SSD Thread: (its good, ignore the anime)

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > SSD Thread: (its good, ignore the anime)

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Kaleidoscopic Gaze posted:

If something shits the bed and I have to reinstall my OS, can I save a copy of the encryption key and rescue the data from my encrypted drive by plugging it into another computer?

Don Lapre posted:

You can, but you need to have backups of any encrypted data.

That's a weird way to spell "no." Re-read what he asked.

# ¿ Sep 8, 2016 13:31

Adbot: ADBOT LOVES YOU

# ¿ May 3, 2024 08:38

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

On normal mATX boards, do the NVMe drives just stick up out of the last PCIe socket, or is there a bracket to hold them? I'm not worried about their weight so much as a cable hooking them and snapping a $700 part.

# ¿ Dec 12, 2016 18:38

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

nielsm posted:

Apart from enterprise level PCIe SSDs, most NVMe ones are for the M.2 slot which mounts it parallel to the mainboard. The M.2 connector is not mechanically compatible with PCIe.

Edit: If it fits in a regular PCIe slot it will also have a bracket for mounting in the back plate.

There's no M.2 slot on this MB (Asrock H97M Pro4) but it got a BIOS update for NVMe boot. I guess it supports it via a PCIe->M.2 adapter board. Are those just passthrough?

M.2 is such a mishmash of non-standard standards. Keying, two different protocols on the same socket, nothing compatible with anything else, and different options for how far out to put the mounting post. I guess the laptop makers are happy with it, because it's basically impossible for a normal end-user to buy an upgrade from anyone but them.

I don't remember the transition from AGP to PCIe being anywhere near this mis-managed, nor PATA->SATA for that matter. Even JEDEC, with 4 incompatible specs on the same connector has managed to keep it clear via version bumps. Was M.3 really a dealbreaker?

# ¿ Dec 12, 2016 21:35

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Harik posted:

There's no M.2 slot on this MB (Asrock H97M Pro4) but it got a BIOS update for NVMe boot. I guess it supports it via a PCIe->M.2 adapter board. Are those just passthrough?

thebigcow posted:

AFAIK, yes. NVMe drives can currently be attached to PCIe through either an actual PCIe slot, an appropriate m.2 connector, or an appropriate u.2 connector.

I've got some christmas amazon giftcards to burn now, and I wanted to up my 256 to a larger NVMe - boot isn't a requirement, worst case I'll toss GRUB on a USB keyfob to boot from.

Usage: Linux development. I spin up VMs to dick around with, so I don't think I need the 960 pro, but I can definitely feel the IO hit on my current SSD (Crucial M4) when I'm rebooting them.

So some practical questions: Does PCIe 2.0 limit the speed? I don't currently have a DGPU so the 3.0 slit is open, but I was considering adding one, which would dump the NVMe to the 2.0 x4 slot.

What PCIe->M.2 adapter will work? I obviously don't want to be running in SATA emulation mode.

# ¿ Jan 10, 2017 05:45

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

crazypenguin posted:

Some back of the envelope estimates suggest it's decompression limited. (i.e. CPU)

Some numbers: JPEG seems to decompress at maybe 20 MB/s of compressed data per core. gzip at about 80 MB/s, I think. With 8 cores, that's about SATA speeds.

It'll be interesting to see what happens when this bottleneck becomes more recognized. Maybe nobody cares? Decompression accelerators? Compression methods with faster decompression? GPU assisted decompression? FPGAs? (Hey, there's reason to want them on servers, and there may be reason to want them on cell phones, why not a reason to want them on the desktop!)

That's pretty insane, LZO 2012 was decompressing at 1.2 GB/s on sandy bridge. Why would anyone gimp their performance by using a slow algorithm?

# ¿ Jan 15, 2017 22:57

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Bob Morales posted:

A couple things: I think jpg decompresses at closer to 80mb/s these days than 20mb/s

That said, a 1mb jpg file is going to compress at say 10:1, so you end up with 10mb of raw pixel data. So you might only be uncompressing 10mb of data per second but you end up creating 100mb per second.

Second, generic data compression algorithms don't compress images very well, and they're are just working on a stream of data. JPEG is analyzing blocks of an image and all that poo poo.

I meant mesh data, which is generic. Point taken, our figures are probably at opposite ends of the algorithm.

# ¿ Jan 16, 2017 01:58

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

So I did get a chance to play around with this 960 EVO in linux.

Wow. Just wow. It's an upgrade I can feel coming from a crucial M4.

Synthetic benchmarks matched what the review sites have, bla bla bla. Nothing interesting there. I tried my actual workload on it.

My big hit is when snapshotting/booting VMs, and I get a few seconds 2.4-2.7GB/s transfer speed. I might not notice if it was in a 2.0 slot and throttled to 2GB/s, but if there was more overhead and it was 1.8 or lower or so I absolutely would. Peak IOs was "only" pushing 30k/sec. Still, way better than SATA.

I'm going to leave it in the 3.0 slot until I get around to buying a GPU. By then I may have bought ZEN or a newer intel that's not so lane starved.

I think the biggest benefit is I'm 100% CPU bound now, I'm just not seeing any IO load anymore.

# ¿ Jan 23, 2017 00:24

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Question about the samsung 850 - does it have a severe problem reading trimmed (zero) sectors?

I was trying to clone a 250gb to another disk, it was a nearly fresh install and I was putting it on a 500 instead of a 250. I thought (at the time I started) that a SSD copy would be faster than waiting around for w10 to get it's head out of it's rear end and do an agonizingly slow install. Instead it took north of 2 hours to clone.

It started at 250MB/sec but pretty quickly dropped down to 25-30MB/s when (I'd assume) it copied the 12+GB of windows 10 trash and started copying empty space.

It's either that or my 840 has sustained write problems, and that's a bad sign.

Neither have anything interesting in SMART.

# ¿ Jun 12, 2017 22:11

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

stevewm posted:

Its your 850 a Pro or EVO? The 850 EVOs had a problem with earlier firmwares where older files would read slower. The 840s have the same problem to a lesser degree. This was fixed in newer firmware updates.

Also what tool are you using for cloning that would bother with empty space? Use something like Macrium Reflect... It only copies occupied sectors, making clones much faster.

dd.

I figured it would be fast enough from SSD to SSD that it wasn't worth downing the whole machine to run a standalone cloning program. I probably should have just to save the worthless writes, but oh well.

BobHoward posted:

Is the destination an 850 EVO? It's a somewhat expected behavior if so.

(good explanation of the effect was here)

I knew about that feature of the EVO, but I didn't think the sustained write dipped down to 25 MB/sec. Source was a 250GB 850 EVO, destination was a 500mb 840 EVO.

Someone benchmarked it and graphed the initial write speed and the drop to the sustained speed, but I don't remember where.

Thinking about it, it could be that I didn't trim the partition I was cloning onto, so it was thrashing eraseblocks of old data instead of having ~400gb free to just use. That would hit the other 840 bug as well, because it had to re-read that old partial EB to move data off it and since it was old and stale it would take longer.

Sorry for the late reply on this one, it was a crazy week.

# ¿ Jun 16, 2017 04:16

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

I just spent some time diagnosing a really obnoxious slowdown in linux and came to the conclusion that my NVMe is too loving fast.

The slowdown was when my swap (on a 960 EVO) was full.

I didn't even notice when I had 16GB of memory running effectively off my NVMe. Running out of it was like hitting a brick wall - everything just stops.

# ¿ Jun 13, 2018 19:05

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

peepsalot posted:

Yeah NVMe swap is pretty great and I partitioned about half my 500GB 960 EVO NVME on my laptop for a while to supplement 16GB physical RAM which allowed me to do some particularly memory hungry tasks. I eventually ended up needing a little more disk space so I dropped it to only 128GB swap.

I'm going to backup & resize my parts a bit to give myself 64gb of swap, because it's nice to spool up a VM without hitches.

# ¿ Jun 13, 2018 20:18

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Would anyone consider a used 860 pro/960 pro from amazon warehouse? I figure there's not much you can do to them that won't just make it DoA (and warrantyable).

They come down pretty good in price from time to time.

# ¿ Aug 26, 2018 03:49

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Klyith posted:

Does the warehouse deal make them cheaper than a new Evo or equivalent TLC NVMe drive? Does your use case actually need a Pro?

One of those would have to be a Yes for me to consider it.

I burned out an EVO doing caching on a 12TB array, yes. I could just keep feeding it lower endurance drives but I think the pro will be more cost-effective in the end. Could be wrong!

If there's a cheaper high-endurance option I'm all for that too.

Atomizer posted:

I've had excellent experiences with Warehouse Deals, and items are usually in better shape than described, and I wouldn't be averse to trying an SSD from there if the price was right, however there are so many deals on new SSDs that it'd have to be a hell of a discount to bother. Samsung SSDs are good, just not mandatory, and there are plenty of other good options. The Mushkin I posted is suspected to have the same hardware (controller and NAND) of the MX500, which is considered to be right up there in quality and desirability with a Samsung Evo, for example. Since you're apparently looking for an NVMe drive, presumably you've seen the 960 Pro which has a 512 GB version as a Warehouse Deal for $185; I also recently posted an Adata 480 GB NVMe for $100, and there's no way in hell a used 960 Pro is worth almost double what that new SX8200 is going for (on sale.)

If a 1 TB mushkin would take more writes than a 500gb pro that'd be an overall better option. This will be drive #3 I've fed to it so I'd prefer it to last. At least I'm getting #2 out before it's completely burnt out, I can put it in something lighter duty.

# ¿ Aug 26, 2018 11:07

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Atomizer posted:

Hmm, it sounds like you have a very specific use-case; without revealing this beforehand you're going to get recommendations assuming you're just looking for a normal consumer drive. You'll definitely want an MLC drive (I don't think they make new SLC ones anymore) and likely an enterprise-grade one designed for durability rather than some consumer drive that's primarily designed to be cheap. You want cheap though which isn't compatible with your rather brutal use-case; most people will not burn out their SSDs before they replace the whole system.

SlayVus' Optane recommendation is probably ideal for your situation.

Well, I did only ask if people trusted a used pro from amazon warehouse, not if it was right for normal usecase :v:

One other thing I'm going to check is cache eviction stats - if the small (250gb evo) was causing too many evictions because the common working set was too large, it would lead to excessive wear. So maybe throwing a cheap 1TB at it would improve performance and lead to lower writes.

# ¿ Aug 28, 2018 01:53

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Phison E12 firmware 12.2 is still the latest with no real problems?

I realized I hadn't actually checked my firmware version when I installed my P34A80 so I did it now.

# ¿ Jun 17, 2019 22:22

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

VostokProgram posted:

Still C and C++. Some hard drives are running an entire RTOS in their firmware

There's some very slow movement towards embedded rust because of all these stupid fuckups that keep happening over and over and over, but I don't expect to see anything more than trivial adoption for a decade.

Protip: If you have any sort of counter start it 1 minute from overflow. Trust me on this, it's the only way to be sure everyone handles it correctly.

What's the current non-garbage controller lineup, phison E12, intel, samsung, micron?
E: nVME controllers I mean

Harik fucked around with this message at 06:59 on Dec 2, 2019

# ¿ Dec 2, 2019 06:37

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Anyone know where the SK hynix BC501 ends up on the garbage spectrum? It's the stock NVMe on the lower-end laptops I got my kids, so the only real worry I have is it dying before I get around to upgrading them from 128gb to a 500 or 1tb phison.

HFM128GDJTNG-8310A

# ¿ Dec 10, 2019 21:34

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

What's the minimum SATA SSD you would put in a server to move a database off 15k SAS? Setup is just mirrored disks that get backed up nightly. 860 pro or could I get away with something like a pair of BX500s?

Can't really throw in a PCIe SSD at the moment otherwise I'd get something more enterprisey, and SAS SSDs seem ludicrously overpriced. More than willing to be wrong on the latter if there's a good one in that space.

# ¿ May 12, 2020 22:26

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

ugh, definitely not then. Hate when companies make massive changes without renaming the product.

What about SAS SSDs? The ones I see are tiny compared to consumer drives, 200gb but SLC.

# ¿ May 12, 2020 22:49

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

DrDork posted:

That again depends on your write tolerance needs are. If it's getting hammered constantly and writing TB of data a day, SLC-based drives make sense unless you're ok with replacing dead drives every so often. If it's mostly read activity, then it's probably not needed.

Curious, though: how is it you've got the option of SAS SSDs but not PCIe ones?

That's easy, I've got open 2.5" SAS bays but not enough free PCIe slots to put in pairs for redundancy. So something like the Intel SSD D3 S4510 series?

# ¿ May 13, 2020 01:52

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Max Wilco posted:

It also apparently needs software from Seagate to activate.

it needs what now?

E: oh it comes with dumb snakeoil poo poo that just initalizes the GPT instead of you going to disk manager.

Trying to decide if I want a third NVMe to dedicate to windows (threadripper, so it'd still be on CPU lanes) or just get a big SATA. What's the current good 2TB NVMe with TLC & DRAM? I've only got PCIe 3.0 so I don't need one of the newer 4.0 ones.

I was looking at another silicon power but apparently they're all QLC now? I've got the older (blue sticker) TLC 1TB.

E2: There is a TLC 2TB SP.

Harik fucked around with this message at 01:15 on Apr 19, 2021

# ¿ Apr 18, 2021 23:29

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

anyway are there any issues with the 2TB TLC silicon power NVMe?

Oh gently caress a duck they changed out the controller without changing the SKU. Performance went from 'meh' to 'dogshit'.

It really is samsung or loot box, huh?

E: Lol WTF is this chart even trying to show, is this a parody?
https://www.storagereview.com/review/xpg-gammix-s50-lite-pcie-4-0-ssd-review

E2:

This is pcie3 so i'm not too concerned with getting a 4.0 drive. Threadripper 2920x so next upgrade will probably be threadripper 69x0 series, at which point i'll be buying 4tb PCIe 4 nvme.

Ok, so the sabrent rocket (3.0) has great revie.... oh it's been downgraded without a SKU change too. gently caress capitalism, gently caress marketing, gently caress beancounters.

Harik fucked around with this message at 18:05 on Apr 19, 2021

# ¿ Apr 19, 2021 16:53

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

yeah they cost exactly as much as samsung so might as well just buy the best.

e: or more, lol. black sn850 $380, 980 pro $350.

# ¿ Apr 19, 2021 18:49

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Klyith posted:

The next best for power use after the sk hynix are bog-standard phision e12 drives like an adata XPG 8200 Pro or a HP EX920, those are available in ~256 size.

The XPG 8200 pro was silicon motion SM2262ENG, not phison e12. It's been downgraded twice since the reviews.

E: My lootbox contained a SM2262G for my 512g 8200 pro

So I got an 8200 instead, but for a higher price. YAY ME! Basically they quit making the pro line and just labeled their 8200 production as "pro". gently caress them so hard.

At this point reviewers should just do the right thing and ignore the given hardware and review a $5 USB keyfob for companies playing these lovely games. There's zero reason to put your weight behind a blatant lie.

Harik fucked around with this message at 16:04 on Apr 20, 2021

# ¿ Apr 20, 2021 15:41

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

That's still bouncing through system RAM though, right?

Adata pinky swears they'll sell you the non-gimped model of the 8200 pro that's using the reviewed chipset: https://www.amazon.com/dp/B07TY2TN64

$209 after $30 'click' coupon. I fell for it, I'll know tomorrow if I got what I ordered or if they're getting a very angry return.

E: Got the SM2262ENG as advertised, nice.

Harik fucked around with this message at 01:17 on Apr 24, 2021

# ¿ Apr 22, 2021 18:00

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

For 2.5" SATA to upgrade an older machine, is there any drawback to dramless TLC? The big bottleneck is going to be the SATA interface but does lack of DRAM matter at those slower speeds?

# ¿ Oct 18, 2022 06:15

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Thanks, I grabbed a cheap teamgroup vulcan Z SATA since it's for an older machine.

WhyteRyce posted:

you guys aren't buying U.2 NVMe drives?!

I am. 12-bay hotswap NVMe, all direct-to-cpu thanks to epyc. it's v. fast even with older 4tb u.2 drives. glorious iops for everyone.

Anecdatapoint: DPG SX8200 Pro 2TB died at around 18 months. 90 TBW before it started taking errors. That seems ... a bit early? I was using it more than most do with 45 erase cycles but the smart data didn't say it was bad.

code:

=== START OF INFORMATION SECTION ===
Model Number:                       ADATA SX8200PNP
Serial Number:                      2L0929AA6JYR
Firmware Version:                   42B9T6TB
PCI Vendor/Subsystem ID:            0x1cc1
IEEE OUI Identifier:                0x707c18
Controller ID:                      1
NVMe Version:                       1.3
Number of Namespaces:               1
Namespace 1 Size/Capacity:          2,048,408,248,320 [2.04 TB]
Namespace 1 Utilization:            1,694,576,869,376 [1.69 TB]
Namespace 1 Formatted LBA Size:     512
Namespace 1 IEEE EUI-64:            707c18 00002092aa
Local Time is:                      Sat Oct 29 22:00:43 2022 EDT
Firmware Updates (0x14):            2 Slots, no Reset required
Optional Admin Commands (0x0017):   Security Format Frmw_DL Self_Test
Optional NVM Commands (0x005f):     Comp Wr_Unc DS_Mngmt Wr_Zero Sav/Sel_Feat Timestmp
Log Page Attributes (0x0f):         S/H_per_NS Cmd_Eff_Lg Ext_Get_Lg Telmtry_Lg
Maximum Data Transfer Size:         64 Pages
Warning  Comp. Temp. Threshold:     75 Celsius
Critical Comp. Temp. Threshold:     80 Celsius

=== START OF SMART DATA SECTION ===
SMART overall-health self-assessment test result: PASSED

SMART/Health Information (NVMe Log 0x02)
Critical Warning:                   0x00
Temperature:                        38 Celsius
Available Spare:                    100%
Available Spare Threshold:          10%
Percentage Used:                    8%
Data Units Read:                    55,144,738 [28.2 TB]
Data Units Written:                 182,457,896 [93.4 TB]
Host Read Commands:                 1,098,800,082
Host Write Commands:                1,581,754,121
Controller Busy Time:               22,262
Power Cycles:                       49
Power On Hours:                     13,262
Unsafe Shutdowns:                   26
Media and Data Integrity Errors:    115
Error Information Log Entries:      27
Warning  Comp. Temperature Time:    0
Critical Comp. Temperature Time:    0
Thermal Temp. 1 Transition Count:   57
Thermal Temp. 1 Total Time:         321

Error Information (NVMe Log 0x01, 16 of 256 entries)
Num   ErrCount  SQId   CmdId  Status  PELoc          LBA  NSID    VS
  0         27     1  0x007d  0x4502      -    792809553     1     -
  1         26     1  0x007c  0x4502      -    792809552     1     -
  2         25     1  0x007b  0x4502      -    792809553     1     -
  3         24     1  0x007a  0x4502      -    792809552     1     -
  4         23     1  0x0079  0x4502      -    792809552     1     -
  5         22     1  0x001e  0x4502      -    792755289     1     -
  6         21     1  0x001d  0x4502      -    792755288     1     -
  7         20     1  0x001c  0x4502      -    792755289     1     -
  8         19     1  0x001b  0x4502      -    792755288     1     -
  9         18     1  0x001a  0x4502      -    792755288     1     -
 10         17     1  0x001e  0x4502      -    792664033     1     -
 11         16     1  0x001d  0x4502      -    792664032     1     -
 12         15     1  0x001c  0x4502      -    792664033     1     -
 13         14     1  0x001b  0x4502      -    792664032     1     -
 14         13     1  0x001a  0x4502      -    792664032     1     -
 15         12     1  0x002d  0x4502      -    792504201     1     -
... (240 entries not read)

Lolling that it considers itself healthy while reads keep stalling and it had unrecoverable errors on a couple sectors. This is why I've got nightly backups.

E: Rated for 1280 TBW, which I'm not remotely close to.

E2: Replaced with samsung 980 pro. Waste of my weekend cloning the disk. I don't backup cccache (and other caches) for obvious reasons but i'd prefer to not do 24 hours of builds to get it populated again.

Harik fucked around with this message at 03:34 on Oct 30, 2022

# ¿ Oct 30, 2022 03:23

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Agreed, which is why I'm restoring the lost files from nightly backup instead of raging that they're gone. Now I just need to extend that backup some more since i see some things i could have potentially lost that weren't critical but would have annoyed me to rebuild.

Sadly the clone wasn't viable, too much metadata corruption and it was slowly eating itself. Wiped it again, re-copied the less important stuff over and letting the backup restore overnight.

Going to RMA it and see if they give a reason why it died so soon. It's got a 5-year on it and I've only had it since last april.

# ¿ Oct 31, 2022 04:56

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

what about sk hynix P31 gold?

that drive would be much more annoying to recover so I don't want to be doing it every 18 months.

# ¿ Oct 31, 2022 22:38

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

90s Solo Cup posted:

I'm 0-2 when it comes to Samsung SSDs, so none of this surprises me.

I went with SK Hynix for my current build.

RIP SK Hynix. They were great drives. Too good for hellworld to let live I guess.

I think I'm 1-4 on samsungs, but I still have a 2TB 980 pro (fw 5B2QGXA7) that hasn't died. Yet. Probably just working to spite me because after the third one died I started doing monthly verifications of my nightly backups...

I guess 1.5 for 4, I've got a 840 evo kicking around in limpalong mode as a boot sector/log destination on my NAS. If it fails the result is the same as replacing it: shut down the server, install new drive that I have lying around, reinstall boot sector.

I posted about it in a previous incarnation of the thread and nothing's changed since then. Still ticking and a few weeks away from 9 years on the hour count. It handles light write loads but shits the bed if you hammer it with a bunch. It was originally a cache drive in front of a RAID array but started faulting at around 10TBW on a 256gb drive. Retired from that role but since it had the bootloader for the NAS on it I left it in and let it just handle logs.

# ¿ Jul 13, 2023 05:09

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Cygni posted:

They still exist, they've just been rebranded as Solidigm after their purchase of Intel's SSD unit. Their P44 Pro is one of the most recommended drives around at the moment.

I thought they changed their flash chips as well and the only thing that remained of SK Hynix was the shareholders?

# ¿ Jul 13, 2023 05:28

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Anything other than enterprise U.2 or SAS SSD (or lmao optane) that gets endurance measurements in DWPD? Before "what on earth do you need it for?" see above where I burned out an 840 evo by trying to use it as the cache for a NAS.

There's decent options there (https://www.newegg.com/p/N82E16820147857 - 2TB, 1DWPD * 5 years, $200) but curious if I'm missing anything.

# ¿ Jul 15, 2023 05:27

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Klyith posted:

How big was the 840? Because besides buying an enterprise drive, you could also just buy a big consumer drive and only use 50% of it. Probably about the same price too, unless you get one of the ebay special "like new" enterprise drives.

A lot of the high endurance numbers on enterprise drives comes from having way more overprovision space. For ex, a samsung drive rated for 5 DWPD has 25% more flash than its nominal capacity. Probably the rest is just having the warranty be closer to the drive's true endurance potential, and being willing to replace more early failures. The TBW specs on consumer drives have always felt more like an excuse to deny warranties than anything else.

I don't think there's a special "enterprise nand" at any rate.

VVV edit: yeah better binning and QC definitely, but I mean it's not special like a totally different product or process.

You don't get 10x the write cycles with a bit of clever overprovisioning, you need the overprov because the load on enterprise is going to be random IO (databases) so you're going to be constantly needing to recycle erase blocks. Consumer stuff often holds big vidjagame installs that remain largely unchanged for months at a time, so it doesn't need as much buffer.

it's absolutely a different product because older nodes mean larger cells which mean they can survive more erase cycles. Has enterprise even started using QLC chips yet? They've always been behind the uptake in newer, easer-to-burn-out cells than consumer gear.

---

I've had good luck with PCIe->M.2 NVMe adapters in the past, bringing NVMe to a motherboard that significantly predated the standard.

has anyone here used a PCIe->U.2 NVMe adapter?
https://www.amazon.com/GLOTRENDS-SFF-8639-Adapter-Desktop-Installation/dp/B099185SSV/

I'm happy to have the drive bolted to the PCIe card, reducing cabling is an end goal here.

kliras posted:

at this point, i might just as well get a pcie nvme adapter and retire some drives

Unless your MB supports slot bifurcation remember that it's one drive per slot.

# ¿ Jul 16, 2023 12:15

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Klyith posted:

The tea leaves predict all drives will fail, have backups.

# ¿ Aug 1, 2023 03:34

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

What's the smallest SSD you'd consider 'reliable' nowadays? Use is a tradeshow demo setup where even 64gb is massive overkill. It's off 99% of the time, minimal writes when running and gets shipped around the country by freight a lot.

USB flash-on-a-10-pin-header? No idea what that form factor would be called. I even have a 32gb one that was left plugged into my supermicro when I got it years ago but I'd prefer to buy new.

Industrial-grade NVMe? I'm probably overthinking this and a samsung pro NVMe is fine and we just buy two of the whole thing and if there's a problem they use the other and I fix it after the show.

# ¿ Mar 14, 2024 14:37

Adbot: ADBOT LOVES YOU

# ¿ May 3, 2024 08:38

Harik: Sep 9, 2001; From the hard streets of Moscow
First dog to touch the stars; Plaster Town Cop

Zorak of Michigan posted:

The form factor is called disk on module.

BobHoward posted:

Retention starts out high when the flash is at 0 write cycles, because that's when the insulation around charge traps or floating gates is in excellent condition. As the flash accumulates write cycles, insulation gets damaged, making the cells leakier, which means they fade faster.

However, people want a single number which tells them "this is the minimum guaranteed data retention time". NAND manufacturers spec the retention time based on the NAND having reached its rated write cycle limit, but this means that brand new NAND will always outperform its datasheet retention time, and not by a small amount either.

This brings us to enterprise vs consumer. They do some binning, as I understand it, but that's not enough to produce the huge increases in write cycle endurance that enterprise NAND enjoys. Instead, they're just choosing to pick a point further out on the curve: more write cycles, worse retention. Enterprise customers are okay with retention times that would be terrible in a consumer product, but want the drive to last longer in service before it has to be scrapped, so they're fine with this tradeoff.

So for this trade show demo application, consumer or enterprise makes no difference. Buy any new SSD, don't write to it a lot, and it should have multi-year power off retention regardless of its nominal specs.

Saukkis posted:

I suggest Intel Optane 16GB M.2 NVMe sticks, they are available plentiful on eBay for $5.

Thanks, this is exactly the info I'm looking for. Those optanes might be perfect but either way, I'm happy with the retention of a low-write SSD. Most importantly I don't need to spec the rest of the hardware around the storage since there's options for anything I choose.

# ¿ Mar 19, 2024 19:45

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > SSD Thread: (its good, ignore the anime)