TR Forums

Optane endurance is rated higher than NAND drives, which I thought correlates with higher retention, yet I saw an Intel employee saying this:

Optane endurance is significantly higher than NAND SSDs with a capability of up to 60 drive-writes-per-day, as compared to 3 drive-writes-per-day with Intel's highest production NAND SSD today. NAND or HDDs may be better aligned for storing static, cold data.

Anyone read anything concrete about 3DXP retention?

Endurance and retention are not necessarily correlated. That just happens to be the case with flash, since retention degrades as the cells wear out. Think of DRAM as being the extreme at the other end of the solid-state memory tech spectrum -- nearly infinite endurance, but retention measured in just milliseconds before it needs to be refreshed.

I do not know the details for current 3DXP devices, but it is entirely plausible that they could degrade less than NAND does in use, yet inherently have worse long-term storage characteristics. Or maybe they just over-provision the 3DXP devices to a larger degree to give them better endurance specs. Or maybe the Intel person was merely referring to cost -- for bulk archival storage of "cold" data, you generally want the cheapest tech that gives you the required retention characteristics.

It might be a cost thing - from what I've seen 3DXP and NAND both have pretty hideous retention characteristics. Refreshing data every 3-6 months is basically required for both.

just brew it! wrote:

Endurance and retention are not necessarily correlated. That just happens to be the case with flash

Good point.

Waco wrote:

from what I've seen 3DXP and NAND both have pretty hideous retention characteristics.

Have you found any concrete retention specs?

It's only 3 months at 40c.

Finally this response from Intel on the endurance questions below:

"The Intel® Optane™ SSD DC P4800X follows the JESD218B-01 standard for data retention: 3 months of powered off retention at 40C. Powered on, data retention is throughout the product warranty period."

Reference: the comments at this EETimes article.

https://www.eetimes.com/document.asp?doc_id=1331494

It's pretty comparable to SSDs stored at 40c write/stored: 14 weeks.

https://www.anandtech.com/show/9248/the ... -retention

The industry still has a long way to go before we can replace hard drives/tape/optical for offline storage.

I thought the attraction of 3DXP was speed of random reads and writes -- it's definitely not aimed at archive purposes afaik.

I don't know about the retention characteristics of 3DXP, but economically/financially, it doesn't even matter. You'd never want to use 3DXP for cold storage.

Retention is for when the drive is not connected to any power source. But why would you want to store a fast, expensive medium like 3DXP in a closet or warehouse? That's what tapes and mechanical HDDs are for. If you own 3DXP, you own it for speed, and speed means it's powered on.

Even if you had an old 3DXP that you replaced with a newer one, it still wouldn't make sense for archive. 3DXP is already small (non-dense), so once it's old, it's going to be even smaller, compared to the density of the newest mechanical HDDs. Filling your closet with 480 GB Intel Optane 900P instead of 12 TB mechanical HDDs makes no sense - both from a money and a physical space perspective.

defaultluser wrote:

The Intel® Optane™ SSD DC P4800X follows the JESD218B-01 standard: for data retention: 3 months of powered off retention at 40C

Thanks. I think I recall this was for exhausted (flash) cells?
Too bad the standard is behind a paywall (they gotta be kidding).

If it's not just a "safe answer" from Intel PR, that's very disappointing. And also easy to test.

K-L-Waster wrote:

I thought the attraction of 3DXP was speed of random reads and writes -- it's definitely not aimed at archive purposes afaik.

Maybe. But there was always hope to one day have a true successor to HDDs, or even *gulp* optical media.

mikewinddale wrote:

economically/financially, it doesn't even matter. You'd never want to use 3DXP for cold storage.

It might matter in the future. Also flash was expensive when it started.

defaultluser wrote:

It's only 3 months at 40c.

Yikes. I wonder if the people buying external SSDs are aware of this? Almost seems like they should be including a battery in the enclosure to sustain a charge, much like the CR2032 for the system BIOS. Don't know what quantity of charge is needed, but I know they have smoke alarm batteries made to last 10 years.

The Egg wrote:

Yikes. I wonder if the people buying external SSDs are aware of this?

AFAIK that's for cells that exhausted their rated write cycles, and specifically for Enterprise-class.
I think the typical quoted figure for Client-class is 1 year at 30C.

There's also this (from a 2010 presentation, not sure what the standard says):

The Egg wrote:

defaultluser wrote:

It's only 3 months at 40c.

Yikes. I wonder if the people buying external SSDs are aware of this? Almost seems like they should be including a battery in the enclosure to sustain a charge, much like the CR2032 for the system BIOS. Don't know what quantity of charge is needed, but I know they have smoke alarm batteries made to last 10 years.

A CR2032 would suffice to sound an alarm when time runs out.
Two AA alkaline cells should be enough to refresh a SSD many times over several years, though (that's my very quick calculation). A CR2032 would suffice to sound an alarm when AA's run out. Not really practical for small external SSDs.

Wirko wrote:

Two AA alkaline cells should be enough to refresh a SSD many times over several years, though (that's my very quick calculation). A CR2032 would suffice to sound an alarm when AA's run out. Not really practical for small external SSDs.

I don't know that that's true - a pair of batteries are unlikely to be able to power the controller and associated flash for very long at all (a regular AA battery has around 3 watt hours of power storage). A typical drive will pull a handful of watts when actively writing, so even a 1 TB drive with a 500 MB/s overwrite speed (which is probably unlikely) would only be able to be updated 3-4 times by a pair of batteries.

I'm still frustrated that there's no public release of retention times, just JEDEC standards to guess from.

Refreshing could require rewriting many or even all of the cells. I don't think any normal battery would suffice.