What are good value SSDs in 2019? 1TB / Phison E12 + Toshiba - Page 10 - Overclock.net - An Overclocking Community

Forum Jump: 

What are good value SSDs in 2019? 1TB / Phison E12 + Toshiba

Reply
 
Thread Tools
post #91 of 134 (permalink) Old 07-29-2019, 04:27 PM
New to Overclock.net
 
Omnidyne's Avatar
 
Join Date: Jan 2018
Location: Louisiana
Posts: 16
Rep: 3 (Unique: 2)
Quote: Originally Posted by Sean Webster View Post
Most of the time the SLC cache clears within minutes. The 1TB Intel 660P takes like 20-30minutes to fold the SLC->QLC. But, remember, when the SLC cache is full, the data on the cache is also faster than the native TLC or QLC. So reads do get a small boost, folding out data isn't always something that is beneficial. Crucial's P1, for example, evacuates data from the cache faster than the 660P in favor of write performance, but that hurts it heavily in PCMark 8 and other application tests.
Hey Sean, just wanted to say thank you for your coverage at Tom's and The SSD Review. You and Chris Ramseyer have provided some quality content over the years.

I have noticed there is a lot of confusion in this forum (and naturally, elsewhere) about QLC NAND. In particular, its performance, given endurance and "reliability". All reviews of the 660p, including yours, for example, gave favorable ratings of the 660p, going as far as to recommend it. As with other review sites, your only criticism being endurance:

"The 660p is a refined and welcome update to Intel's SSD 6 series. The 660p proves that Intel’s QLC NAND is ready for the mainstream and easily earns our top value recommendation for the low-endurance/budget segment. If you haven’t upgraded to an SSD yet and your workloads tend to be mundane, Intel’s SSD 660p is a good fit."

"The Intel SSD 660p offers high capacity, solid performance in real-world workloads, and impressive power efficiency. With prices at just $0.20-per-GB, nothing on the market comes close to it in value, but our recommendation comes with a caveat. You should select a drive with higher endurance if you commonly use productivity applications or have heavy workloads."

There are statements that the 660p suffers from "hard-disk like" performance or that the typical consumer (OS or gaming) would be better off purchasing a hard disk over the 660p or even that the 660p isn't suitable for OS usage. Are any of those statements accurate? I have been unable to find one metric where the 660p performs similar to a hard disk, even under Anandtechs extremely heavy testing conditions. I find nothing of the sort in your review. Every review shows the 660p far outpacing hard disks in every metric.

As for endurance, you state that the lower endurance of the 660p is a non-issue for most consumers:

"In reality, most consumers don’t need that much endurance if their average use case involves mostly office applications, web browsing, and content streaming."

My understanding is that the endurance rating of SSDs in general is a non-issue with the exception being enterprise or specific heavy workloads. Even the DRAM-less (without HMB) WD Blue SN500 carries a respectable endurance rating. I'm under the impression that the SLC cache buffer largely plays a role in increasing endurance significantly because pSLC enjoys higher endurance than TLC NAND but lower than true SLC. Essentially, 1GB of writes to an SSD does not equal 1GB of endurance lost/ used.

Hopefully you can clear this up for everyone and again, thank you for your contributions.

Intel Core i5-8400 | Gigabyte Z370P D3 | Corsair CX550M | NVIDIA GTX 970 Founders Edition | ADATA SX850 256GB - SandDisk MLC 120GB | WD 1TB | G.SKILL Ripjaws V Series 16GB (4 x 4GB) DDR4 2666 | Windows 10 | Corsair Carbide SPEC-03
Alien Space Heater
(12 items)
CPU
i5-8400
Motherboard
Gigabyte Z370P D3
GPU
GTX 970 Founders Edition
RAM
G.Skill Ripjaws V
Hard Drive
ADATA SX850
Hard Drive
Western Digital Blue
Power Supply
Corsair CX550M
Case
Corsair Spec 03
Operating System
Windows 10
Monitor
Acer SA230
Keyboard
Gigabyte Force K83 Cherry MX Red
Mouse
Logitech
▲ hide details ▲

Last edited by Omnidyne; 07-29-2019 at 04:49 PM.
Omnidyne is offline  
Sponsored Links
Advertisement
 
post #92 of 134 (permalink) Old 07-29-2019, 10:43 PM
Storage Nut
 
Sean Webster's Avatar
 
Join Date: Nov 2010
Location: Boca Raton, FL
Posts: 21,076
Rep: 1660 (Unique: 967)
Quote: Originally Posted by JackCY View Post
Update to 12.3 did not change the temperature sensor reading, it stayed the same about 15C below what it should be.

Uploaded the 12.3 update to OP.
Thanks for the ideas! Btw, if you have any questions on anything or feel to chat, feel free to PM me or direct me to a post.

Quote: Originally Posted by ThrashZone View Post
Hi,
Got the samsung 970 evo
Didn't get a screen shot but read was just over spec's 3550 read CDM
Write was also just over spec's at 2350
Was a little toastier than the other

Get to return the 8200 pro's to Kohl's for free return trip
Once i get a 325.us gift card credit I'll pick up 1-970 plus 500gb and 1-970 evo 1b for storage maybe 2tb
Although 860 evo 1tb are stupid cheap I saw yesterday.
Sweet! I'd say to just get the biggest single drive you can get. I have gone through too many. Now I just have a 12Gb/s SAS SSD array, a broken 1.5TB Optane SSD 905P, and a random NVMe boot drive from my box of them. lol

Quote: Originally Posted by Omnidyne View Post
Hey Sean, just wanted to say thank you for your coverage at Tom's and The SSD Review. You and Chris Ramseyer have provided some quality content over the years.

I have noticed there is a lot of confusion in this forum (and naturally, elsewhere) about QLC NAND. In particular, its performance, given endurance and "reliability". All reviews of the 660p, including yours, for example, gave favorable ratings of the 660p, going as far as to recommend it. As with other review sites, your only criticism being endurance:

"The 660p is a refined and welcome update to Intel's SSD 6 series. The 660p proves that Intel’s QLC NAND is ready for the mainstream and easily earns our top value recommendation for the low-endurance/budget segment. If you haven’t upgraded to an SSD yet and your workloads tend to be mundane, Intel’s SSD 660p is a good fit."

"The Intel SSD 660p offers high capacity, solid performance in real-world workloads, and impressive power efficiency. With prices at just $0.20-per-GB, nothing on the market comes close to it in value, but our recommendation comes with a caveat. You should select a drive with higher endurance if you commonly use productivity applications or have heavy workloads."

There are statements that the 660p suffers from "hard-disk like" performance or that the typical consumer (OS or gaming) would be better off purchasing a hard disk over the 660p or even that the 660p isn't suitable for OS usage. Are any of those statements accurate? I have been unable to find one metric where the 660p performs similar to a hard disk, even under Anandtechs extremely heavy testing conditions. I find nothing of the sort in your review. Every review shows the 660p far outpacing hard disks in every metric.
Thank you for the compliment!

To answer your Q: Yes and no. People who state that it is like an HDD or that you are better off with an HDD are one of two people i'd say. Elitists who just want to talk trash because it's not the best. Or, people who don't understand how their workloads really work and what is actually needed to satisfy it and just reiterate what those elitists say. (which is far too common)

An SSD has no moving parts, while HDDs do. This alone makes SSDs hundreds of times faster than HDDs, just look at the random latency differences. And, in addition to latency, HDDs also have seek time, which adds even more time to tasks.

Measured latency of a 2.5" SATA external HDD vs 2.5" SATA external SSD using iometer:
4K random QD1 read --> 8.2ms vs 0.16ms (some SSDs are as fast as 0.02-0.03ms)
4K random QD1 write -> 3.2ms vs 0.09ms (And that's with the HDD at its fastest. In use, they can be much more latent, although, so can some crappy SSDs)

Sure, direct QLC writes are horrible. But, with the pSLC cache in place, most daily tasks like I noted before only need up to 2-3GB MAX. And that QLC buffer empty within ~20 minutes of running TRIM or Intel's toolbox SLC cache clearing feature, that is if you filled it that is. Just restoring the static cache alone is much quicker than the dynamic cache. TLC SSDs with small static caches refill the buffer much faster that QLC drives too. Within a few minutes, sometimes within tens of seconds.

But now, let's refocus. This talk has just been about writes, right? A lot of what we need an SSD for in our applications is actually read performance. QLC can kinda keep up at that at all times. Although, as it's been said, TLC SSDs can be faster. Just depends on what you're doing.

6.5GB zip file read:
1TB Intel SSD 660P - ~1.25GB/s
1TB Samsung 970 EVO- ~2.25GB/s
1TB Crucial MX500 - ~500MB/s

Final Fantasy XIV StormBlood benchmark - total load time of 5 scenes/levels:
1TB Intel SSD 660P - 19.947s
1TB Samsung 970 EVO - 18.661s
1TB Crucial MX500 - 20.938s

Game load times are surprising. After speaking with Allyn from PC Perspective (now at Intel), he told me that he found that most games load via 4K sequential. So, usually, the drives with the fastest 4K read speed will have load the game (and applications) the fastest.

Quote: Originally Posted by Omnidyne View Post
As for endurance, you state that the lower endurance of the 660p is a non-issue for most consumers:

"In reality, most consumers don’t need that much endurance if their average use case involves mostly office applications, web browsing, and content streaming."

My understanding is that the endurance rating of SSDs in general is a non-issue with the exception being enterprise or specific heavy workloads. Even the DRAM-less (without HMB) WD Blue SN500 carries a respectable endurance rating. I'm under the impression that the SLC cache buffer largely plays a role in increasing endurance significantly because pSLC enjoys higher endurance than TLC NAND but lower than true SLC. Essentially, 1GB of writes to an SSD does not equal 1GB of endurance lost/ used.

Hopefully, you can clear this up for everyone and again, thank you for your contributions.
Yeah, usually consumers will write ~5-10TB of data to the drive a year. So, you can typically just figure out what you need from that. Really, for most it isn't a factor to worry about too much. Sometimes its there it just to have an extra level of warranty limitation to reduce a company's RMA liability. Often times SSDs can exceed their ratings, it just takes a long time lol.

Write amplification (WA) varies as you stated, depending on the pSLC cache use and workloads (sequential vs random). Usually, WA for desktop usage usually isn't too high, especially with write caching enabled by default on Windows (helps make random writes more sequential). Thus, people will usually see maybe 1-3x actual writes, meaning 1GB of writes turns into 1-3GB of actual NAND writes. Sequential tasks =~1:1 while random can vary greatly 1:4 for example. Garbage collection and other background management tasks can also increase actual NAND writes too. Controllers that integrate real-time compression, like SandForce controllers of days past, can even hit negative WA values.

I AM THE SSD BENCHMARK GURU
Storage Essentials | My Build Logs | My Server Build Log | flickr | Steam | Depression and I
Some say he once took a backup of the entire NSA database. Others say he lays SSDs like chickens lay eggs. All we know is, it's Sean Webster!
Le Main Rig
(15 items)
Terrorbyte V2
(16 items)
Test Bench
(8 items)
CPU
i7-6700K
Motherboard
ASRock Z170 OC Formula
GPU
Gigabyte GTX 960 WindForce 2X OC
RAM
G.SKILL Ripjaws V
Hard Drive
480GB ADATA SX8200
Hard Drive
1TB Samsung 960 EVO
Hard Drive
2TB HP EX950
Power Supply
Corsair RM650
Cooling
Noctua NH-U12S
Case
Fractal Design R5
Operating System
Windows 10 Pro 64-Bit
Monitor
BenQ GW2765HT
Monitor
BenQ GW2765HT
Keyboard
Patriot Viper V760
Mouse
Patriot Viper V560
CPU
Intel i7 4790K
Motherboard
ASUS Maximux IV Extreme
RAM
Corsair Vengeance Pro
Hard Drive
128GB Samsung 830
Hard Drive
480GB SanDisk Extreme II
Hard Drive
2TB WD Red
Hard Drive
2TB Samung Spinpoint F4
Hard Drive
2TB Hitachi Deskstar 7K3000
Hard Drive
1TB Samsung Spinpoint F3
Hard Drive
3TB Toshiba DT01ACA300
Hard Drive
Seagate Barracuda 7200.14
Power Supply
Corsair 650TX
Case
Norco RPC 4224
Operating System
Windows 10 Pro
Other
LSI 9261-8i
Other
Intel RES2CV240
CPU
i5-4670k
Motherboard
Asrock Z97 Extreme6
RAM
32GB Kingston HyperX Beast 2400MHz
Hard Drive
128GB Samsung 850 Pro
Power Supply
850W Be Quiet! Dark Power Pro 10
Cooling
Thermaltake Water 3.0 Ultimate
Case
Thermaltake Urban T81
Operating System
Windows 8.1 Pro
▲ hide details ▲



Last edited by Sean Webster; 07-29-2019 at 11:01 PM.
Sean Webster is offline  
post #93 of 134 (permalink) Old 07-31-2019, 07:36 AM
New to Overclock.net
 
Join Date: Dec 2010
Posts: 3,795
Rep: 296 (Unique: 218)
Quote: Originally Posted by Sean Webster View Post
To answer your Q: Yes and no. People who state that it is like an HDD or that you are better off with an HDD are one of two people i'd say. Elitists who just want to talk trash because it's not the best. Or, people who don't understand how their workloads really work and what is actually needed to satisfy it and just reiterate what those elitists say. (which is far too common)
Hey Sean! Long time no see. Haven't been active here in years.

One thing that concerns me with QLC are peak latencies with the drive full. High latencies can really make a drive feel unresponsive. I tend to fill up my SSDs so I require consistent performance even when they're close to full. That said, I'm quite curious what AnandTech means by full here and whether an extra 10-20% OP might help reduce peak latencies and improve performance consistency.

















rui-no-onna is offline  
Sponsored Links
Advertisement
 
post #94 of 134 (permalink) Old 07-31-2019, 12:25 PM
New to Overclock.net
 
Join Date: Dec 2010
Posts: 3,795
Rep: 296 (Unique: 218)
Quote: Originally Posted by Sean Webster View Post
Write amplification (WA) varies as you stated, depending on the pSLC cache use and workloads (sequential vs random). Usually, WA for desktop usage usually isn't too high, especially with write caching enabled by default on Windows (helps make random writes more sequential). Thus, people will usually see maybe 1-3x actual writes, meaning 1GB of writes turns into 1-3GB of actual NAND writes. Sequential tasks =~1:1 while random can vary greatly 1:4 for example. Garbage collection and other background management tasks can also increase actual NAND writes too. Controllers that integrate real-time compression, like SandForce controllers of days past, can even hit negative WA values.
P.S.
Technically, SandForce drives can get WA below 1x if dealing with mostly compressible data but they can't actually hit negative WA values.
rui-no-onna is offline  
post #95 of 134 (permalink) Old 07-31-2019, 03:57 PM
New to Overclock.net
 
Omnidyne's Avatar
 
Join Date: Jan 2018
Location: Louisiana
Posts: 16
Rep: 3 (Unique: 2)
Quote: Originally Posted by Sean Webster View Post
Yeah, usually consumers will write ~5-10TB of data to the drive a year. So, you can typically just figure out what you need from that. Really, for most it isn't a factor to worry about too much. Sometimes its there it just to have an extra level of warranty limitation to reduce a company's RMA liability. Often times SSDs can exceed their ratings, it just takes a long time lol.

Write amplification (WA) varies as you stated, depending on the pSLC cache use and workloads (sequential vs random). Usually, WA for desktop usage usually isn't too high, especially with write caching enabled by default on Windows (helps make random writes more sequential). Thus, people will usually see maybe 1-3x actual writes, meaning 1GB of writes turns into 1-3GB of actual NAND writes. Sequential tasks =~1:1 while random can vary greatly 1:4 for example. Garbage collection and other background management tasks can also increase actual NAND writes too.
I've read that garbage collection and wear-leveling are the main causes of WA?

Quote: Originally Posted by Sean Webster View Post
Controllers that integrate real-time compression, like SandForce controllers
If I remember correctly, you stated you believe the 660p might be compressing the LBA map?

Quote: Originally Posted by rui-no-onna View Post
Hey Sean! Long time no see. Haven't been active here in years.

One thing that concerns me with QLC are peak latencies with the drive full. High latencies can really make a drive feel unresponsive. That said, I'm quite curious what AnandTech means by full here and whether an extra 10-20% OP might help reduce peak latencies and improve performance consistency.
You pulled results from the Anandtech "Destroyer" and "Heavy" benches; scenarios the majority of users would never subject the 660p to. The results you posted from the "Light" are more than acceptable for consumer workloads, as Anandtech states:

"On the Light test, the average read latency of the 660p stays comfortably below that of SATA drives even for the worst-case full drive test run, and only the average write latency shows a serious problem from filling up the whole drive and not giving it enough time to empty the now-reduced SLC cache."

"Otherwise, the 660p doesn't have any worrying QoS problems on this test and users won't notice serious pauses from the drive."

"The best-case latency scores from a freshly-erased 660p are acceptable for a high-end NVMe SSD and excellent for an entry-level drive. In the worst case of a full drive, the average latency is far higher but still low enough that the drive won't actually feel much slower.

They answered your question quite directly, I'd say.

Quote: Originally Posted by rui-no-onna View Post
I tend to fill up my SSDs so I require consistent performance even when they're close to full.
So, are you saying you're a workstation user that needs an ultra-low latency SSD? Because this is not made for that and isn't marketed as such. If you're not a workstation user the 660p is more than acceptable, as Anandtech makes very clear. Do you think they would recommend a drive that stutters under ultra-light workloads such as OS usage or gaming?


I almost wish Anandtech would take the storage benches out of their reviews because people seem to have issues understanding them. If you're not a power user the 660p is more than adequate in every way.

Intel Core i5-8400 | Gigabyte Z370P D3 | Corsair CX550M | NVIDIA GTX 970 Founders Edition | ADATA SX850 256GB - SandDisk MLC 120GB | WD 1TB | G.SKILL Ripjaws V Series 16GB (4 x 4GB) DDR4 2666 | Windows 10 | Corsair Carbide SPEC-03
Alien Space Heater
(12 items)
CPU
i5-8400
Motherboard
Gigabyte Z370P D3
GPU
GTX 970 Founders Edition
RAM
G.Skill Ripjaws V
Hard Drive
ADATA SX850
Hard Drive
Western Digital Blue
Power Supply
Corsair CX550M
Case
Corsair Spec 03
Operating System
Windows 10
Monitor
Acer SA230
Keyboard
Gigabyte Force K83 Cherry MX Red
Mouse
Logitech
▲ hide details ▲

Last edited by Omnidyne; 07-31-2019 at 04:08 PM.
Omnidyne is offline  
post #96 of 134 (permalink) Old 08-01-2019, 07:18 AM
New to Overclock.net
 
Join Date: Dec 2010
Posts: 3,795
Rep: 296 (Unique: 218)
Quote: Originally Posted by Omnidyne View Post
You pulled results from the Anandtech "Destroyer" and "Heavy" benches; scenarios the majority of users would never subject the 660p to. The results you posted from the "Light" are more than acceptable for consumer workloads, as Anandtech states:

"On the Light test, the average read latency of the 660p stays comfortably below that of SATA drives even for the worst-case full drive test run, and only the average write latency shows a serious problem from filling up the whole drive and not giving it enough time to empty the now-reduced SLC cache."

"Otherwise, the 660p doesn't have any worrying QoS problems on this test and users won't notice serious pauses from the drive."

"The best-case latency scores from a freshly-erased 660p are acceptable for a high-end NVMe SSD and excellent for an entry-level drive. In the worst case of a full drive, the average latency is far higher but still low enough that the drive won't actually feel much slower.

They answered your question quite directly, I'd say.
Note, I posted peak 99th percentile latencies for Destroyer, Heavy and Light. I don't really game nowadays but reckon that's like when the GPU gets decent average frame rates but you still feel it stutter at times because the minimum frame rates drop.


Quote:
So, are you saying you're a workstation user that needs an ultra-low latency SSD? Because this is not made for that and isn't marketed as such. If you're not a workstation user the 660p is more than acceptable, as Anandtech makes very clear. Do you think they would recommend a drive that stutters under ultra-light workloads such as OS usage or gaming?

I almost wish Anandtech would take the storage benches out of their reviews because people seem to have issues understanding them. If you're not a power user the 660p is more than adequate in every way.
Don't know if my use qualifies as workstation/power user but I do run a couple of light load VMs and deal with thousands of small files when I'm actually sitting in front of the PC. Multiple instances of Firefox and Chrome with 10-20 tabs each. Occasionally some torrents in the background when I'm trying new Linux distros. Often simultaneous read/write to same drive. I think my usage tends to be more I/O intensive rather than CPU or GPU intensive.

Granted, I certainly wouldn't be using HDDs for my usage (those are reserved for the NAS). I don't need to have the fastest SSDs or even ultra low latency but I do prefer better behaved SSDs with relatively consistent performance even when full. QLC is the future but at this time, it kinda seems like when we made the switch from MLC to planar TLC when you could find, say, the Samsung 830 at same price or cheaper than the Samsung 840.
rui-no-onna is offline  
post #97 of 134 (permalink) Old 08-01-2019, 07:29 AM
Overclocker in training
 
ThrashZone's Avatar
 
Join Date: Apr 2017
Posts: 7,526
Rep: 72 (Unique: 55)
Hi,
Got the samsung 970 evo plus 500gb installed last night only did one run just like the 970 evo 500gb with CDM

Read was a little lower than spec's hit 3250 spec's 3500
Write was surprising high it hit 3034 spec's is 2500

So I'll take that over the XPS any day done and dusted

I'm done with as ssd it's the most inconsistent piece of rubbish I've ever seen.

CPU
i9-9940x With Heatkiller IV Pro and VRM Plexi-Copper water block
Motherboard
ASUS x299 Rampage VI Apex
GPU
Titan Xp with copper Water Block
RAM
Trident Z 3600C16 4x8gb's b-die default timings 16-16-16-36
Hard Drive
Samsung 970 Evo Plus 500gb M.2/ Samsung 850 Pro 256gb for os and 500gb evo for favorite games plus other wd hdd's for data and backup's
Optical Drive
LG Blu-Ray WH16NS40
Power Supply
evga 1200-P2
Cooling
2-280GTX Black Ice Nemesis rads with D5 pump-res combo and D5 top before and after rads.
Case
corsair 450D with added 2nd floor to house radiator on top
Operating System
Win-7 pro & 10 pro Win-7 Primary os.
Monitor
ASUS VG248QE 24" 144Hz
Keyboard
Logitech G910 Orion spectrum
Mouse
Redragon Perdition
Audio
Built in realtek and Insignia 2.0 soundbar HSB318
CPU
i7-5930k with Heatkiller IV Pro Plexi-Copper and koolance vrm water blocks
Motherboard
X99 Sabertooth
GPU
EVGA 1080ti FTW3 with Water block
RAM
Trident-Z 3200C14 4x8gb's b-die default timings 14-14-14-34
Hard Drive
Samsung 970 Evo 500gb M.2/ Samsung 850 Pro 256gb for os and 500gb evo for favorite games plus other wd hdd's for data and backup's
Power Supply
EVGA 1000-P2
Cooling
2-240 GTX Black Ice Nemesis rads/ D5 pump-res combo and D5 top before and after rads.
Case
Corsair 450D with added 2nd floor to house radiator on top
Operating System
Win-7 and 10 pro Win-7 Primary os.
Monitor
AOC G2460PG 24"G-Sync 144Hz
Mouse
Red Dragon Perdition
Audio
Built in realtek and Insignia 2.0 soundbar HSB318
Hard Drive
eLeaf MELO-3 4ml
Power Supply
SE US18650VTC6 3120mAh 30A 3.6V 18650 Li-ion Battery - GREENx4
Case
ANGIS Legion MOD 200w Geek Vape
Operating System
VapeWild RazzleBerry 50-50-0mg nic
▲ hide details ▲
ThrashZone is offline  
post #98 of 134 (permalink) Old 08-01-2019, 08:08 AM
New to Overclock.net
 
Join Date: Dec 2010
Posts: 3,795
Rep: 296 (Unique: 218)
Quote: Originally Posted by Omnidyne View Post
I've read that garbage collection and wear-leveling are the main causes of WA?
It's not so much garbage collection itself but rather all writes and deletes that got the SSD into a "dirty" state.

SSDs write to a page but when they erase, they have to erase an entire block. They also can't overwrite pages directly so without wear leveling and LBA remapping, what would happen on overwrite is the SSD reads the entire block, erases the block and then writes the block (with the updated pages). Obviously, that's slow. It can also kill SSDs quickly as you'd have blocks with really high write amplification and all their P/E cycles used up (like old Indilinx SSDs, iirc).

With newer, smarter SSD controllers, what happens when you overwrite is the controller writes to another (clean/empty) location on the SSD and just notes the new address in its LBA map. It marks the old address as dirty so garbage collection can clean it up next time it does its thing. This is called dynamic wear leveling and it actually reduces write amplification.

You likely also have data that doesn't change on your SSD. If your SSD has, say, 60% static data and you're just writing/deleting to the same 40% with dynamic data/free space, that's bad. You'll have 60% of cells with just 1 or 2 P/E cycles used and a ton of wear on the other 40%. This is where static wear leveling comes in. The SSD controller moves the static data occasionally so you're not just hammering the same blocks over and over. Obviously, static wear leveling increases WA because it causes some extra writes but it does so to keep P/E cycle usage even. Otherwise, you can get into a situation where, say, 60% of the SSD is at 100% health while the other 40% is already dead.

TL;DR
Normal writes (primarily overwrites) cause write amplification.
Random 4K writes cause higher WA than large sequential writes.
Wear leveling (static) also increases WA but end result is good for overall SSD health.

Last edited by rui-no-onna; 08-01-2019 at 08:11 AM.
rui-no-onna is offline  
post #99 of 134 (permalink) Old 08-03-2019, 09:05 AM - Thread Starter
Overclocker
 
JackCY's Avatar
 
Join Date: Jun 2014
Posts: 10,013
Rep: 339 (Unique: 240)
Quote: Originally Posted by Omnidyne View Post
QLC and 660p
Latencies and write speed on large transfers (movies, VMs, ...) it's in reviews but not on Tom's there is not that full drive write graph as other reviews there.



And I think it only gets worse as you fill the drive.

Meanwhile you can pay <10% or so more for a faster TLC drive that doesn't tank below HDD write speeds.
The QLC drives cost too much for what they offer. QLC would make sense if they increased offered capacity to 4TB at least and drop prices to half. Till then they can keep the QLC.
JackCY is offline  
post #100 of 134 (permalink) Old 08-03-2019, 09:53 AM
New to Overclock.net
 
Join Date: Dec 2010
Posts: 3,795
Rep: 296 (Unique: 218)
Quote: Originally Posted by JackCY View Post
Meanwhile you can pay <10% or so more for a faster TLC drive that doesn't tank below HDD write speeds.
The QLC drives cost too much for what they offer. QLC would make sense if they increased offered capacity to 4TB at least and drop prices to half. Till then they can keep the QLC.
This is pretty much where it's going eventually. Situation was the same with the transition from MLC to TLC. Although I expect we'll still have TLC for performance and QLC will be bulk-ish storage or large all-in-one (OS+data) drive for laptops, NUCs, pre-built desktops, etc.
rui-no-onna is offline  
Reply

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off