|
Some of my friends also use fio at their work.
|
#
¿
Sep 1, 2016 08:03
|
|
|
|
| # ¿ Apr 23, 2024 11:03 |
|
|
Found something interesting from Microsoft about using NVDIMM-N: https://channel9.msdn.com/events/Build/2016/P470 It's a bit old (back from April), but I don't believe I saw discussion about it.
|
|
#
¿
Sep 22, 2016 16:30
|
|
|
Unlucky7 posted:I know everyone says to ignore the OS optimizations in Magician, but what is this RAPID mode thing? I can't use it since I am on Windows 10, and apparently they do not offer support for it, yet. Is it actually useful? It's glorified ramdisk, don't turn it on.
|
|
#
¿
Oct 1, 2016 22:15
|
|
|
Can I just grab an M.2 PCI-E SSD like a 600p or a 960 EVO, stick it in a adapter and plug it into my MB (ASUS Z87-C) and have it work without any other issues (chipset or whatever) as a data drive? I won't be needing to boot from it, but I have no idea how this poo poo works re PCI-E lanes.
|
|
#
¿
Mar 9, 2017 14:27
|
|
|
Ataxerxes posted:Thanks for the help an info everyone, I got my 850 Evo installed yesterday and also installed Macrium Reflect. My Windows install takes about 20GB, can I transfer it to a partition on the SSD with Macrium? Also, I read that you should leave some space on an SSD unused. Partition management in Windows says there are about 470GB space in the SSD, so does it do that automatically? You see 470 GB rather than 500 because drive makers use SI meaning of Gigabyte (10^9) while Microsoft is calculating the size in Gibibytes (2^30) but denoting it with the wrong unit (GiB vs GB). mmkay fucked around with this message at 09:30 on Mar 11, 2017 |
|
#
¿
Mar 11, 2017 09:28
|
|
|
Synthbuttrange posted:Wierd. I looked at the installed components and it's 122mb, but the installer downloaded a 900mb file. Maybe it had x64/x86 and different OS versions inside?
|
|
#
¿
Apr 29, 2017 09:27
|
|
|
Saukkis posted:Or if you are not ready to upgrade yet, then install Win 10 on the NVMe, upgrade the firmware and wipe the NVMe. Would this actually work? I didn't think Sandy Bridge era motherboards had UEFI drivers capable of booting from NVMe drives.
|
|
#
¿
Dec 17, 2018 11:59
|
|
|
Klyith posted:Nah that "SSD flash uses 4k pages, so align your partition to 4k" thing is common advice for SSDs, you're not crazy. I believed that was a fact until 10 minutes ago too. I just suddenly had the wait a second moment thinking if a flash cell doesn't even have a constant number of bits in it, the larger structures probably weren't static either. You can read up on Open Channel specification if you're interested in what the firmware in a regular SSD generally has to deal with.
|
|
#
¿
Mar 3, 2019 21:42
|
|
|
To throw some more info into the mix - the drive generally wants to do the garbage collection as late as possible (though not too late) - it reduces the amount of data it needs to relocate. Though it probably matters more for data center SSDs. Also I think this is the first time I saw someone else mention Open Channel - these are pretty cool, you get to learn all the geometry and media management bullshit a NAND drive has to keep up with and it's very enlightening.
|
|
#
¿
Aug 17, 2020 07:01
|
|
|
Also when doing random write tests on NAND SSDs you ideally want to fill up the drive beforehand, then start doing random writes for some time and THEN actually check the performance. The results with/without garbage collection in the drive can be pretty big as your writes are going to be contending with internal read/write/erases, or just going straight to the media when the drive is mostly empty, or not really internally 'defragmented'. Not that it really should matter for your average use case, you don't do enough 4k random writes to really notice the shift in performance.
|
|
#
¿
Dec 11, 2020 21:57
|
|
|
For consumers without any intensive workloads - low qd (like 4 at max) random read/writes (possibly mixed).
|
|
#
¿
Dec 19, 2020 10:47
|
|
|
priznat posted:I was thinking about it more, Optane still uses DRAM right? I'm interested in drives for PCIe benchmarking but most NVMe SSDs are a pain because you have to precondition them before you can run write benchmarks. And then the performance drop off due to wear levelling after a lot of that, which Optane will mitigate. But the preconditioning, blargh. Currently use NVRAM devices because capacity doesn't matter just want raw speed! Only available in Gen3 though. The reason NAND SSDs need preconditioning is that (especially) during random write workloads the data inside the drive will be increasingly defragmented (valid/invalid blocks) and it will need to relocate some of it in order to erase/program the next user input. This a) takes up bandwidth available to the user b) until the distribution of valid/invalid blocks is uniform across NAND dies, the performance will be unstable. Both NAND and Optane still will do relocations due to wear leveling, but it's not anywhere near as significant as garbage collecting. I don't believe Optane (SSDs) has the same magnitude of DRAM inside the drive as NAND (assuming this is what you mean), but don't quote me on that. The PM modules have DRAM for sure, but I don't recall any ratios there.
|
|
#
¿
Apr 7, 2021 07:44
|
|
|
Not really, most of the day to day workloads is low qd random i/o which won't let you use the bandwidth advantages. The latency boost is going to be there, but you probably won't notice it. Also unless you copy movies back and forth all day long, you won't notice the sequential speed difference between Samsung and WD, so don't base your choice on that alone. mmkay fucked around with this message at 06:44 on Jul 2, 2021 |
|
#
¿
Jul 2, 2021 06:42
|
|
|
High queue depth sequential writes+reads. Or if you want to just see some stupid, actually meaningless results (since the drive won't be doing anything, while it's still empty), just sequential reads.
|
|
#
¿
Oct 6, 2021 19:18
|
|
|
Also note that manufacturers will use the SI units (so Giga = 10 ^ 9), while Windows prints the SI symbol, but actually means the binary one (Gibi = 2 ^ 30).
|
|
#
¿
Nov 1, 2021 10:34
|
|
|
Most of the lifespan of an SSD is tied to how many writes per day it can do for the warranty period. Current QLC drives have a value around 0.1-0.2 (so for a 1TB drive you can write 100-200GB every day for 5 years without issues), while older TLC ones have a value of 1 or so. The TLC drives would probably also have been rated for 5 years, just with a higher WPD level. I don't remember though if that value is derived from the worst case (4K random writes use up the drive the most), or average (some mix with sequential). If you're writing that much data in a day, you're probably writing sequentially (assuming typical home usage).
|
|
#
¿
Dec 23, 2021 07:50
|
|
|
In an ideal world those would just be new models.
|
|
#
¿
May 23, 2022 06:30
|
|
|
Which tool can dump this NAND geometry info?
|
|
#
¿
Jan 18, 2023 13:03
|
|
|
It's different media - you store 4 bits per cell (Quad) versus 3 (Triple). It's cheaper/bigger for the same price, but it degrades more easily and performs slightly worse.
|
|
#
¿
Feb 25, 2023 10:54
|
|
|
What's the goto option of 1-2TB M.2 SSD? It's going to be mainly storing Steam games and some random crap I have on my HDD (that I'm hoping to replace with this), so it doesn't really need massive performance (or write endurance I guess).
|
|
#
¿
Mar 31, 2023 08:14
|
|
|
Thanks for the suggestions. My other SSDs are actually old 840 and 850 EVO SATA drives (750GB total), so nothing special. As mentioned the only purpose I'll ever have for this is to replace the spinning HDD, so I'm fine with QLCs. Eventually I might get around to upgrading my boot drive to something speedier/to get rid of the SATA cables if I ever feel like it.
|
|
#
¿
Mar 31, 2023 16:02
|
|
|
From what I remember from a video (https://youtu.be/8YBeriMsDS0) it uses slightly more DRAM than Microsoft's default max value allows for (64 MB vs 70-80 IIRC), but I don't remember the reasoning. It also detects sequential read workloads and preloads them into the OS (which is where some of that RAM usage might go into) Might be some other stuff, but these are the things I remember.
|
|
#
¿
May 19, 2023 07:06
|
|
|
Shaocaholica posted:Is SSD latency getting better generation over generation or has it stagnated for a while? It's getting worse. Each increase of the number of stored bits (SLC > MLC > TLC > QLC) increases capacity, but it means you need to do more work to program/read the data.
|
|
#
¿
May 20, 2023 07:37
|
|
|
priznat posted:
They're still around (Solidigm).
|
|
#
¿
Jun 27, 2023 06:46
|
|
|
What do you mean by longevity? The drives top out performance at 4 gen 4 PCIe lanes, so 5 seems like a waste. But if you're not using them, then you might as well stick them in, you can just move them around later if you buy a Gen 5 NVMe drive.
|
|
#
¿
Jul 7, 2023 12:32
|
|
|
Klyith posted:If you installed windows fresh, it will be able to do UEFI. Just flip the BIOS switch and it should pick up the UEFI boot partition. I had a Windows 7 install that I was slowly upgrading in versions over the years, but there's a Microsoft tool that can change the boot partition from MBR to GPT. I don't know the exact dark magic details it needs to do (seems like there's a lot of stuff that can break if you need to shuffle around partition and file system metadata), but it can do a dry run IIRC to see if it's at least possible.
|
|
#
¿
Sep 22, 2023 07:03
|
|
|
priznat posted:Only intel and micron were interested in producing it, and frankly NAND lifespan is still plenty good especially with a lot of redundancy for enterprise products. I think performance was better for the DRAM modules, but RAM prices fell just as it entered the market, so it was squeezed out off the memory mode. And the persistent memory mode needed stack adaptation, so it was too slow to get going.
|
|
#
¿
Mar 18, 2024 08:41
|
|
|
|
| # ¿ Apr 23, 2024 11:03 |
|
|
WhyteRyce posted:ZNS seems pretty dead in favor of FDP because of the hard requirement of a complete rewrite even though the former gives you a real cost advantage.
|
|
#
¿
Mar 18, 2024 16:47
|
|