HDD performance in a nutshell
Last week, we posted our new SSD test suite with some initial results. We included a few references to hard drive performance, and at the time that seemed sufficient, but let's really put the hard drive vs. solid state performance debate to rest. We all know SSDs are faster, particularly at random accesses; hard drives, in contrast, offer massive amounts of storage at bottom-basement prices. For archiving large files and backups, hard drives can be great, and they've been around so long that they're a proven technology. That doesn't mean they don't fail on occasion, but recovering data from a HDD (via a recovery service) tends to be about half to one-fourth the cost of recovering data from a failed SSD. You've all got a backup strategy in place, right? Good. Moving on….
The venerable hard disk drive, or HDD, in 3TB capacity, circa 2012.
Even the worst of modern SSDs tend to smoke HDDs when it comes to performance metrics. And over time, HDDs are still prone to file fragmentation, which is why that well-worn copy of Windows 7 running off an HDD can take several minutes to boot up after a few years of use. Many people still use HDDs, however, probably in large part because they simply don't know any better. So here's your chance: The next time someone asks you why they need to pay more money to get an SSD, point them at this article.
For the tests today, we've got the same collection of SSDs used in our OCZ Trion 100 480GB review. The Trion costs just $0.33 per GB—and the 960GB model goes even lower at less than $0.32 per GB—but neither one can hope to match the price of HDD storage. Our Seagate 3TB drive, for example, costs just $0.03 per GB. Yes, that's a full order of magnitude more expensive (per GB).
Now, we don't like to beat dead horses, but as we noted in the review, the Trion 100 just isn't worth the asking price, considering faster SSDs can be had for roughly the same price. But this does bring up an interesting point: If the Trion 100 is the slowest "modern" SSD we've tested, how does it fare against a decent HDD? To settle the matter, we ran our storage suite on a Seagate ST3000DM001 3TB drive. Let's not mince words: Some of the tests are painfully slow to run. It was so bad that on the random IO testing with AS SSD, we dropped back to a 1GB test instead of 10GB, and it still took over ten times as long to finish the test. There's that order of magnitude business again; it's likely to be a common refrain.
Our storage test bed consists of a modern Skylake processor in a Z170 motherboard. It doesn't really affect performance with a hard drive, but we're keeping things consistent. Here's our test bed, and then we'll get to the results of our testing.
Orders of magnitude, in charts
Our benchmarks include a mix of real-world and synthetic testing. We normally run each test multiple times, though we admit that on the HDD we just took the results of a single run for the lengthier tests. Here are the results, with the HDD highlighted in all the charts.
The AS SSD Sequential testing gives us the best-case pure transfer performance. This is as good as it gets for both hard drives and solid state drives. Note also that the hard drive (and SSDs) is basically empty at the time of testing; if it were full of fragmented files, performance on the HDD could be substantially lower. The same is only partially true for SSDs; some can get in a degraded performance state after lots of use, but TRIM and other features will keep a modern SSD much closer to "like new" performance. (Note that the SSDs have at least had a full drive's worth of writes, however, so we're not showing absolute best-case SSD performance.)
If you look at the sequential transfer rates, hard drives don't look so bad. Sure, a slow SSD is typically twice as fast, but you could live with that, right? Meanwhile, a modern NVMe drive boasts a 10X performance increase, but we're talking about a $90 3TB HDD vs. a $1050 1.2TB SSD—$0.03 per GB vs. $0.87 per GB means it's also 30X as expensive. Interestingly, the Intel SSD 750 "only" costs 3X as much per GB as the OCZ Trion 100 and other inexpensive SSDs, so if you do lots of sequential transfers you don't even get diminishing returns. It's also worth pointing out that write performance on the Trion 100 tends to start high, at over 350MB/s, but over the length of the test sequence the transfer rate steadily declines. For bursty workloads, the Trion actually fares much better; it's just sustained writes where it starts to struggle.
Moving over to AS SSD's Random testing, suddenly the problem with HDDs becomes strikingly obvious. The slowest SSD in our charts is over 40x as fast as the HDD in read performance, and over 60x faster in write performance… and that's at a queue depth of one. SSDs shine when they have more stuff going on, and at QD64 the SSDs are 250x to 1,150x (that's over three orders of magnitude!) faster at reads, and 125x to 1,150x faster at writes. Suddenly, that 30x advantage in price per GB that the HDD holds doesn't look quite so compelling, does it?
We use IOmeter as a second test of theoretical read/write performance for sequential and random workloads, the difference being that we mix the reads and writes and take the geometric mean of the five mixes. The mixing of reads and writes has a clear impact on all the drives, though it's most pronounced on the HDD and less on the SSDs. Sequential read/write speeds are pretty similar to the pure write speeds shown in AS SSD on several of the drives, though the OCZ Trion and Intel SSD 750 both perform better in the IOmeter write testing.
The mixed random read/write, on the other hand, continues to separate the contenders from the pretenders, and the HDD can only look on in wonder. If you're still running a hard drive for your OS, this is a great illustration of why a system with an SSD can boot the OS in under 15 seconds, while you might end up waiting several minutes for a hard drive to become fully ready for use. Windows will load all sorts of drivers and kernel files, and the access pattern ends up looking more random than sequential in many cases. The differences become a bit less pronounced once things settle down—assuming that you have sufficient system RAM, of course—but even mediocre SSDs will easily beat the fastest HDDs.
We've covered mostly theoretical testing so far, but what does this mean in the "real world"? We'll start with a file copy test, creating a clone of just shy of 20GB of data (the contents of our Steam Batman: Arkham Origins folder). This provides a split workload of 50/50 reads/writes, though the use of PowerShell tends to deliver results closer to sequential performance than random IO. (Interestingly, doing the file copy within Windows Explorer tends to be about 10–15 percent slower than using PowerShell.) We're back to the slower SSDs only being twice as fast as a clean hard drive, while the fastest NVMe drives are nearly an order of magnitude faster; a moderately fragmented HDD would only make matters worse.
Last up is PCMark 8's Storage test. The overall score shows less than a 2x difference between HDDs and SSDs, which makes sense: Over the course of a day using your computer, having an SSD won't usually make you twice as efficient. Certain tasks will go much faster with the SSD, but for office work, most of the time the PC is waiting on you—with or without an SSD. In contrast, the Bandwidth result focuses on the times when you're waiting on the computer, specifically when you're waiting for storage. It illustrates why systems with SSDs feel so much faster: Those times where you're waiting impatiently for the PC to boot your OS or load an application are what you notice the most.
Stick a fork in the HDD
There's a saying: "Once you go SSD, you'll never go back." It's absolutely true. Personally, I was fine with my HDD storage (on an old Bloomfield system, if you must know). I didn't usually shut down the system at night (power bills be damned!), so it was right where I left it in the morning. About the most painful aspect was when I would occasionally need to restart Chrome, complete with my 20 or so active tabs—it might take 15–20 seconds before all of the tabs were finished loading. Basically, I didn't really think I needed an SSD enough to warrant the expense. And then I got a decent-sized 120GB SandForce drive and suddenly I needed lots more SSD capacity. 240/256GB is typically enough, at least for my purposes, but 480/512GB is the sweet spot where you mostly stop worrying about how much free space is still available.
Several years later, prices on SSDs have fallen to the point where most enthusiasts won't even touch a new PC without one. And as someone who routinely ends up troubleshooting computer problems for my family and friends, doing malware scans on hard drive–based PCs is pure torture. Scan. Reboot. Scan some more. Reboot some more. And speaking of malware, running an active anti-virus utility can often make a huge difference in how fast a PC feels, particularly if you're using a hard drive; all those extra disk accesses to check for malicious files and such quickly add up. Switching to an SSD, again, provides a healthy improvement to the feel of any PC, even systems that go as far back as the Core 2 Duo days!
With the talk of Intel and Micron's XPoint Technology, which claims 1,000x increases in endurance and performance, along with a 10x increase in capacity, we're rapidly approaching the point where hard drives may finally stop showing up in most new systems. Similarly, Samsung's increasingly dense V-NAND chips (and other vendor's 3D NAND technology) promise better performance and data densities with reduced pricing. There will still be people that want/need several terabytes of storage (or more), and for archival/backup purposes HDDs are still great. But the days of the spinning disk are numbered. Considering the largest HDDs are currently sitting at 6GB, SSDs have already surpassed that limit and continue to grow—albeit with stratospheric pricing. Samsung, for example, showed off a 16TB 2.5-inch SSD at the recent Flash Memory Summit.
We're now at the point where a decent 250GB-class SSD can be had for well under $100. Sure, Crucial's BX100 isn't the fastest kid on the block, but it's better than the Trion and priced to move. For those that don't need more than 100–200GB of storage—and trust me, I've serviced a lot of PCs over the years where the HDD only had 50–100GB of data—such a drive should be a no-brainer. When we get to the point where a 500GB SSD costs around $80, only the pinchiest of penny-pinchers will continue to shun SSDs for their OS and applications. As far as we're concerned, that day can't come fast enough. Here's hoping that 2016 proves to be the year that we cross the tipping point and start seeing SSDs on any new PC priced above $400.
