Thread 106907505

eMMC and HDD bros, our response?

>>106907505 (OP)
>our response?
Fuck you.

>>106907505 (OP)
Most SSDs are DRAMless rendering them slower than HDD when copying big files

remember when these were like 4000 bucks

>>106907505 (OP)
>money can't buy time
Who said that, ever? Money lets you buy the services of others to do shit you don't want to waste time doing.

>>106907870
Hello 2012, how is it going?

>>106908800
that still takes time

>>106907505 (OP)
the emmc in my chromebook doesn't even work with linux, I bought a samsung fit plus. 5MB/s 4k random write, 10 times faster than a 5400rpm HDD

>>106908956
you may laugh but way too many sata ssd are being sold dramless to this day
nvme is the only reliable thing you can buy anywhere and rest assured that it won't be a dramless freezefest

>Money can't buy you time
Literally no one says that. It's a made up argument to strawman against

>>106909009
Post chromebook pic with timestamp.

>>106907505 (OP)
>SDD

>>106910073
why

>>106908800
The point is that the initial time investment is designed to be higher than your return unless you have a good safety net like being a jewish nepo baby.

>>106909263
Imagine the time it takes to recover terabytes of irreplaceable data when your allotted 10 writes on the MLC cells are exceeded.

In contrast, hard disc media has a fundamental write durability on the order of 10^12

We have build machines with SSDs in them and they consistently die in 6 to 9 months. There are a bunch of less-obvious advantages to HDDs as well:

In secure environments, pattern-wiping drives does not harm the media. In an SSD it seriously cuts down on the lifespan. On pattern-wipe can destroy a SSD because of the insanely low write durability.
With SSDs, we’re not even sure it’s really wiping the data, it might just be marking it zeroed or something.

>>106910257
ever heard of encryption?

>>106910257
>We have build machines with SSDs in them and they consistently die in 6 to 9 months.
Nigga what?
SSDs dying within a few months would trigger an internal investigation at the Google DC I worked at.
They where concerned when a handful of decade old sata SSDs failed because of a bug in borg
HDDs meanwhile fail in the hundreds a day and no one bats an eye

The hell you retards doing?

>>106910257
>We have build machines with SSDs in them and they consistently die in 6 to 9 months
I've run my games off of my NVME for 3 years now with no issue. I've had an SSD for my OS for almost 10 and it's still running strong.

>>106910196
Just wanted to see your battleship in action. Nice and cool. Only annoyance must the the TN.
Well, happy Winters, dear anona. May God bless you with good health and happiness

>>106910748
> encryption?
Yes, but that is also even riskier for recovery if something goes wrong. You do make good point though. It depends on how the encryption is done, even some of the self-encrypting hard drives has a performance drop — it cut it in exactly half. Which I thought was odd, but that's what the specs said.

>>106910787
> what are you retards doing
A build box absolutely rapes the SSDs.
Windows makes tons of temporary files, .LIBs, logs, PCHs, PDBs, in addition to objs and exes full of debugging information, and some of the linker options rape the swap file due to insane memory requirements, then it all gets copied around, compressed, packaged, archived, signed, and then deleted.
It runs 24/7, sometimes running 4 builds simultaneously. It occasionally clears the backlog by sunday, and gets a few hours of idle time before the east coast comes in.

HDDs used to last years.
My recommendation if you have a SSD is lots of ram, and turn on write caching. You cannot have too much RAM. Buildboxes just build, nothing critical is lost should it die (on write failure)

>>106911743
>A build box absolutely rapes the SSDs.
What models of SSDs?

>>106911743
why not tmpfs?

>>106907870
>Most SSDs are DRAMless
>>106910257
>In contrast, hard disc media has a fundamental write durability on the order of 10^12
>We have build machines with SSDs in them and they consistently die in 6 to 9 months.
>>106911743
>A build box absolutely rapes the SSDs.
pure copium
stop being poor
pic related is orders of magnitude faster, more reliable, and cheaper than HDD's
Sorry you can't into SAS much less NVME

>>106907870
>Most SSDs are DRAMless
Even when this is true
>rendering them slower than HDD when copying big files
This is not true.

>>106911806
> brands
Not always sure… that’s IT…, knowing our company, they’re not industrial grade SLCs though. It was various brands over the years, samsung has come up.

>>106911848
> tmpfs
If you give windows or linux enough free ram, it uses it for disc cache, and it’s a bit difficult to turn on write caching (there is a waiver dialog, lol) on windows.
We tried ramdisks, but surprisingly performance wasn’t that great (although the guy running the tests it was a bit if a tard).
The OS caches the pages before it even gets to the disk subsystems.
>>106911867
> most SSDs are DRAMless
Great, but I’m talking about system RAM, we blow through the dinky RAM on the SSDs in probably seconds. Plus, see above.
It’s generally a bunch of trade-offs.
Either way, RAM is faster and wears on the order of 10^16. Mostly it’s about minimizing the wear.

>>106912181
> Rendering slower to ssds is not true
Agreed, SSDs are usually much faster on build machines but our important storage is written across many HDDs simultaneously in RAID so it’s actually pretty fast (i.e. the source code itself).

>>106912181
^THIS
plus even with the shittiest qlc drives these days, the slc cache is so fucking huge in most cases that you need like 400+ gb of sequential writes before that becomes apparent
and besides even if you had an ssd with only a 5 gb slc cache that goes down to 10 mb/s once that cache is exhausted, that ssd would still absolutely mog an hdd in 99.9% of daily use cases
like an hard drive takes 5 minutes to boot windows, an ssd with an abysmally slow slc cache is still loading windows + your web browser under 10 seconds no matter what, reads aren't affected and most of those reads are also random
basically hdds always suck while a dramless ssd with a small slc cache only sucks sometimes, how the fuck is that worse? no idea, the brain of these third worlders that cannot even afford the shittiest dramless qlc ssd just doesn't exist

>>106907505 (OP)
RAID0 10K RPM spinning rust!

>2025
>not Optane

>>106911743
>A build box absolutely rapes the SSDs.
Don't use consumer QLC drives then?

If you're slamming them that hard, run a bunch of old 3 or 10 drive writes per day drives. S3600, S3610 are 3, S3700, and S3710 are 10, and that's rated for random writes which have heavy internal write amplification. If you're paving out large batches of files together their endurance is potentially upwards of a full order of magnitude higher.

If you need nvme performance, which it doesn't sound like you do since you're using spinners, there's tons of modern options offering upwards of a million IOPS that can handle 3 drive writes a day on 15TB drives. If you're paving out 45TB of garbage every day, you're workload is actually insane and you should have bought optane drives from the start.

There's so many high end used SSDs available on the market that it's honestly nuts not to consider them if you need cheap disposables for torture workloads.

>>106914794
I wish optane were still around, but sadly it's too expensive and too niche of a product.

>>106907505 (OP)
raid

>>106916122
continued
>>106910257
>In secure environments, pattern-wiping drives does not harm the media.
So? Datacenter SSDs usually support secure instant erase, and even the ones that don't can be fully trimmed in a couple of minutes. Yeah there's compliance crap, but even the DoD is typically fine with those options if you clearly ask for permission and they're validated drives. Nobody outside of government actually gives a fuck.

If you want security fast, run them through a shredder or buy a cheap log splitter on craigslist. Drives are cheap enough now that unless you need to shred dozens (or realistically hundreds) of TB a week it should be a rounding error in the budget of a "secure" facility. Stuffs down at 50-60 dollars a TB when you're buying in bulk now. A thousand bucks a week is less than half the salary of a vaguely competent mid level engineer, and less than a third of what he costs the company factoring in taxes, his 401k/pension, and other bureaucratic crap, and that's shedding 60-80TB worth of drives every month.

None of this post makes any sense. The fuck are you people doing?

>>106913183
i mentioned tmpfs because if you're actually chewing through disks you should be able to calculate how much ram you need to have system-only storage and built everything else on volatile and basically never wear it out

>>106907505 (OP)
Who uses HDD as a boot drive in a well functioning country?

>>106907505 (OP)
I don't know, maybe cost/capacity? A 16tb hdd is like $300 while the same ssd is $1500.
I don't really care that my anime takes 100ms to start instead of 10.

>>106907505 (OP)
https://en.wikipedia.org/wiki/Jevons_paradox

CPU speed and multicore is a great example of this
look how much dogshit botnet stuff we have running in the background at all times now
because it's cheap and seems to cost basically nothing (relative to what's already running)
the same is true of SSD and any other finite resource
increase the efficiency and the demand for it increases way beyond whatever your improvement factor was

reminder that seagate and wd stocks will double in value soon

>>106907917
>block your path...

>>106907870
I have several DRAMless SSDs in my servers and they are still faster than hard drives writing terabytes etc. And don't even get started on read speed.