Intel Unleashes Powerful SSDs For Big Data Era

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
The Intel Solid-State Drive Data Center Family for PCIe, announced today by Renee James in her Computex keynote, features substantial performance gains over SATA SSDs and traditional hard disk drives. The Intel SSD DC P3700, P3600 and P3500 Series offer extraordinary throughput, lower latency and reduced total cost of ownership. Utilizing NVMe technology, the new SSDs were designed to meet the needs of the modern data center and support the growing demands for storage as a result of the explosion in data. The SSD Data Center Family for PCIe delivers faster access to data for businesses and end users and ensures a consistently amazing and reliable experience across a range of applications.
 
Anything addressed on increasing the write lifespan of such devices?
 
Anything addressed on increasing the write lifespan of such devices?

Can this die soon, please? Every damn thread regarding SSD's has someone who hasn't been on a tech site in 4 years spew some variant of this comment. It's getting old.
 
Can this die soon, please? Every damn thread regarding SSD's has someone who hasn't been on a tech site in 4 years spew some variant of this comment. It's getting old.

I think its a good metric to keep on hand, I don't like TLC drives simply because by design they won't last as long as mlc or slc. That and TLC drives don't seem to be offering that much of a discount.
 
I don't like TLC drives simply because by design they won't last as long as mlc or slc.

Will you care that much if your drive only lasts 12 years instead of 15? By then I expect we will not be using flash for SSDs.

That and TLC drives don't seem to be offering that much of a discount.

That I agree. The EVOs are actually more expensive than a lot of MLC drives. Although doesn't the EVO have some SLC on each nand and they come with the write behind caching software that other drives do not have.
 
Will you care that much if your drive only lasts 12 years instead of 15? By then I expect we will not be using flash for SSDs.

I care more for the use of the drive. If I'm in need of a storage drive, then TLC isn't a big issue, if I need a drive to be a cache drive for a large pool then I want that drive to be designed to be written to as much as possible.
 
Will you care that much if your drive only lasts 12 years instead of 15? By then I expect we will not be using flash for SSDs.
It seems I recently read that the endurance of TLC isn't quite as bad as expected, but I've also read that data decay is a fair bit faster. I'm not sure if/when that can be overcome.

That I agree. The EVOs are actually more expensive than a lot of MLC drives. Although doesn't the EVO have some SLC on each nand and they come with the write behind caching software that other drives do not have.
I'm sure that TLC would go mainstream if prices better reflected the difference in endurance, etc. between the NAND techs.
-

TLC probably wouldn't be so bad for large storage drives that don't get written to constantly. I'm thinking along the lines of relatively low cost 8TB+ (16TB drives are already planned for 2016) media server drives.
 
These drives cost between $1.50 to $3 per GB, which may seem incredibly high to many of you, until you realize that these drives have enough performance to replace SATA SSDs in RAID. For example, a single one of the $3 per GB drives (the P3700) can replace 4 to 6 of the best SATA SSDs you can buy in RAID.
 
According to the review it's bootable... I think I just found my next system drive!
 
I read a lot of people talking about garbage collection over the years, yet nobody could explain how it could work if it really existed...

Will you care that much if your drive only lasts 12 years instead of 15? By then I expect we will not be using flash for SSDs.

That I agree. The EVOs are actually more expensive than a lot of MLC drives. Although doesn't the EVO have some SLC on each nand and they come with the write behind caching software that other drives do not have.

No, they use a tech that writes on TLC as if it was SLC, so only one bit per cell instead of three, improving performance but using three times the space. Then copy that data elsewhere to recover the space. This tech should actually shorten the lifespan of the drive by causing write amplification.
 
Write endurance is not a bad question to ask regarding MLC drives and smaller fab processes, although it really has become a diminishing concern with most enterprise focused drives, especially from intel.

Even so, for database and certain write-heavy applications, it's something to keep in mind when switching from an older SLC based SSD - at the very least you want to know what the expected lifespan is and compare it to the amount of data you expect to write over the expected drive lifespan.

For the vast majority of write applications, these drives will be more than sufficient as far as endurance goes, and for personal use or web apps / write-light enterprise workloads, the P3500 and P3600 are a great value.

The big story for me is progress in moving away from the SATA/SAS bottlenecks and on bootable and well-supported PCIe interfaces like NVMe, and the optimization of the driver stack as we start approaching 1 million IOPS in a single compact chassis.
 
As far as I'm concerned write endurance is a complete non-issue.... even if a production environment.

On a high-use DB server I would anticipate replacing my drives at least once every five years whether I felt they needed it or not.

Drive performance going to these SSDs is too damned high when compared to 15Krpm arrays or SATA SSDs to worry about it.
If you have a server other than a large-capacity file server running on spinning drives you're doing it wrong.
 

Great argument there !

For the vast majority of write applications, these drives will be more than sufficient as far as endurance goes, and for personal use or web apps / write-light enterprise workloads, the P3500 and P3600 are a great value.

I wouldn't call paying several times more for no actual gain great value ! If you don't need the performance, don't buy it.

As far as I'm concerned write endurance is a complete non-issue.... even if a production environment.

On a high-use DB server I would anticipate replacing my drives at least once every five years whether I felt they needed it or not.

Drive performance going to these SSDs is too damned high when compared to 15Krpm arrays or SATA SSDs to worry about it.
If you have a server other than a large-capacity file server running on spinning drives you're doing it wrong.

I don't understand your argument at all, what if your workload means the drive is dead in 2 years ?
 
Can this die soon, please? Every damn thread regarding SSD's has someone who hasn't been on a tech site in 4 years spew some variant of this comment. It's getting old.

You mad? Plenty of people have had dead SSDs in a short time.
 
I don't understand your argument at all, what if your workload means the drive is dead in 2 years ?


Then I would still happily replace my drives every single year.
SSD's are that cheap.
The performance they grant is that significant.

Let's assume I was writing a petabyte per year, per SSD... and I was worried about endurance.
Replacing the SSD every single year should not be a deterrent to its use.
If anything requirements of that sort of write volume with half the performance afforded from SSDs should allocate annual replacement into its budget.

A 15k Cheetah can give up to 125MBps sustained write.
This intel SSD can give nearly 2GBps sustained write.
Assuming the cheetahs can scale linearly (they can't) then you'd need 16 of them.

Sixteen 15k drives are far more expensive than this SSD.
Sixteen drives with fast spinning parts have a worse MTBF than the Intel SSD does.


FWIW, TechReport has been running a write endurance test for a while.
http://techreport.com/review/26058/the-ssd-endurance-experiment-data-retention-after-600tb

Here is 600 TB written without a major performance drop due to endurance failures.

The bottom line:
If performance matters, you should be using SSDs

There is absolutely no reason to worry about write endurance failures. Zero.

This intel drive advertised 36 Petabytes of write endurance.... I'd love to see any spinner on the planet obtain that level of abuse.
 
Last edited:
You mad? Plenty of people have had dead SSDs in a short time.

How many of these deaths were caused by wearing out the NAND? I'd bet very very few. And its not like hard drives don't have infant mortality.
 
My main concern with SSDs is "Sudden unexpected death syndrome", not write endurance. But write endurance is certainly a factor to look at when deciding what to buy, at least for certain purposes.
 
HDDs have that too. More often than SSDs do.

This is why you do backups.

I agree that hard drives have much higher annual failure rates then most SSDs however there is at least some chance of predicting failure and some ability to recover at least some of the data. I have predicted failure in almost every case of the 75+ drives I have RMAd at work over the last few years. With SSDs there is no chance of predicting failure and the end user can not recover. Although I suspect in a lot of cases the manufacturer could get your data back provided the mapping table is not destroyed however to my knowledge they are not doing this yet.
 
I would like to see a source on hdd vs ssd failure rates. http://www.ibas.no/ have stated that they see a larger amount of failed SSDs compared to HDDs relative to the amount of units sold. This can of course be related to my point about "sudden death syndrome". If SSDs tend to fail without warning more often than HDDs then that would lead to more people needing data recovery on SSDs. Also, based on personal experience HDDs tend to usually give some notice before they fail completely while SSDs just die. I don't have enough SSDs to say anything statistically significant about that though...
 
Last edited:
I don't have enough SSDs to say anything statistically significant about that though...

I suspect I will never ever be in control of a sample size that would be statistically significant although I wish I had thousands of drives to play with instead of hundreds of hard drives and dozens of ssds.


have stated that they see a larger amount of failed SSDs compared to HDDs

I was going by the 2% to 10% annual failure rates in HDs that I see discussed from time to time versus the manufacturer quoted failure rates from Intel Samsung and Crucual (usually less than 1%). Although manufacturers do lie. And yes I am deliberately ignoring OCZ.
 
Last edited:
You mad? Plenty of people have had dead SSDs in a short time.

Thank you for your anecdotal evidence. I shall archive it right next to "Conspiracy Theories" in my "crazy stuff found on the internet" folder.
 
I think I'll be buying a P3500. I see the P3600s and P3700s for sale on a few sites like ShopBLT but no P3500 yet...
 
Well data is written two times, that's called write amplification in my book.

Then your book is in error -- oversimplified to the point of being misleading and useless.
 
Last edited:
How many of these deaths were caused by wearing out the NAND? I'd bet very very few. And its not like hard drives don't have infant mortality.

Indeed...it's mainly crappy firmware or incompatibilities between a controller and a drive that seem to cause most of the issues. In any event, most SSDs have life indicators on them that'll let you know when you're approaching the end of a drive's write lifespan.

I think I'll be buying a P3500. I see the P3600s and P3700s for sale on a few sites like ShopBLT but no P3500 yet...

Some sites my have SKUs up for the P3600 and P3700, but have no stock on them. I'm watching (and waiting) too...;)
 
According to the review it's bootable... I think I just found my next system drive!
But I'm also wondering if it's using NVMe protocol and not AHCI.

I'll probably wait until Windows 9 comes out and build my next system on x99 with one of these as boot drive.
 
Then your book is in error -- oversimplified to the point of being misleading and useless.

And your answers are not simplified ? Can you explain how the tech doesn't cause write amplification ?
 
Alrighty then guys!

Who is going to do it the [H]ardocp way and get one for these beauties and test it for the rest of us! :D

Man, that DC P3600 800 GB PCI-E looks like the killer kit to get. Those 400GB are not that fast at writing, the P3700 are too expensive, and the P3500 is neither out nor available in 800GB.

Sooo! Who is getting that DC P3600 800 GB PCI-E? :p :D
 
I am waiting on the DC P3500 to be released. It is suppose to be 3 rd quarter this year. But don't know where to find any info on the release date.
 
Back
Top