Overclock.net › Forums › Components › Hard Drives & Storage › SSD › SSD write endurance
New Posts  All Forums:Forum Nav:

SSD write endurance - Page 4

post #31 of 64
Note that P/E cycle specification are minimum numbers, not average.

In addition, there are tests which do write random data to SSDs and all have demonstrated life spans well above P/E cycles

In addition, ECC, wear-leveling, parity, and adaptive signalling are techniques already in use to help extend the life of NAND.
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
post #32 of 64
Thread Starter 
Quote:
Originally Posted by rui-no-onna View Post

You keep quoting 3x write amplification but that's just a conservative estimate by most websites. I reckon real world write amplification is lower. Alas, I don't have write amplification numbers for my TLC SSD's (840 500GB and 840 EVO 1TB) since I haven't written enough data to them yet.
I think the point he was making is he's a power user relatively speaking and he's only written 17TB to the drive in 2.7 years. For your average user, having only 128TB NAND writes on a 128GB just isn't a cause for concern. Besides, 128GB is fairly small and I doubt folks would be capturing raw video at bitrates of 1.5Gbps to a measly 128GB drive.

The thing is I doubt folks who tend to use their computers for 10 years or more (until the computer dies) are the type to write tons of stuff to their drives. The ones likely to do that are people familiar with technology and they're more likely to use high capacity HDD's or SSD's as well as replace hardware more often. Really, the only thing I'd do if I'm setting up a system with SSD for a non-techie is to overprovision the SSD.

I think 3X is kind of realistic.

If I do a straight file copy on my SSD, it causes about a 1.5-1.8X write amplification. The non-stop tiny writes caused by vmware are causing a 5 - 10X write amplification though, so averaged out I could see 3X being my actual WA.
Quote:
Originally Posted by DuckieHo View Post

Note that P/E cycle specification are minimum numbers, not average.

In addition, there are tests which do write random data to SSDs and all have demonstrated life spans well above P/E cycles

In addition, ECC, wear-leveling, parity, and adaptive signalling are techniques already in use to help extend the life of NAND.

Wear leveling allows you to use the full P/E cycles of the entire drive. Without it drive lifetimes would be under a year since the moment 1 nand cell wears out the drive's useless. The other things are great as well, but once a drive starts getting errors that are being corrected by ECC, I wouldn't want to use it anymore.

And I understand those are minimum values, not average, but I wouldn't want to use a boot drive past it's minimum guaranteed P/E cycles. Just like I wouldn't use my platter disks that work fine, but have SMART failures as a boot drive (though I do use them for torrenting. Wouldn't bother me one bit if I lost that data).
post #33 of 64
Quote:
Originally Posted by demoship View Post

I think 3X is kind of realistic.

If I do a straight file copy on my SSD, it causes about a 1.5-1.8X write amplification. The non-stop tiny writes caused by vmware are causing a 5 - 10X write amplification though, so averaged out I could see 3X being my actual WA.

Curious how you got that 1.5-1.8x number for what I would presume to be sequential writes. I calculated the write amplification on my 2-yr old Samsung 830 256GB OS drive (containing page file, appdata, temp folders, etc) and it comes out to just 1.43x.
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
post #34 of 64
Quote:
Originally Posted by demoship View Post

Wear leveling allows you to use the full P/E cycles of the entire drive. Without it drive lifetimes would be under a year since the moment 1 nand cell wears out the drive's useless. The other things are great as well, but once a drive starts getting errors that are being corrected by ECC, I wouldn't want to use it anymore.

And I understand those are minimum values, not average, but I wouldn't want to use a boot drive past it's minimum guaranteed P/E cycles. Just like I wouldn't use my platter disks that work fine, but have SMART failures as a boot drive (though I do use them for torrenting. Wouldn't bother me one bit if I lost that data).

NAND are still readable for at least one year after no longer being writable.

If you look at every single real-world test, the NAND vastly outperfom their minimum specifications.

The case you make is like saying we should throw out HDDs after 4 years since their failure rate on average is higher.
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
Once again...
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 920 [4.28GHz, HT] Asus P6T + Broadcom NetXtreme II VisionTek HD5850 [900/1200] + Galaxy GT240 2x4GB G.Skill Ripjaw X [1632 MHz] 
Hard DriveOSMonitorKeyboard
Intel X25-M 160GB + 3xRAID0 500GB 7200.12 Window 7 Pro 64 Acer H243H + Samsung 226BW XARMOR-U9BL  
PowerCaseMouseMouse Pad
Antec Truepower New 750W Li Lian PC-V2100 [10x120mm fans] Logitech G9 X-Trac Pro 
  hide details  
Reply
post #35 of 64
Thread Starter 
Quote:
Originally Posted by rui-no-onna View Post

Curious how you got that 1.5-1.8x number for what I would presume to be sequential writes. I calculated the write amplification on my 2-yr old Samsung 830 256GB OS drive (containing page file, appdata, temp folders, etc) and it comes out to just 1.43x.

Well, if I copy a bunch of big files, the Lifetime Writes # goes up by roughly 1.5X the size of what I copied, a bunch of small files and the value's a bit higher.
post #36 of 64
Quote:
Originally Posted by DuckieHo View Post

NAND are still readable for at least one year after no longer being writable.

If you look at every single real-world test, the NAND vastly outperfom their minimum specifications.

Theoretically readable but I've yet to see an SSD fail gracefully based on JEDEC recommendations. That said, like you said, I haven't seen any NAND-related SSD failure before it reaches the P/E cycle specification either based on torture tests done by various websites. At the very least, we know a Samsung 840 EVO 120GB is good for 120TB NAND writes (assuming it doesn't experience controller failure or something) and that's more than what your average user will likely write in their computer's serviceable life.

Quote:
Originally Posted by DuckieHo View Post

The case you make is like saying we should throw out HDDs after 4 years since their failure rate on average is higher.

Exactly. Going by this, I'd be replacing my HDD's every 3 years just because.
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
post #37 of 64
Quote:
Originally Posted by demoship View Post

Well, if I copy a bunch of big files, the Lifetime Writes # goes up by roughly 1.5X the size of what I copied, a bunch of small files and the value's a bit higher.

The values to pay attention to are Host Writes and Wear Leveling Count. Those are the values that tell you what your write amplification is. On my Samsung 840, that's F1 (Total LBA Written) and B1 respectively. One thing I like about my Intel 330 and Sandisk Extreme, there are S.M.A.R.T. values for NAND writes and Host writes so it's easy to calculate for write amplification.
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
post #38 of 64
Thread Starter 
Quote:
Originally Posted by rui-no-onna View Post

The values to pay attention to are Host Writes and Wear Leveling Count. Those are the values that tell you what your write amplification is. On my Samsung 840, that's F1 (Total LBA Written) and B1 respectively. One thing I like about my Intel 330 and Sandisk Extreme, there are S.M.A.R.T. values for NAND writes and Host writes so it's easy to calculate for write amplification.

I have a samsung 840 pro, my SMART attributes don't include either of those.
post #39 of 64
Quote:
Originally Posted by demoship View Post

Warning: Spoiler! (Click to show)
I have a samsung 840 pro, my SMART attributes don't include either of those.

It does (just named differently with HDTune vs CrystalDiskInfo). (B1) Wear Range Delta is your wear leveling count (raw data currently at 1) so you've used 1 P/E cycle.

I'm not sure if the Data column in HDTune is raw data or the calculated value in bytes.

If (F1) LifeTime Writes is raw data and based on host writes in terms of LBA, you'd have to multiply that value by 512 to get the value in bytes.

Your total writes so far:
346,233,066 * 512 bytes = 177,271,329,792 bytes = 165 GiB

I recommend installing CrystalDiskInfo for displaying your S.M.A.R.T. information for easier viewing.
Edited by rui-no-onna - 4/8/14 at 1:06pm
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
Garnet
(11 items)
 
Lucifiel
(13 items)
 
Metatron
(13 items)
 
CPUMotherboardRAMHard Drive
Intel Core i5-3450S Intel DQ77KB Corsair 16GB DDR3 1600 SO-DIMM Samsung 830 256GB 
Hard DriveHard DriveCoolingOS
Crucial m4 256 mSATA Samsung 840 500GB Intel BXHTS1155LP Windows 7 Ultimate x64 
KeyboardCaseMouse
Logitech K800 Lian Li PC-Q05B Logitech M570 
CPUMotherboardGraphicsRAM
Core i7-860 Gigabyte GA-H55N-USB3 EVGA GTX 460 1GB GDDR5 Kingston 2x4GB DDR3 1333 
Hard DriveOSPowerCase
G.SKILL Phoenix Pro 120GB Windows 7 Ultimate x64 Silverstone ST45SF 450W Silverstone Sugo SG05B 
CPUMotherboardGraphicsRAM
Core i7-860 Biostar TH55B HD MSI GT 240 1GB GDDR5 Kingston 2x2GB DDR3 1333 
Hard DriveOSMonitorPower
Intel X25-M 120GB Windows 7 Ultimate x86 Samsung 22" Antec EA-380D 
Case
Rosewill R101-P-BK 
  hide details  
Reply
post #40 of 64
Thread Starter 
Quote:
Originally Posted by rui-no-onna View Post

It does (just named differently with HDTune vs CrystalDiskInfo). (B1) Wear Range Delta is your wear leveling count (raw data currently at 1) so you've used 1 P/E cycle.

I'm not sure if the Data column in HDTune is raw data or the calculated value in bytes.

If (F1) LifeTime Writes is raw data and based on host writes in terms of LBA, you'd have to multiply that value by 512 to get the value in bytes.

Your total writes so far:
346,233,066 * 512 bytes = 177,271,329,792 bytes = 165 GiB

I recommend installing CrystalDiskInfo for displaying your S.M.A.R.T. information for easier viewing.

Wear range delta isn't leveling count. It's the delta between the most worn block and least worn block (hence the delta). It's basically an indication of how well the wear leveling is working.

I'm not sure what lifetime writes represents. Now that you mention it, I think I read it wrong. I was reading that as if the first 3 digits were GB (I was off by 3 zeros, lol)
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: SSD
Overclock.net › Forums › Components › Hard Drives & Storage › SSD › SSD write endurance