This is a really poor article. It's basically looking at spec sheets and current prices to determine the "best" drive. There doesn't seem to be any empirical testing whatsoever. They also seem to recommend product lines that are significantly pricier. eg. their "High-Capacity Desktop" recommendation is "14TB Seagate BarraCuda Pro" for $515. Is that really better than getting a 12TB easystore for $190 and then shucking it? The cost per terabyte of the easystore is less than half, and I really doubt that the recommended drive would last twice as long. Also, despite the article being less than 2 days old, most of the recommended drives are either not available, or available at significantly higher prices that listed.
They do specifically discuss "shucking" the drives from USB enclosures and recommend against it. I do wonder how well this list compares against the reports from Backblaze.
>While this is easy enough to do, the user experience might not be optimal - obtaining warranty services is pretty much ruled out
AFAIK easystores are relatively easy to shuck non-destructively. If you do it right and don't destroy the enclosure, and your drive breaks, you might still be able to submit it for warranty. Regardless, I'm skeptical about the actual value of the warranty is worth the premium (double the price).
>the default TLER settings might need alteration (which is not always possible with commercial off-the-shelf NAS units)
True, although it's likely not useful unless in business-critical workloads. You're probably fine without TLER if you're just using it to store movies/games.
>the drive is going to be used in non-critical scenarios.
What's meant by "critical scenarios" here? Either you care about the data or not. If you care about it, you will keep backups and/or use parity to ensure that one drive failure won't wipe out your data. If you're only storing on one drive, you're one unexpected disk crash from losing it. This applies regardless of whether you're using the cheapest consumer drive, or the most expensive "enterprise" drive.
Backblaze only covers the drives they have in production, which is a fairly small variation. Any smaller drives you see in their stats will be older (with more data), which means you might not see any data if you're buying a more recent smaller drive, and they may not have data on the highest end as they want to hit a good price to size/performance value for what they buy.
It's more "here's a review of what we have" and less a "here's a survey of what exists to buy", although you might be able to make some educated guesses about the latter from the former (but I would consider those low accuracy).
> Also, despite the article being less than 2 days old, most of the recommended drives are either not available, or available at significantly higher prices that listed.
In my experience with checking prices for the SSD Buyer's Guide, it is impossible to get to the end of the list before something earlier in the list has already changed in price or availability. Amazon is the worst in this respect; I'm pretty sure the act of checking prices causes price changes, and less-popular models are constantly going in and out of stock.
It seems like people who need moderate tens of terabytes of storage would be _more_ interested in the performance and reliability of an SSD, not less. If you have demanding storage needs aren't you then more likely to be willing to pay $100/TB for storage?
> you would need much more SATA ports on a machine
Why? There are 16TB SSDs u.2 SSDs on the market. That is comparable to top-of-the-line hard drives in terms of bytes per port and much, much better in terms of bytes per volume.
U.2 is a different kind of port, and one that's in notoriously short supply on all but the most expensive platforms. Even if you could get U.2 SSDs at the same price per TB as hard drives, you still need to spend a lot more to get them connected and keep them cool (enterprise SSDs don't have low-power idle states).
This conversation is increasingly weird. Enterprise SSDs definitely do have low-power modes. The Samsung PM983 I've used at scale draws 4W when idle or 11W active. A 14TB Seagate HDD like the one recommended in the article draws 5W minimum. The SSDs have a 5 milliwatt non-operating mode they can exit in less than 30ms, which is dramatically better than the latency of a non-operating hard drive (and turning a hard drive off wears it out, too). The SSD wins in this regard.
It's also virtually impossible to overheat an SSD. Samsung SSDs have critical temperatures above 90C. You won't reach this temperature just by operating the drive. The only way I've been able to make a Samsung SSD trip the critical temperature alarm was by pointing a paint-stripping air gun into the fan inlet of a machine while running an fio workload. HDDs by contrast are trivial to overheat, and are destroyed by doing it.
> The SSDs have a 5 milliwatt non-operating mode they can exit in less than 30ms,
Which mode exactly are you referring to? Because the 983 DCTs (retail version of the PM983) that I have don't support features like NVMe APST, and idle at 3+ W (for the M.2 version, which is lower-power than the U.2 version).
> It's also virtually impossible to overheat an SSD.
Maybe in a properly designed rackmount enclosure with fans that never drop below 4k RPM. But in a more typical consumer case, enterprise U.2 drives will absolutely overheat, and I have the logs to prove it. (The Samsung drives you refer to are definitely better about this than the kind of U.2 drives that use 15mm thick cases and controllers with more than 8 channels—which I believe includes all U.2 drives with capacity in excess of 8TB.) Consumer cases that still have 3.5" bays tend to provide plenty of airflow for them, but that's not the case for 2.5" bays.
Maybe we should avoid the specifics of Samsung, since they are in the habit of shipping different firmware to retail and bulk buyers. I've never used the retail 983, and I have no experience with the u.2 either. All I can say for sure is you cannot overheat the stick version. If it overheats that's because your datacenter is on fire.
I can replicate this at home with a Samsung 970 PRO m.2 in an Intel NUC. It just will not overheat. The only thing I can do to make it shut down is to put blankets over the NUC in which case the CPU has gone PROCHOT anyway.
Bottom line, it seems to me that power and heat are advantages of SSD, not HDD. It seems like people buy HDD purely because of GB/$, which makes sense to me at scale but not at home. The care and management of hard drives, and their atrocious performance, seems to outweigh the price advantage. It's really hard to manage HDDs successfully and even if you are doing it perfectly, it still means you are replacing the HDD every 3-5 years. What a pain.
> It's really hard to manage HDDs successfully and even if you are doing it perfectly, it still means you are replacing the HDD every 3-5 years. What a pain.
Neither HDDs nor SSDs will last forever.
It's as easy as sticking a pair of drives in a cheap NAS enclosure and listening for the warning alarm when SMART predicts failure or a drive fails.
I did have to replace the previous NAS after 5 years-- not because of failure, but because I needed more storage and wanted more performance. Otherwise it might have lasted quite a bit longer.
An SSD that's only being read really will last ~forever. It has the same wear-out process as a CPU or RAM. A hard drive will wear out based on power-on hours and start/stop cycles, regardless of how much you use it. I'm not saying you should blindly assume that your SSD will never break, since they can and do randomly stop working, but it seems clear that reliability is a clear win for SSD over HDD, especially in consumer applications where the mere passage of time is the thing that will kill hard drives.
> It has the same wear-out process as a CPU or RAM.
This isn't really true-- flash isn't really like DRAM; leakage currents tend to increase even if you don't reprogram things (and to properly fight leakage-related data corruption, you really need to periodically rewrite and do the erases that further compound dielectric breakdown). And it's especially not true of board-level assemblies, where capacitor failure and solder joints become important.
> A hard drive will wear out based on power-on hours and start/stop cycles, regardless of how much you use it.
Some failure modes look like this in SSDs, too. Studies have shown that SSD age is a much better predictor of failure probability than usage. Depending on how firmware handles things and the cell characteristics, -insufficient use- of a SSD can cause failure, too.
I mean, I did some work on firmware for early SSDs and error rate / lifetime analysis and I've had SSDs lose data in a drawer unused, but maybe you know better. Yes, SSDs are better than HDDs, but not as drastically so as one would think.
I <3 flash. But I also like having big dumb spinny disks. I'm rather glad I can have both.
The article is about consumer hard drives, but I guess if you have an infinite budget and access to any hardware you want there's no point at looking at hard drives.
If we take my case at home, I would need about 30 TB of total capacity to be at ease. Let's disregard my previous comment about backups of backups and go for "just" 60 TB total capacity.
Where I live I can only reliably source 14 TB hard disks at €400 a piece and 4 TB SATA SSDs for €550 (I didn't look at NVME since the number of ports possible on one machine is too small).
So I would need 5 disks for 2000€ or 15 SSDs for 8250€.
Almost any motherboard comes with 6 SATA ports but for the SSDs I would need to add at least one PCI Express raid card to get more SATA ports.
FWIW I could buy the cheapest new car available in France for the same price as the SSDs (Dacia Sandero base model).
Home users that want to store tens of TB of data are highly unlikely to need fast access to most of it. A large personal movie collection never needs more performance than a single hard drive offers. An incremental backup archive is not performance-critical while adding new backups, and restoring from backup is infrequent enough that it's not worth paying an order of magnitude higher for better restore performance. A video game collection can see some benefit from faster storage, but it's still cheaper and easier to shuffle games in and out of local ~1TB SSD storage than to establish high-speed access to a solid-state NAS.
Depends on access patterns. If you have tons of video, audio, sensor data, the access is mostly sequential and processing it is mostly CPU bound (unless you have a remarkable workstation). If a hard disk costs about half per TB than an SSD, you can RAID-1 2 of them for twice the reliability and read speed. More disks, more reliability and more speed.
Here where I live I can get a 6TB serious HPE SSD for €1781.63 or a 4TB consumer grade unit for €671.88. For that amount of money, I can get a 3-drive RAID-1 of Hitachi 10TB disks or, better, a 2-drive RAID-1 (certainly not as fast as the HPE board) and keep the third disk in another machine and do backups there (because the reliability of an SSD means nothing if the data is erased).
Not necessarily you can have terabytes of just video footage or photos which isn't that performance intensive or just plain old backup. Depends on the use case and price. Plus I have had a Samsung SSD fail on me and had to RMA it. So I would tell anyone SSDs are not fail proof.
Consumer + lots of storage usually equals lots of cold data that you don't care how fast it is. Lots of cold data without strict performance requirements calls for spinning disks still-- whether we're talking about enterprise or consumer. Paying $100/TB for things that I might access once per year and will still be close to instant when I choose to (video, photo) is silly when I can pay $250 for a RAID-1 pair of 12TB drives.
Backblaze does it the best way IMHO, they review the drives based on the time they have them in production running and based on the effective failures to write operations & consider the $/Tb as well and they do this every quarterly and publish the result.
Check those out once and see if you can get the data you need
I read somewhere recently (might have been here, I don’t recall for sure), that one of the problems with the recommendations from Backblaze is that the drives change constantly and there can be variations over time in manufacturing. You could buy something today that has the same model number as what Backblaze bought 6-12 months ago, and it’s not really the same drive.
They’re still my preferred resource for this, but it’s not without its caveats.
That is a very valid criticism, but at the same time, could anyone really do any better? If the manufacturer is going to change the build for the same SKU within the time it takes to collect meaningful failure rate data, then I don't know that you can construct a more useful measurement technique than Backblaze has. The data on specific drive models may be of limited reliability, but you do at least get enough data to rank brands based on typical long-term reliability by brand and product line.
But then you need the datacenter operator to use the same generic drives you would. Backblaze is the only one who publishes results for the drives they have, which are more or less close to what a prosumer would pursue.
I find it very hard to browse all these comparison sites. For example i need a very simple question: whats the very best SSD i can buy 1TB for under or near $500. So much confusion, especially Samsung with all the EVO and PRO and numbers and whatever models sheesh.
> whats the very best SSD i can buy 1TB for under or near $500
The problem is that you have already made some kind of decision that $500 is a reasonable threshold for getting 1TB of SSD, and some assumption that you stand to gain anything at all by paying extra for the same capacity. That's kinda backwards; you should start by figuring out how much performance and endurance you need out of your 1TB of storage, just like you've apparently already figured out that 1TB is the capacity you need.
You might be fine with a $99 1TB SSD, or maybe you need a $320 1TB SSD. But it's pretty odd to expect a purchasing guide to be a useful tool to figure out how best to overpay for your storage.
38 comments
[ 2.7 ms ] story [ 90.3 ms ] thread[1] https://slickdeals.net/newsearch.php?src=SearchBarV2&q=easys...
>While this is easy enough to do, the user experience might not be optimal - obtaining warranty services is pretty much ruled out
AFAIK easystores are relatively easy to shuck non-destructively. If you do it right and don't destroy the enclosure, and your drive breaks, you might still be able to submit it for warranty. Regardless, I'm skeptical about the actual value of the warranty is worth the premium (double the price).
>the default TLER settings might need alteration (which is not always possible with commercial off-the-shelf NAS units)
True, although it's likely not useful unless in business-critical workloads. You're probably fine without TLER if you're just using it to store movies/games.
>the drive is going to be used in non-critical scenarios.
What's meant by "critical scenarios" here? Either you care about the data or not. If you care about it, you will keep backups and/or use parity to ensure that one drive failure won't wipe out your data. If you're only storing on one drive, you're one unexpected disk crash from losing it. This applies regardless of whether you're using the cheapest consumer drive, or the most expensive "enterprise" drive.
Isn't that almost exactly the purpose of Backblaze's HD stats?
https://www.backblaze.com/b2/hard-drive-test-data.html
It's more "here's a review of what we have" and less a "here's a survey of what exists to buy", although you might be able to make some educated guesses about the latter from the former (but I would consider those low accuracy).
In my experience with checking prices for the SSD Buyer's Guide, it is impossible to get to the end of the list before something earlier in the list has already changed in price or availability. Amazon is the worst in this respect; I'm pretty sure the act of checking prices causes price changes, and less-popular models are constantly going in and out of stock.
Not everyone has a photo library or media library or game collection that fits onto a reasonably priced SSD.
So in the end the choice is SSDs if you need fast storage or hard drives if you need cheap(er) storage.
Even if SSDs were priced the same as hard drive you would need much more SATA ports on a machine to achieve a big storage capacity.
Why? There are 16TB SSDs u.2 SSDs on the market. That is comparable to top-of-the-line hard drives in terms of bytes per port and much, much better in terms of bytes per volume.
It's also virtually impossible to overheat an SSD. Samsung SSDs have critical temperatures above 90C. You won't reach this temperature just by operating the drive. The only way I've been able to make a Samsung SSD trip the critical temperature alarm was by pointing a paint-stripping air gun into the fan inlet of a machine while running an fio workload. HDDs by contrast are trivial to overheat, and are destroyed by doing it.
Which mode exactly are you referring to? Because the 983 DCTs (retail version of the PM983) that I have don't support features like NVMe APST, and idle at 3+ W (for the M.2 version, which is lower-power than the U.2 version).
> It's also virtually impossible to overheat an SSD.
Maybe in a properly designed rackmount enclosure with fans that never drop below 4k RPM. But in a more typical consumer case, enterprise U.2 drives will absolutely overheat, and I have the logs to prove it. (The Samsung drives you refer to are definitely better about this than the kind of U.2 drives that use 15mm thick cases and controllers with more than 8 channels—which I believe includes all U.2 drives with capacity in excess of 8TB.) Consumer cases that still have 3.5" bays tend to provide plenty of airflow for them, but that's not the case for 2.5" bays.
I can replicate this at home with a Samsung 970 PRO m.2 in an Intel NUC. It just will not overheat. The only thing I can do to make it shut down is to put blankets over the NUC in which case the CPU has gone PROCHOT anyway.
Bottom line, it seems to me that power and heat are advantages of SSD, not HDD. It seems like people buy HDD purely because of GB/$, which makes sense to me at scale but not at home. The care and management of hard drives, and their atrocious performance, seems to outweigh the price advantage. It's really hard to manage HDDs successfully and even if you are doing it perfectly, it still means you are replacing the HDD every 3-5 years. What a pain.
Neither HDDs nor SSDs will last forever.
It's as easy as sticking a pair of drives in a cheap NAS enclosure and listening for the warning alarm when SMART predicts failure or a drive fails.
I did have to replace the previous NAS after 5 years-- not because of failure, but because I needed more storage and wanted more performance. Otherwise it might have lasted quite a bit longer.
This isn't really true-- flash isn't really like DRAM; leakage currents tend to increase even if you don't reprogram things (and to properly fight leakage-related data corruption, you really need to periodically rewrite and do the erases that further compound dielectric breakdown). And it's especially not true of board-level assemblies, where capacitor failure and solder joints become important.
> A hard drive will wear out based on power-on hours and start/stop cycles, regardless of how much you use it.
Some failure modes look like this in SSDs, too. Studies have shown that SSD age is a much better predictor of failure probability than usage. Depending on how firmware handles things and the cell characteristics, -insufficient use- of a SSD can cause failure, too.
I mean, I did some work on firmware for early SSDs and error rate / lifetime analysis and I've had SSDs lose data in a drawer unused, but maybe you know better. Yes, SSDs are better than HDDs, but not as drastically so as one would think.
I <3 flash. But I also like having big dumb spinny disks. I'm rather glad I can have both.
If we take my case at home, I would need about 30 TB of total capacity to be at ease. Let's disregard my previous comment about backups of backups and go for "just" 60 TB total capacity.
Where I live I can only reliably source 14 TB hard disks at €400 a piece and 4 TB SATA SSDs for €550 (I didn't look at NVME since the number of ports possible on one machine is too small).
So I would need 5 disks for 2000€ or 15 SSDs for 8250€. Almost any motherboard comes with 6 SATA ports but for the SSDs I would need to add at least one PCI Express raid card to get more SATA ports.
FWIW I could buy the cheapest new car available in France for the same price as the SSDs (Dacia Sandero base model).
Here where I live I can get a 6TB serious HPE SSD for €1781.63 or a 4TB consumer grade unit for €671.88. For that amount of money, I can get a 3-drive RAID-1 of Hitachi 10TB disks or, better, a 2-drive RAID-1 (certainly not as fast as the HPE board) and keep the third disk in another machine and do backups there (because the reliability of an SSD means nothing if the data is erased).
Check those out once and see if you can get the data you need
They’re still my preferred resource for this, but it’s not without its caveats.
>For Sata: DC S3610 1.6tb Intel SSD -- under $200 right now on ebay
>For NVMe DC P3605 1.6tb Intel SSD -- usually around $200 on ebay (addin card) or a little more as U.2 drive.
[0] https://www.webhostingtalk.com/showthread.php?t=1818777&NL=W...
The problem is that you have already made some kind of decision that $500 is a reasonable threshold for getting 1TB of SSD, and some assumption that you stand to gain anything at all by paying extra for the same capacity. That's kinda backwards; you should start by figuring out how much performance and endurance you need out of your 1TB of storage, just like you've apparently already figured out that 1TB is the capacity you need.
You might be fine with a $99 1TB SSD, or maybe you need a $320 1TB SSD. But it's pretty odd to expect a purchasing guide to be a useful tool to figure out how best to overpay for your storage.