Stereodude
Not really a
That's not necessarily true. It depends on the flash cell design.
SSDs - State of the Product?
- Thread starter ddrueding
- Start date
ddrueding
Fixture
Connectivity is going to be a major concern as well. As Lunar alluded to, even now newer storage interfaces are coming out. Also, the unknown risk of putting something offline on a shelf is, to me, unacceptable.
I'll start another thread about this part of it...
I'll start another thread about this part of it...
sechs
Storage? I am Storage!
No company of any kind of size ever did back-ups to optical disks or USB thumb drives.
There doesn't seem to be a lot of penetration of SSDs into big storage. While I'm sure that the spinning disk is well on its way out on desktops, when you're deploying petabytes, a small price delta gets magnified pretty quickly.
jtr1962
Storage? I am Storage!
This sums things up pretty well: http://www.enterprisestorageforum.c...sd-vs.-hdd-performance-and-reliability-1.html
The article is two years old however. SSD prices have dropped since then relative to HDD prices, so so of their conclusions may have changed slightly. I think we'll experience a gradual phase out of spinning disks in enterprise storage which will happen after HDDs are mostly gone from personal computers, but I'd be surprised if spinning disks were used at all in, say, a decade. The roadmap for increasing HDD capacity seems to peter out around then.
Didn't you say the same thing about 6 years ago? 
A 2TB 850 Pro SSD costs the same as last year. The 4TB 850 TLC (which has far less durability than a HDD) is $1500. That's not exactly inspiring for price parity.
10 years is a long time to wait. Windows 12 will probably be the OS then and all our data will be controlled by it.
A 2TB 850 Pro SSD costs the same as last year. The 4TB 850 TLC (which has far less durability than a HDD) is $1500. That's not exactly inspiring for price parity.
10 years is a long time to wait. Windows 12 will probably be the OS then and all our data will be controlled by it.
jtr1962
Storage? I am Storage!
If I did that means we have four years to go. It's a pretty fair bet HDDs will mostly be dead in personal computing market by then.Didn't you say the same thing about 6 years ago?
How many spindles can that 2TB SSD replace? In many enterprise applications using HDDs they actually use more drives even if capacity isn't needed to spread the data across more spindles for faster access. It's actually possible we've already hit price parity in some enterprise applications when you consider that. BTW, the regular 850 has practically the same specs as the 850 Pro for about 2/3rds the price. What exactly is that extra ~$300 getting you? Maybe there's some esoteric spec that matters in enterprise applications but frankly I'm not seeing it. Also, if you want to do a fair comparison, then compare the 2TB 850 Pro at ~$950 to 2TB enterprise HDDs like this. Not such a huge price difference any more, and really getting the much, much faster 850 Pro for ~1.6x the price seems like a no brainer.
The $1500 for the 4TB 850 doesn't seem all that horrible either when you look at enterprise drives with the same capacity. Every time a higher capacity SSD comes out, it puts downward price pressure on smaller sizes, especially for consumer drives.
Another thing I've heard regarding HDDs versus SSDs is at some point HDDs will just be unable to deliver the performance needed to work with the ever faster requirements in the enterprise arena. At that point it won't even matter if you have price parity or not. Nobody will want HDDs even if you gave them away for free.
Hopefully Microsoft's model of wanting to control your machine and your data will fall flat on its face with Windows 10 and you'll get maintream adoption of Linux. If not, Windows 12 might be called the "Empire Edition".
Last edited:
Of course the 10K and 15K SAS performance drives are going away first, but the Amazon or the Google, etc. need to store exabytes of data in bulk at low prices. Whether it is in the enterprise capacity or cheaper drives like the BarkBlazers, that's all 3.5" drives for economy. I'd be buying piles of SSDs, but 10x the cost of HDDs is far too much for bulk storage.
There needs to be some fundamental improvements in the flash memory and the vertical stacking is not a long-term solution.
There needs to be some fundamental improvements in the flash memory and the vertical stacking is not a long-term solution.
jtr1962
Storage? I am Storage!
I've heard there's no fundamental physical limit on vertical stacking. There might ultimately be a process limit. In any case, the next few years in storage should be interesting. Obviously cheap, bulk storage will be the last thing to go. HDDs will probably hit some fundamental limit on future capacity increases which in turn will limit how low the price per TB can go. SSDs will hit their own limits eventually also but my understanding is we stillhave quite a way to go.
sechs
Storage? I am Storage!
Two-thirds if we're talking about 2.5" drives. Only a fifth on the 3.5" front.
Samsung announced a 16GB SSD, but I can't imagine how ungodly the price is. It'll be a good while before the cost premium for solid state is worth it.
My point is that with 48 layers, you'd think the price would be a lot lower. They need to do more than that to reduce costs. Is NAND the only answer for SSD? I thought there would be something better eventually.
jtr1962
Storage? I am Storage!
I think we're still on a major learning curve. The raw materials for any kind of semiconductor are ridiculously inexpensive. The real cost is in the process. I have to think as we get better at it, yields will go up and the cost will come way down. The yields for 3D NAND are probably a closely guarded secret, but it wouldn't surprise me if they're under 10% given the process complexity described here. That implies we could reduce cost by up to a factor of ten just by increasing yield.
We've all seen articles about the next big thing which will replace NAND. So far, none of these have panned out. I'd love to see something better myself but so far no luck. Xpoint looks promising on the speed, durability, and capacity front but it's looking more like a premium solution for enterprise applications than mainstream bulk storage. MRAM was the next big thing a few years back but it hasn't scaled up so well. I'm sure eventually something will come along, hopefully before I'm on Social Security.
It's rather past Q2 and Mushkin doesn't even have a 2TB reactor drive, much less 4TB. More Vorpwares.
jtr1962
Storage? I am Storage!
It would be great to hit that price point for a 4TB SSD but only if they don't sacrifice reliability in the process.
Prices won't go down very quickly when there is no competition. At the very least we should see 2TB drives from the other mainstream manufacturers. There's no way I would buy drives from a "Munchkin" company.
CougTek
Hairy Aussie
This things needs active cooling. Lots of it :
.
That's 60W of dissipation, just for a 3.5in drive.
Last edited:
Seagate's bullet point #3 says, "New level of power efficiency: 4TB/watt"
So 60 watts when in use and 15 watts when idle?
From the Ars article, "We do have a figure on power consumption: 1W per terabyte. Other enterprise SSDs tend to sit at around 2W-per-TB at idle."
What is the "in-use" vs "idle" time of an SSD in a large data center? Is the drive going to be idle 99.37% of the time, so one 60 TB Seagate will use less power than 60 one TB SSDs even if the Seagate uses more power when active because of it's size?
Does DD's CC have a $30k limit so we can find out? /snark
ddrueding
Fixture
It does, but I just put a car on it. It would fit both, but then I wouldn't be able to pay it off this month.
Even 20W continuous output would get darn hot without significant active airflow. Particularly if you have many of them in a chassis.
It would fit both, but then I wouldn't be able to pay it off this month.Even 20W continuous output would get darn hot without significant active airflow. Particularly if you have many of them in a chassis.
jtr1962
Storage? I am Storage!
Those numbers don't make sense to me. For example, here's Anandtech's review of the 4TB EVO 850. Active idle power consumption is 290mW. That's really the number which matters. Scale that to 60GB and you arrive at a reasonable 4.35 watts. That's probably the real-world continuous power consumption, especially given this drive's intended use as faster than HDD bulk storage. Probably it wouldn't be accessed often enough to significantly exceed the active idle power consumption. I would also imagine the 60 watts is a purely theoretical figure if you had every flash chip in the drive reading or writing simultaneously. If the drive regularly exceeded more than a few watts, it undoubtedly would have a hefty heat sink, perhaps even a fan. The fact it doesn't tells me there's no need for it.
The 1.6TB Samsung and 800GB Intel SSDs in Half-Height, Half-Length (HHHL) form factor that we have in our rack server have heatsinks on them already top help dissipate the extra heat. The Samsung and Intel specification states 25W max and 4W idle for the drives. In the case of the 60TB drive I suspect it's not only the NAND that may output more heat but also needing an advanced controller to manage the much larger size of the device that may require more power to process it all. That and likely larger amounts of RAM for caching indexes and possibly even super caps to provide a short amount of power during power outages to flush the writes.
jtr1962
Storage? I am Storage!
Interesting. I'd personally like to see more detailed specs from Seagate. I know the HHHL drives have much higher IOPS and STR than consumer SSDs. That probably accounts for the much larger power consumption. The Seagate's case design is just totally incompatible with even a 15W continuous power consumption, which is why I'm puzzled.
This is true that these drives I mentioned aim to have much higher IOps and STR than consumer drives but a 60TB SSD we're discussing is still not a consumer drive (at least not for the near future). To make 60TB worth of data manageable and feasible by any system that uses a single bus interface (SAS/NVMe/etc) connected to that amount of storage the controller needs to be able to manage a lot more data than any typical SSD we currently know of today. I realize this drive isn't targeting a user in need of super fast IO but the drive needs to have some significant IO in order to deal with a massive 60TB worth of information from such a condensed physical size and limited interfaces to get to it.
Mercutio
Fatwah on Western Digital
Apropos of nothing going on in the thread right now, has anyone seen a dual slot M.2 adapter that actually works and can address both drives independently? I've tried a couple that didn't work and I've yet to see one that does. I have single drive adapters and they're fine but I'd rather have one piece of hardware than two.
jtr1962
Storage? I am Storage!
Well, Merc thinks we'll probably have sub-$200 1 TB SSDs by Xmas, which implies $400 2TB drives. The smaller sizes already make lots of sense. 2TB may not make as much sense now, but once we get to under $200/TB it'll probably start to.
The 4TB Samsung still costs way too much.
Purely anecdotal but I've been watching SSD prices for a long time. It seems the price per GB drops by half about every two years on average. If this trend continues we'll be at $50/TB by the end of 2020.
The 4TB Samsung still costs way too much.
Purely anecdotal but I've been watching SSD prices for a long time. It seems the price per GB drops by half about every two years on average. If this trend continues we'll be at $50/TB by the end of 2020.
I have five 1TB class SSDs, but they are too small for practical storage purposes and too slow for desktop OS compared to M.2. I use a couple of them externally without enclosures sometimes, but splitting data is such a pain.
The larger drives should be less per GB.I have five 1TB class SSDs, but they are too small for practical storage purposes and too slow for desktop OS compared to M.2. I use a couple of them externally without enclosures sometimes, but splitting data is such a pain.
I find the other cost that doesn't get factored into pricing out new larger drives is the cost per port on the PC it will be connected to. Most of us have a limited number of SATA ports on our motherboards and unless you're willing to down the road of adding in a controller card you still have to factor in what it costs per port to connect a device. Why I think this is important is that as you buy larger drives I think the cost actually can go down in some situations because you're consuming less SATA ports. I realize it's not as simple to factor into the cost but when I look at my current workstation I have all 8 SATA ports used so adding another drive becomes more expensive for me. I either need to decommission an old drive or add a new controller.
This is especially true on a Z77 chipset motherboard with only two 6 Gbit SATA ports.
Here is some news on Samsung's 860 series m.2 formfactor NVMe drives. Some may have already seen this but worth a look. Looks pricey but seemingly decent performance numbers at face value especially when it comes to 4k random writes when compared to the 950 series. Wattage will be interesting to see on the larger drives once those get published.
Likely because they may not fit in a lot of applications with a large heatsink mounted to them.
Keep in mind:
Also:
CougTek
Hairy Aussie
I hesitated to post this here or in the Firefox thread, but there you go.
In short, Firefox will eat the lifespan of your SSD unless you modify its default parameters.
In short, Firefox will eat the lifespan of your SSD unless you modify its default parameters.
12GB per day does not seem like much. Sometimes I write a TB in a day.
mangyDOG
Learning Storage Performance
Have your tried one of these: https://www.startech.com/Cards-Adap...ds/pcie-m2-ngff-ssd-adapter-card~PEXM2SAT32N1 ?
Takes two m.2 SATA SSDs and one NVMe m.2 drive.
Have you found one that works? Looking for something like this to access client m.2 drives on my workbench pc.
Cheers,
mangyDOG
jtr1962
Storage? I am Storage!
I found some interesting reading on SSD data retention:
http://www.extremetech.com/computing/205382-ssds-can-lose-data-in-as-little-as-7-days-without-power
http://www.militarysystems-tech.com...-industrial-military-and-avionic-applications
Combining the two sources, I was able to come to some general conclusions. First look at this chart showing data retention time in weeks:
Note this these data retention numbers are when the flash is at its rated number of cycles. However, from the second source we have this:
It appears data retention times are at least 10 to 20 times longer when the cells are at 1% of rated life or less (~100 writes for MLC and 1000 writes for SLC). Given typical usage patterns, it's highly likely many SSD users will never even get to 100 writes. I'm at about 13TB of writes on my 240GB SSD after over 4 years. This translates to roughly 50 writes per cell. In any case, it's looking like the numbers in the chart can be multiplied by 5 to 10 for most users. If you write data at 40°C with a consumer SSD and store it at 25°C, then the data retention time is about two years when the drive is at the rated number of writes. Given that most drives will see far less use, you can safely multiply the numbers in the chart by 5 to 10, giving you 10 to 20 years of data retention.
It gets even better. If someone was hell bent on using an SSD for archiving, you could put it in the refrigerator at 5°C. Note that data retention times go up by a factor of two for every 5°C decrease in temperature. So now you're up to 160 to 320 years. In fact, even if the drive were at the rated number of cycles it would still retain data for over 30 years.
Bottom line-I worried somewhat about losing data when using either USB drives or SSDs in external drive enclosures but it appears to be a non-issue. In the case of USB drives especially you'll likely never see more than a few tens of writes, if that. That puts their data retention times at 10+ years even at room temperature.
http://www.extremetech.com/computing/205382-ssds-can-lose-data-in-as-little-as-7-days-without-power
http://www.militarysystems-tech.com...-industrial-military-and-avionic-applications
Combining the two sources, I was able to come to some general conclusions. First look at this chart showing data retention time in weeks:
Note this these data retention numbers are when the flash is at its rated number of cycles. However, from the second source we have this:
It appears data retention times are at least 10 to 20 times longer when the cells are at 1% of rated life or less (~100 writes for MLC and 1000 writes for SLC). Given typical usage patterns, it's highly likely many SSD users will never even get to 100 writes. I'm at about 13TB of writes on my 240GB SSD after over 4 years. This translates to roughly 50 writes per cell. In any case, it's looking like the numbers in the chart can be multiplied by 5 to 10 for most users. If you write data at 40°C with a consumer SSD and store it at 25°C, then the data retention time is about two years when the drive is at the rated number of writes. Given that most drives will see far less use, you can safely multiply the numbers in the chart by 5 to 10, giving you 10 to 20 years of data retention.
It gets even better. If someone was hell bent on using an SSD for archiving, you could put it in the refrigerator at 5°C. Note that data retention times go up by a factor of two for every 5°C decrease in temperature. So now you're up to 160 to 320 years. In fact, even if the drive were at the rated number of cycles it would still retain data for over 30 years.
Bottom line-I worried somewhat about losing data when using either USB drives or SSDs in external drive enclosures but it appears to be a non-issue. In the case of USB drives especially you'll likely never see more than a few tens of writes, if that. That puts their data retention times at 10+ years even at room temperature.
Did you get that one?
sechs
Storage? I am Storage!
At what kind of temperatures should I expect to see an M.2 SSD to throttle?
Just installed my first and, despite the fact that my BIOS doesn't seem to know anything about drive, it's working fine with Windows.
Just installed my first and, despite the fact that my BIOS doesn't seem to know anything about drive, it's working fine with Windows.