SSD Thread: (its good, ignore the anime)

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

WhyteRyce: Dec 30, 2001

Even in Samsung's white papers it didn't show it beating Intel in QoS and still suffered (unsurprisingly) from the same read/write speed imbalance as typical NAND drives. The big threats are cost and it being good/close enough to those key Optane performance metrics. The QoS numbers looks good but will have to hold up

And at this point it's basically gen 1 3dxp vs. extremely polished and understood NAND

WhyteRyce fucked around with this message at 18:03 on Mar 8, 2018

# ¿ Mar 8, 2018 17:59

Adbot: ADBOT LOVES YOU

# ¿ May 13, 2024 06:48

WhyteRyce: Dec 30, 2001

Optane 800p reviews going up
https://techreport.com/review/33338/intel-optane-ssd-800p-58-gb-and-118-gb-solid-state-drives-reviewed
http://www.tomshardware.com/reviews/intel-optane-ssd-800p,5497.html
https://www.tweaktown.com/reviews/8545/intel-optane-ssd-800p-58gb-118gb-2-nvme-pcie-review/index.html

WhyteRyce fucked around with this message at 18:50 on Mar 8, 2018

# ¿ Mar 8, 2018 18:43

WhyteRyce: Dec 30, 2001

These 800ps can still apparently be used as a cache but they are (barely for the 60) large enough to be used as a regular storage device

# ¿ Mar 8, 2018 18:59

WhyteRyce: Dec 30, 2001

Connectors and protocols being decoupled are all the rage now

# ¿ Mar 16, 2018 06:37

WhyteRyce: Dec 30, 2001

Thermal management is something certain types of customers care very much about

# ¿ May 13, 2018 16:30

WhyteRyce: Dec 30, 2001

Oh they could have definitely messed up the implementation but calling it something marketing wanted might be misleading if it's something a customer specially asks for

# ¿ May 13, 2018 21:18

WhyteRyce: Dec 30, 2001

NVMe is great but somehow consumer enthusiasts and gamers think they are the only high end users in the world

# ¿ Oct 7, 2018 04:20

WhyteRyce: Dec 30, 2001

BIG HEADLINE posted:

They're good for moving large files, but unfortunately most gaming/general computing is still geared toward small files that HDDs have an easier time handling because your grandma thinks an SSD is some ~soshalist~ thing and she doesn't want it in her 'puter, dagnabit.

Not just speed, there is a whole bunch of admin configuration ability options for all sorts of stuff that enterprise customers want and demand.

# ¿ Oct 7, 2018 15:36

WhyteRyce: Dec 30, 2001

I think dual port is for high availability (connect to multiple hosts) not high bandwidth

WhyteRyce fucked around with this message at 00:39 on Oct 21, 2018

# ¿ Oct 21, 2018 00:27

WhyteRyce: Dec 30, 2001

I just want 2.5 or 5 gigabit to take off

# ¿ Nov 22, 2018 21:05

WhyteRyce: Dec 30, 2001

Why do people stress over the cell write lifespan when HW and FW design purposely build a solution around that and the overall drive write per day lifespan is a sufficient measurement for most cases?

# ¿ Nov 24, 2018 04:41

WhyteRyce: Dec 30, 2001

Encrypted posted:

The problem with intel's 660p is its failure mode when the drive reaches it's conservative EOL estimate, as the drive simply locks up and goes into a read only mode.

This is going to be even more troublesome for recovering a nvme instead of the regular ol 2.5"

Just curious, but why? Read only mode isn't dead mode and couldn't you just read/clone off the data? I've never had to recover an EOL drive but I'd imagine that's preferable to a drive passing it's EOL and just letting you continue to write to the drive and gently caress things up

# ¿ Nov 24, 2018 04:45

WhyteRyce: Dec 30, 2001

DrDork posted:

With a SATA drive, you can get your replacement drive, slap it in, install an OS, recover off your locked drive and off you go. With a NVMe drive, you most likely only have one m.2 slot, so you have to grab a SATA drive as a temp deal, or find a working system with an open m.2 slot. Obviously not impossible, but more annoying than before.

e; the Intel drive behavior is dumb for home users because it will probably be fine for far more than 200TB; most drives don't lock until they actually start recording and suffering errors, rather than meeting some super conservative semi-arbitrary write count. That behavior makes far more sense in an enterprise environment, though.

It's "arbitrary" in the sense that it's probably what the drive is spec'd and warrantied for so Intel doesn't want you going beyond those limits. Could it go beyond that? Probably. Does Intel make any promises or guarantees for the drive operating outside of spec? No. Do I see scenarios where users decide to keep running it like a car with the fuel light that just went on and then end up having something bad happen that they then want to blame someone for? Absolutely.

Are we even at the point where Joe Consumer is even hitting this drive EOL threshold at a frequency and speed soon enough anyway?

# ¿ Nov 24, 2018 06:35

WhyteRyce: Dec 30, 2001

BIG HEADLINE posted:

My parents' current hand-me-down still has an Intel X-25M in it because the machine *only* has SATA II. It's fine, though I admit I should probably check it's health with Intel's SSD Toolbox the next time I'm over there.

Granted, though - they don't fit what you'd call a heavy duty I/O usage case.

Yeah and that's like a decade old and if it went into some kind of read-only mode after that long you'd be hard pressed to claim planned obsolescence. And even then it's still a preferable alternative to your SSD just eating itself alive via corruption or your spinning rust getting a click of death that your SMART monitor might not have warned you in time over.

# ¿ Nov 24, 2018 07:21

WhyteRyce: Dec 30, 2001

Real pros would have an nvme drive and just change the throttle thresholds themselves

# ¿ Dec 8, 2018 05:36

WhyteRyce: Dec 30, 2001

Is this like when people had issues getting AHCI to work on Windows XP

# ¿ Dec 16, 2018 21:45

WhyteRyce: Dec 30, 2001

I have a feeling that the number of people that know enough to over provision their drives to preserve performance but also are too cheap to buy a bigger SSD when they run right up to the (now lower) capacity limit is not that huge

Or maybe it is larger than I thought but at that point you have no one to blame but yourself

# ¿ Jan 22, 2019 19:17

WhyteRyce: Dec 30, 2001

MLC for 2, TLC, and QLC are what everyone else uses so don't be a jerk and just stick with it

# ¿ Jan 23, 2019 21:00

WhyteRyce: Dec 30, 2001

priznat posted:

Just get a SATA 860 evo yeah, that�s how I roll.

Anyone have a chart with nvme vs sata price per gig? I�m wondering when they�re going to converge. There�s not much reason nvme should be more other than a bit more work on the pcb side running 4 rx tx lanes vs 1. This�d also be the added pins on the controller IC, slightly higher bus speeds etc.

SATA controllers are old, well known, and (I'm assuming) require little to no tweaking. In general, the NVMe spec is still growing, new features being added in, and controllers are being refreshed and updated. Price premium is a thing that is milked hard but there is lots of actual work that is going on that you never see because you don't really care about a bunch of stuff the NVMe spec lets you do

WhyteRyce fucked around with this message at 02:39 on Feb 27, 2019

# ¿ Feb 27, 2019 02:31

WhyteRyce: Dec 30, 2001

I got cheap once and bought an open box mobo from Newegg and of course the pins were bent

# ¿ Feb 27, 2019 04:01

WhyteRyce: Dec 30, 2001

I believe the recommended configuration for Optane is either polling mode or hybrid polling because interrupt mode can't keep up

# ¿ May 28, 2019 07:07

WhyteRyce: Dec 30, 2001

There is pulling ones and zeroes off, and then there is actually pulling off data into something that is actually readable or usable. How much of the later actually happens?

But yeah you kind of are at the mercy of the company implementing things correctly if you rely on sanitize and secure erase. I think nvme 1.3 brings in multiple santize operation types which you can kick off yourself

WhyteRyce fucked around with this message at 21:27 on Jun 1, 2019

# ¿ Jun 1, 2019 21:21

WhyteRyce: Dec 30, 2001

I believe the nvme 1.3 santize overwrite option will do all physical blocks not just the logical range so good enough should be more than good enough for most people

WhyteRyce fucked around with this message at 01:47 on Jun 2, 2019

# ¿ Jun 2, 2019 01:42

WhyteRyce: Dec 30, 2001

22110 only for me plz

# ¿ Jun 22, 2019 08:07

WhyteRyce: Dec 30, 2001

Combat Pretzel posted:

Yeh. But even proper arguments lead to nothing.

I had a discussion with someone earlier speculating how PCIe 4.0 would improve with "4K editing" so hard. Inquiring about his actual bandwidth usage, I get something like he's at "90% utilization" according to task manager. So I tell him that means nothing, because that metric gets derived from IO response times and queue depth. Repeatedly asking about actual reported bandwidth usage, essentially wanting to point out that if one of the best SSD controllers on the market (claimed to have an 970 Pro) can't reach the advertised maximum bandwidth under realistic workloads, neither would a PCIe 4.0 SSD, led to absolutely nothing. But Corsair says on their page it'll do 5GB/s.

It's maybe a smidge frustrating.

And no, it wasn't in r/pcmasterrace.

It's usually the same people that say "hey why is Optane slower than this Samsung drive"

# ¿ Jun 26, 2019 05:24

WhyteRyce: Dec 30, 2001

Potato Salad posted:

The future of consoles is game cartridges with 3dxpoint nonvolatile ram FIGHT ME

it would be the only console with the right type of RAM

# ¿ Jun 26, 2019 23:08

WhyteRyce: Dec 30, 2001

Malcolm XML posted:

I wouldn't trust the on drive encryption. https://www.kb.cert.org/vuls/id/395981/

I mean, the thing you linked to shows drives that are not affected by this specific issue and drives that are patched

# ¿ Jul 29, 2019 18:49

WhyteRyce: Dec 30, 2001

Potato Salad posted:

The point is that manufacturer crypto implementation is a black box that:
1) most importantly, is almost certainly logically flawed
2) less importantly, has state backdoors

Ok so dump your Intel and AMD platforms while you are at it too

Assuming that just because that one product is broken and therefore all are broken and that companies can't improve and learn is off

WhyteRyce fucked around with this message at 19:20 on Jul 29, 2019

# ¿ Jul 29, 2019 19:16

WhyteRyce: Dec 30, 2001

BangersInMyKnickers posted:

I starting writing a post to explain the difference between software and hardware and the fundamental issue with this type of hardware encryption implementation from companies where the functionality is secondary to their core business but I'm just going to say: lol dipshit

The core function is to sell a product to consumers. Big customers who care about certain features or just force it to happen and demand things work and will drop their business with you if you blow raspberries in their face. When things like this break you can be sure that someone's rear end is getting chewed out and that same chewed up rear end is then bitching out the engineering managers who are then forced to figure out how this happened and what they do to improve it. You can remain paranoid your entire life of everything and that's fine but don't assume things will never get better or companies don't have any reason to ever fix things or any incentive to do things right

WhyteRyce fucked around with this message at 19:32 on Jul 29, 2019

# ¿ Jul 29, 2019 19:24

WhyteRyce: Dec 30, 2001

EoRaptor posted:

You get what you pay for.

I kind of bristle at these comments because security is something a lot of big enterprise customers care about and do ask and pay for via product requirements and sessions where they beat up on some poor customer engineer and verify with their own quals. Now companies can do a really bad job at it and not know what they are doing when they try (or just do a quick hack job in some low margin client drive so YOLO), but the implication that there is no focus or effort or customer ask on it or someone won't get bitched out for loving something up is false.

WhyteRyce fucked around with this message at 22:52 on Sep 23, 2019

# ¿ Sep 23, 2019 22:43

WhyteRyce: Dec 30, 2001

Seamonster posted:

Here's to hoping 660p prices just tank. Can't wait to get rid of SATA cables once and for all.

Your monkey paw wish results in everyone moving to U.2

# ¿ Sep 26, 2019 23:52

WhyteRyce: Dec 30, 2001

priznat posted:

Well, same here, and why?

I assumed that most hardware labs look like Frankenstein monstrosities

# ¿ Sep 30, 2019 06:57

WhyteRyce: Dec 30, 2001

Let's talk about all the cool stuff you have to build with cardboard boxes and nylon standoffs

# ¿ Sep 30, 2019 15:02

WhyteRyce: Dec 30, 2001

DLC when

# ¿ Sep 30, 2019 16:55

WhyteRyce: Dec 30, 2001

Bob Morales posted:

As if QLC wasn't slow enough. Gotta stay competitive with SMR hard drives!

Honestly, if PLC leads to some 32TB consumer level drives I'm fine with that

# ¿ Sep 30, 2019 19:12

WhyteRyce: Dec 30, 2001

Peanut3141 posted:

This isn't going to happen. It's only 25% more dense than QLC, while being 1/3 the write speed if the TLC->QLC experience is anything to go by.

32TB was more hyperbole on my part but SSDs that huge capacity drives that rival spinning drives already exist but are in the enterprise line of drives. I think a lot of people will be fine with rotational drive performance in cold, cold storage cases if the cost/density is compelling

# ¿ Sep 30, 2019 23:05

WhyteRyce: Dec 30, 2001

Peanut3141 posted:

SSDs with 32+ levels per cell is not a good option for cold storage. The cells will lose their charge over time and this will happen far more quickly than HDDs, tape or optical media degrade. [timg]This AnandTech article[/timg] is a bit dated, but 3-12 months was the nominal deterioration time for unpowered NAND when we were living in an MLC/TLC world. QLC/PLC will not fare as well.

SSDs excel at random read and write, which QLC/PLC compromise for a small gain in density. The use case is baffling.

Cold tier storage as in data that is not accessed frequently and latency is ok, not stuff meant for actually storing offline

WhyteRyce fucked around with this message at 01:30 on Oct 1, 2019

# ¿ Oct 1, 2019 01:26

WhyteRyce: Dec 30, 2001

ilkhan posted:

Enterprise is just fine with spinning disk level linear performance and ssd level random performance. Just keep stacking them wider until you get your required linear performance.

I'm not saying it's for improving cold tier storage performance, just that customers are already expecting a certain level of performance at this tier so if you can get compelling capacity/cost (and acceptable endurance) then you'll have customers. Not every SSD has to have amazing SLC performance to have a customer, that stuff is for your hot tier

# ¿ Oct 1, 2019 01:55

WhyteRyce: Dec 30, 2001

You guys do know that QLC exists in Intel drives other than the cheap client stuff. And that these dense enterprise drives are specifically targeted at HDD replacement and storage consolidation scenarios in enterprise environments. PLC is probably going to be targeted at these same areas

Not everything has to be targeted as a flag ship performance product. Even as a home user I'd love a big giant slow SSD for my non performance critical stuff if the price and capacity was compelling vs those schucked WDs I'm using

QLC/PLC has significant trade-offs but it's not targeted at every situation

WhyteRyce fucked around with this message at 04:04 on Oct 2, 2019

# ¿ Oct 2, 2019 03:40

Adbot: ADBOT LOVES YOU

# ¿ May 13, 2024 06:48

WhyteRyce: Dec 30, 2001

Atomizer posted:

Yeah but the raw QLC NAND performance is still mediocre, to say nothing of PLC. Plus, have you seen the price of the D5-P4x20? 8 TB for well over $1k isn't what I'd consider a compelling price.

Enterprise drives cost more than consumer drives so not a 1:1 comparison to consumer costs and you're paying for more than just the raw NAND. If you want to try and compare within the same product family, I think the QLC varients were roughly $500~$600 less than their TLC equivalent 8TB drives

# ¿ Oct 2, 2019 04:16

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

«‹›7 »