TR Forums

I've got a WD Green 4 TB from a few years back. It's starting to get low on space and I'd like to replace it with something larger. It replaced a 1 TB WD Green and that provided a very nice boost in read and write speed, so I'd like to see a similar bump in speed again. 8 TB in capacity is about what I'm looking for.

My case has plenty of airflow, so heat's not a concern, but I'm wondering how much noise difference there would be if I went with a 7200 RPM drive?

It'll be used in a desktop system for mass storage, so it'll spend most of its time idle if that can be an issue?

Thanks!

Assuming you stick with lower RPM, I don't think you will get a comparable speed boost this time, as I believe the 8TB models use more platters with only modest increase in linear bit density.

Also make sure you get a PMR drive, not SMR. Performance of SMR drives for non-sequential writes is abysmal (they are intended mainly for archival/backup use).

I don't think any of the WD Greens use SMR but I am not 100% certain.

Noise-wise, higher RPM alone isn't going to make a huge difference, since all drives use fluid dynamic bearings these days. But a 7200 RPM drive's head actuator will likely be tuned for higher performance (shorter seek times); more platters also means the head actuator assembly is heavier. So you will likely hear the drive when it is doing a lot of seeking, and the vibration caused by rapidly accelerating/decelerating that heavier head assembly may also be amplified by your case (depending on the design of your drive cage).

If you're going to be using it as a single drive I would probably get a 7,200 RPM unit to enable slightly better performance than a 54-5900RPM drive. I'm assuming you're using an SSD as your OS/programs drive rather than a WD green, but I don't want to assume that incorrectly either.

If you haven't already thought of it, I would take the time to consider how you're going to backup all your data. Possibly sticking the 4TB drive that is coming out in an external enclosure to put the MOST important things on and get that offsite or at least disconnected from your main system. One suggestion if you don't mind voiding a warranty would be to get an 8TB external drive which for whatever reason are cheaper than buying the internal equivalent, shuck the 8TB drive from within it, use the enclosure for your 4TB drive and you'd be all set - other than not having a warranty anymore.

Alternatively, you could just use the 8TB as an external to backup everything on the 4TB as well as storing less critical (not needing backup) things until the 1 year warranty is up and then make a swap - you would already have all the important stuff on the 8TB and the 4TB could become the backup at that point.

https://www.amazon.com/Seagate-IronWolf ... B01M1BUBSO

or the 10 TB version:

https://www.bhphotovideo.com/c/product/ ... _sata.html

I went with the 8TB Seagate off Newegg. Got it for $239, so it was a pretty good deal.

I've got an Intel 750 400GB as my system drive and will use the 4TB drive as a temporary backup until I get a real solution. Thanks, guys!

just brew it! wrote:

Assuming you stick with lower RPM, I don't think you will get a comparable speed boost this time, as I believe the 8TB models use more platters with only modest increase in linear bit density.

Also make sure you get a PMR drive, not SMR. Performance of SMR drives for non-sequential writes is abysmal (they are intended mainly for archival/backup use).

Two points:

1. Density has gone up quite a bit, so even the lower RPM drives still reach nearly 200 MB/s sequential at the beginning of the platters.
2. Seagate is the only company shipping SMR drives that are drive-managed. That said, bursty writes (less than 30 GB at a time) on SMR drives are actually quite a bit *better* than traditional HDDs. They just suck if you hammer them.

Waco wrote:

Two points:

1. Density has gone up quite a bit, so even the lower RPM drives still reach nearly 200 MB/s sequential at the beginning of the platters.

I just spot checked some of the 3TB and 4TB drives in my home desktop and server; these are drives which were purchased in the past 1-2 years (so they should have platter densities pretty close to the current state-of-the-art), and are 7200 RPM drives. They're all doing around 190 MB/sec on the outer tracks. I suppose it's possible that linear bit density has improved significantly in the past year or two, but I was under the impression that most of the capacity gains lately have been coming from tighter track pitch, not linear bit density.

Waco wrote:

2. Seagate is the only company shipping SMR drives that are drive-managed. That said, bursty writes (less than 30 GB at a time) on SMR drives are actually quite a bit *better* than traditional HDDs. They just suck if you hammer them.

Yeah, I suppose my rather dim view of SMR drives is due in part to the fact that I deal with datacenter applications at my day job. Hard drives getting hammered is par for the course where I work...

I will likely be dealing with HGST host-managed SMR drives in the medium-term, unless SMR as a tech falls out of favor.

just brew it! wrote:

I will likely be dealing with HGST host-managed SMR drives in the medium-term, unless SMR as a tech falls out of favor.

I highly doubt SMR is going to fall out of favor - I'd bet that it'll be in every drive with the exception of NL-SAS in the next few years as cheap NAND continues to push on the HDD market.

Seagate has made leaps and bounds improvement on behavior for their newer SMR drives, so I wouldn't write them off just yet.


EDIT: For density, I guess I was assuming a 4-5 year old 4 TB drive versus a modern drive. There's a 30% or so bump in density on the newer platters IIRC.

Does anyone have details about the specific terribleness of SMT drives? Other than obviously the horrible re-write speeds.

I'm asking in part because I mostly use my large drive for media storage and playback, the contents haven't changed (other than adding a few large files) for years and I'm looking for a replacement.

Vhalidictes wrote:

Does anyone have details about the specific terribleness of SMT drives? Other than obviously the horrible re-write speeds.

I'm asking in part because I mostly use my large drive for media storage and playback, the contents haven't changed (other than adding a few large files) for years and I'm looking for a replacement.

I run many thousands of them in production for mostly-read workloads. The gotchas are the inconsistent write performance (if you do small random writes or large sustained writes) and rebuild speeds (for the same reasons).

Read performance is essentially the same as a CMR drive.

Waco wrote:

just brew it! wrote:

Assuming you stick with lower RPM, I don't think you will get a comparable speed boost this time, as I believe the 8TB models use more platters with only modest increase in linear bit density.

Also make sure you get a PMR drive, not SMR. Performance of SMR drives for non-sequential writes is abysmal (they are intended mainly for archival/backup use).

Two points:

1. Density has gone up quite a bit, so even the lower RPM drives still reach nearly 200 MB/s sequential at the beginning of the platters.
2. Seagate is the only company shipping SMR drives that are drive-managed. That said, bursty writes (less than 30 GB at a time) on SMR drives are actually quite a bit *better* than traditional HDDs. They just suck if you hammer them.

Does "drive-managed" mean they're zeroing out empty space like TRIM on an SSD?

evilpaul wrote:

Does "drive-managed" mean they're zeroing out empty space like TRIM on an SSD?

No... SMR drives don't have wear leveling issues, but (like SSDs) some trickery is required to make them look like a "normal" random-access disk device. SMR drives are divided into "zones", which are typically a few hundred MB in size. Within a zone, data must be written sequentially. Reads can still be done randomly, like on a normal drive.

So updating a sector requires reading the entire zone into a buffer, replacing the modified sector, then writing the entire zone out again.

In a drive-managed SMR drive, this behavior is hidden from the host system by the drive firmware. Random writes get redirected to a cache; this cache is several GB in size, and typically kept on a non-SMR part of the disk. In the background, the drive performs the (slow) process of updating the SMR zones with the newly written data. Of course, reads of recently written data also need to be redirected to the cache, since that data is not in its final location yet.

There are also host-manged and host-aware SMR drives. Host-managed drives don't have the extra logic to "hide" the underlying storage layout, requiring that the host rearrange write operations to be sequential within each zone; this requires a SMR-aware file system on the host. Host-aware SMR drives allow random writes from the host (using a caching algorithm like drive-managed SMR), but also expose the zone geometry to facilitate more efficient operation for hosts which have SMR-aware file system support.

I suppose you could draw some analogies with SSD TRIM for host-managed and host-aware SMR drives. The host explicitly tells the drive "I am about to start rewriting zone X", which is sort of like TRIMming that entire zone. I believe reading beyond the rewritten portion of a zone on a host-managed drive results in a read error though (not zeroes).

The 8TB Red is a helium-filled drive that actually idles even quieter than your old 4TB Red, performance is pretty pedestrian at 5400MB/s but firmware and areal density improvements mean it'll be around 50% quicker than your current 4TB Red - sequential transfers of large media files tend to run at about 150-180MB/s

The Seagate Ironwolf 10TB is also a decent drive, as it is also helium-filled. I use 12 of them in a NAS here are they're reasonably quiet and the performance per disk of 200-250MB/s for sequential transfers is about as good as it gets for mechanical drives.

Importantly, both the 8TB Red and 10TB Ironwolf are the good old "standard" PMR technology rather than SMR which provides 30% more capacity per platter at the cost of order-of-magnitude write latency increases. If you've never experienced the pain and misery of using an SMR drive as a general purpose desktop drive, count yourself lucky. Their only valid usage case is in surveillance/DVR boxes where everything is purely sequential.

If you're at all concerned about noise, don't get anything faster than 5900rpm and remember that even ignoring the spindle speed, more platters to spin means more noise. This is why I wouldn't even consider a non-helium drive over 4TB these days (but I have particularly strong hate for the high-pitched whine of overloaded 7200rpm bearings and the vibration it causes throughout the rest of the case. YMMV!)

My most recent drives purchased were two 8tb WD Red drives. From a SSD to the Red I get about 160mb-180mb/sec depending on file size. They're quiet, fast and cheap. I'd highly recommend them.

Before that were 2x 4tb HGST about two years ago, I'd recommend them but only because I like to have at least two brands in my archive. They act weird (transfer speeds dropping to single mb/sec but Task Manager showing 100% drive utilization) so I'm hesitant to give them much praise. No dead drives mind you, but none of the other drives in my server do that. My vintage 4TB Greens are still going strong, and even older ones are in cold storage now.

Also have to 4tb-5tb Seagate external drives that I got super cheap. Normally I'd never recommend them but depending on the sale it's not a bad idea to pick one up along with some internal drives just in case you need to take your storage with you. Of course there's external WD drives that hold up very well too.

My biggest thing is to just make sure your archive isn't all made up of the same drives - get different mfg, different "brand lines" (red, green), different lot numbers...something to ensure that if there was a defect on a certain lot of drives, all of yours aren't affected. That's why I usually replace two out of my eight drives every year or two, to keep cycling them and to put others into cold storage or move to a offsite location (your entire music/movie collection makes a great gift to a relative across the country )

just brew it! wrote:

evilpaul wrote:

Does "drive-managed" mean they're zeroing out empty space like TRIM on an SSD?

No... SMR drives don't have wear leveling issues, but (like SSDs) some trickery is required to make them look like a "normal" random-access disk device. SMR drives are divided into "zones", which are typically a few hundred MB in size. Within a zone, data must be written sequentially. Reads can still be done randomly, like on a normal drive.

So updating a sector requires reading the entire zone into a buffer, replacing the modified sector, then writing the entire zone out again.

In a drive-managed SMR drive, this behavior is hidden from the host system by the drive firmware. Random writes get redirected to a cache; this cache is several GB in size, and typically kept on a non-SMR part of the disk. In the background, the drive performs the (slow) process of updating the SMR zones with the newly written data. Of course, reads of recently written data also need to be redirected to the cache, since that data is not in its final location yet.

There are also host-manged and host-aware SMR drives. Host-managed drives don't have the extra logic to "hide" the underlying storage layout, requiring that the host rearrange write operations to be sequential within each zone; this requires a SMR-aware file system on the host. Host-aware SMR drives allow random writes from the host (using a caching algorithm like drive-managed SMR), but also expose the zone geometry to facilitate more efficient operation for hosts which have SMR-aware file system support.

I suppose you could draw some analogies with SSD TRIM for host-managed and host-aware SMR drives. The host explicitly tells the drive "I am about to start rewriting zone X", which is sort of like TRIMming that entire zone. I believe reading beyond the rewritten portion of a zone on a host-managed drive results in a read error though (not zeroes).

This is spot-on, but I'll note that the zones are nowhere near a few hundred MB in size on drive-managed drives (think tens of MB). It's not a published spec, but it's been determined experimentally on the commonly available drives via a few very interesting papers (and those numbers hold true from my experiments as well).

The newer SMR drive-managed drives have a lot of new logic to increase performance for non-perfect workloads. You can absolutely still hammer them into submission, but in general, the problems with RAID rebuilds of the past (fully sequential drive overwrites) are gone. Bursty consumer-type writes of a few tens of GB should be no real issue and subjective performance can be better than traditional HDDs.

Waco wrote:

This is spot-on, but I'll note that the zones are nowhere near a few hundred MB in size on drive-managed drives (think tens of MB). It's not a published spec, but it's been determined experimentally on the commonly available drives via a few very interesting papers (and those numbers hold true from my experiments as well).

That's somewhat surprising. Modern drives store on the order of 2 MB/track; if we assume a 5 platter design (10 surfaces), that's 20 MB per cylinder. With only "tens of MB" per zone, you'd be overlapping only 2-4 cylinders in each SMR zone, leaving a lot of potential capacity on the table.

It'll certainly help performance though, since doing the read-modify-write operation on a zone will be a lot faster.

just brew it! wrote:

That's somewhat surprising. Modern drives store on the order of 2 MB/track; if we assume a 5 platter design (10 surfaces), that's 20 MB per cylinder. With only "tens of MB" per zone, you'd be overlapping only 2-4 cylinders in each SMR zone, leaving a lot of potential capacity on the table.

It'll certainly help performance though, since doing the read-modify-write operation on a zone will be a lot faster.

I've seen proposals for many hundred / gigabyte size tracks but the performance would be so bad in typical consumer workloads that they've avoided pushing the zones so wide for a handful of percent capacity gain. The zone widths can't easily approach the amount of DRAM on the HDD either (for drive-managed and host-aware designs) if performance is to be maintained at even a minimal level. There are many tricks manufacturers are playing to keep minimum performance bounded in "bad" workload conditions and almost all of them rely on the zone sizes not growing overly large.

I could go further in-depth on how these things work, but suffice to say, they're incredibly complicated.

I'll mention that TiVo Bolt is one of the many applications for which Seagate's problem-plagued SMR drives are totally unusable. After several hours or a few days on-line, the Seagate drive corrupts the file system and makes the DVR completely unusable. I'll be sticking to other brands of drives from now on.

Depending on what you're using the drive for I'd either get a Seagate Archive 8TB or a Barracuda 8TB drive. If it's just for media storage the value of the Archive drive can't be beat, but if it's for general purpose usage I'd get the Barracuda. They're both air drives so they're cheaper than helium drives (and you don't have to worry about the helium leaking out.) I'd also consider the Toshiba X300 8TB since that is a new 8TB air drive, but I have heard that Toshiba drives are louder than the competition. YMMV

P.S. I also really hate when people call non-SMR drives PMR. All modern drives are PMR, even the SMR ones. I like to use CMR (conventional magnetic recording) and SMR (Shingled magnetic recording) to differentiate between the two.

JustAnEngineer wrote:

I'll mention that TiVo Bolt is one of the many applications for which Seagate's problem-plagued SMR drives are totally unusable. After several hours or a few days on-line, the Seagate drive corrupts the file system and makes the DVR completely unusable. I'll be sticking to other brands of drives from now on.

An SMR drive is a horrible choice for a DVR. Especially since all a DVR does is hammer the drive with writes. Don't hate the brand just because you used one of their drives in a use-case it was never designed for.

I hate the company because Seagate obfuscates which of their drive models have the inferior SMR technology and which will actually work for common applications. If the fact that certain models or revisions in a series of drives use SMR technology were available in the marketing materials for that drive, I would have been able to avoid the aggravation and lost time by using the models or versions that did not have the problematic SMR.

JustAnEngineer wrote:

I hate the company because Seagate obfuscates which of their drive models have the inferior SMR technology and which will actually work for common applications. If the fact that certain models or revisions in a series of drives use SMR technology were available in the marketing materials for that drive, I would have been able to avoid the aggravation and lost time by using the models or versions that did not have the problematic SMR.

It's pretty easy to figure out. If it's an internal drive and has the word "Archive" in the name it's SMR. If it's an external drive only the 8TB 3.5" are SMR. Seagate literally only makes 1 model of 3.5" drive that is SMR. (edit: oh and their older 5TB 3.5" external drive, but AFAIK those aren't made any more and seem to be low in stock everywhere)

This drive right here is one of the crappy ones that uses SMR:
https://www.newegg.com/Product/Product. ... 6822178871
Avoid.

For Seagate drive-managed:

Certain 3 TB and 5 TB 2.5" drives are SMR (it's harder to tell for 2.5" drives). This obfuscation really annoys me.
Archive 5 TB and 8 TB 3.5" drives are SMR (if you shuck one, it's a 99% chance it's SMR). The 5 TB drive is measurably worse in every performance metric and has been EOL'd since last year. Avoid if possible, IMO.


The newer drives will probably handle DVR duty just fine, as long as the bitrate your storing at isn't excessive. Any filesystem that gets *corrupted* from slow hardware, though, is terrible to begin with.