will compress the files to around 10% of original size if that is any indication....
I am stuck with the choice of Marvel with the performance pro or sandforce with the 520 and a quick look shows the intel is around £9 cheaper here at £145 so I was leaning in that direction.
I will pull the trigger probably Wednesday so if anyone has a better idea or if there is perhaps a more pressing need for info on another drive I would consider an option if it is not more expensive
EDIT: there is a drive that is £20 cheaper from Sandisk but although they are a known force with flash know how I wonder how their own 24 nm toggle NAND would fare.
Last edited by OldChap; 03-23-2012 at 10:30 AM.
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
If you can get the files down to 10% you should see a real benefit with the 520. Both faster write speeds and less wear.
^ Appears to be the dream scenario for SF. I think he would prosper with such an arrangement.
@oldchap -- The SanDisk Extreme uses 24nm Toggle NAND from their joint venture with Toshiba. For your application I don't know that it matters, but the Intel certainly has better SMART data and has been in the market a few months longer. You have one of those workloads that could benefit greatly from SandForce's deduplication scheme, so I think that is a fantastic idea. Which SF is up to you, as they appear to be quite a bit more expensive over there than here in the States. But the 520's SMART data and warranty are appealing, and while the Extreme might be slightly faster in some areas, your workload is going to be equally fast [on practically any SF].
Last edited by Christopher; 03-23-2012 at 02:33 PM.
Reply With Quote
Bookmarks