SSD myths

SSDs, like most disruptive technologies, face some questions and resistance. People will grasp at any straw to avoid adopting them. Thanks to this resistance, a number of SSD myths arose. Here are the myths I see repeated over and over again, and the truth, based on my experience actually using the things.

Note: I originally wrote this way back in 2010. The drive technologies I speak of as state of the art are rather aged now. But the principles still hold today, and will continue to do so. Hard drives have gotten better, but SSD have gotten better at a more rapid pace.

1. Mechanical HDDs are more reliable than SSDs.

SSD myths
Even after all these years, there’s still a fair bit of misunderstanding about SSDs. Let’s bust some SSD myths.

Not really. The reliability of mechanical HDDs varies widely. Some do tend to be very reliable, while others are known for failing spectacularly at inopportune times. IBM’s 75GXP drive, for example, was so notoriously unreliable that it drove IBM out of the hard drive business. Prior to the 75GXP, IBM had a reputation for building very high quality drives, and in my personal experience, IBM drives failed very rarely. Seagate has (or had) a reputation for making bulletproof drives, but the last Seagate drive I bought died after a few months of very light use.

The reliability of mechanical hard drives varies so much that it’s very dangerous to make such flat statements.

SSDs are based on the same technology as a number of proven devices, such as USB thumb drives, digital camera memory cards, and even the firmware on PC motherboards. The underlying technology is well proven and highly reliable. I’ve been using it for at least 10 years.

2. You can only write to an SSD a limited number of times.

Now we’re getting somewhere. It is true that a memory cell has a finite number of writes before it turns into a pumpkin.

However, that finite number of writes keeps ticking upward as the technology advances, and SSDs incorporate a technique called wear leveling that pushes this limit off even further.

The idea is that if you have a drive that contains 10 blocks of memory, the drive ensures that any writes get distributed evenly between those 10 blocks, rather than allowing any single block to take the whole load. Since the majority of the blocks in any drive do tend to sit static, this is very effective.

Under real-world conditions, if you write continuously to an SSD, it will still last eleven years. This has been true since 2007. That’s longer than the expected shelf life of the drive, which is 10 years.

Mechanical hard drives have warranties that last 1-5 years. Some mechanical drives last 10 years. Others fail within a few hours, days or weeks.

If you use the drive under normal usage patterns, they’ll last centuries. So, thanks to wear leveling, the shelf life of the memory chips themselves is the limiting factor.

Considering that most performance-minded enthusiasts will replace their equipment every 2-3 years anyway, this argument just doesn’t hold anymore.

3. SSDs are bad at some things.

I love the phantom “some things.” My Honda Civic is bad at some things. But people buy them by the millions anyway, because those “some things” happen to be hauling and driving really fast. Since their strengths, like high gas mileage and reliability, outweigh their weaknesses, people buy them and generally are happy with them.

In the case of SSDs, the “some things” was one thing specifically: random writes. But this was a problem with entry-level drives in early 2009 and earlier, specifically, those using a variant of the JMicron JMF602 series of controller. With any current, reputable drive, this is no longer an issue.

Yes, random reads and writes are slower with SSDs than sequential reads and writes. But the same is true of conventional HDDs too.

The Western Digital Velociraptors are the very best, fastest SATA hard drives. They deliver a speed of 1.63 MB per second during random writes. That’s a far cry from their 300 MB per second peak speed, or their more realistic peak of 118 MB per second during sustained sequential reads.

Many SSDs deliver random write speeds of over 140 MB per second. See my Bestselling SSDs page for some solid suggestions.

Keep in mind this is what SSDs are bad at. At their worst, they’re still faster than the best conventional HDD at its best.

Even aging SSDs, such as the Intel X25 series and the very cheapest entry-level OCZ Agility or Onyx drives, will deliver random write performance that’s 3-10 times faster than a Velociraptor.

There may still be some JMicron-based drives and some other older technology floating around, but their capacities are so small you won’t want to buy one. The netbook I bought in 2009 had a JMF602-based drive in it. And yes, their performance under the worst of circumstances is crippling. By crippling, I mean PC/XT speeds for those of you who’ve been around a while. In modern terms, the system will just grind to a halt for a few seconds under heavy load. But defragmenting the drive, to take the randomness out of large writes, helps that.

You can safely avoid the problem entirely by buying a current drive manufactured by Intel, Samsung, or Crucial. There are other companies that make good drives too, but those three are safe choices.

3. I can just RAID some regular drives together and get SSD-like performance.

Yes, a RAID array of fast drives will deliver sequential throughput comparable to that of an SSD. But random reads and writes will still be slower, and the seek times will be much slower. The seek times of the very best HDDs are on the order of 3 ms on paper (in real-world tests it’s usually twice that). Meanwhile, the seek times on the best SSDs are on the order of .1 ms.

The seek time on an SSD is, worst case, 30 times faster. That’s comparable to the difference between a modern HDD and an HDD from a mid-1980s IBM PC/AT. You’ll notice it. If you ever actually used a PC/AT, you know what I mean.

4. You need SATA 3 to really take advantage of an SSD.

Not really. It took a while for SATA 3 interfaces to appear on SSDs and we bought them anyway. While some SSDs can saturate a SATA 2 bus during sustained sequential I/O, realistically, it’s not doing it all the time. During random reads and writes, which is what SSDs spend the majority of their time doing, even the best drives still stay below the limit of the SATA 2 bus.

For that matter, some of them dip into SATA 1 territory. Yes, you can hook an SSD up to a SATA 1 interface. I’ve done it.

Many SATA 1 chipsets have problems with non-Intel drives, because they don’t negotiate speeds properly. But even when you connect an Intel drive to a SATA 1 bus, you still benefit from the low seek times. And wouldn’t you rather have an SSD chugging along at a consistent 150 MB per second than a hard drive ranging from 5 to 118 MB per second depending on what it’s doing?

5. SSDs offer no real advantage in power consumption over conventional HDDs.

This stems from a widely discredited article published at Tom’s Hardware Guide in 2008. The methodology behind this article was poor, and the SSDs in the test are no longer in production anyway.

Modern SSDs consume anywhere from 0.4-0.75 watts when sitting idle, and the fastest drives consume up to 4.2 watts when in use. But many of them stay below 2 watts.

Conventional HDDs consume anywhere from 2-8 watts while idle, and 5-10 watts when in use.

At minimum, an SSD will save you 1 watt of power the whole time your computer is on. That’s not enough to pay for the SSD of course, but it’s also not zero.

6. SSDs are too expensive to justify.

In 2011 this was questionable, but the price difference is narrowing quickly.

Many of the same people who complain about 1 TB SSDs costing $250 think nothing of spending $100 extra to get an enthusiast-grade motherboard, or the top-of-the-line CPU, or for extra memory. Or spending $300 extra to get all three.

An SSD is a performance enhancement first, and a storage medium second.

Buy whatever size SSD you can afford and load your most critical stuff on that. That means your operating system and your most demanding applications. Then use a cheap conventional HDD to store your picture/MP3/movie collection. Those things don’t benefit from the added speed of an SSD all that much anyway.

Do that, and the stuff you care about most runs fast, while your piles and piles of data take advantage of cheap conventional storage.

Just because something isn’t inexpensive enough to store everything you own doesn’t mean you should discredit it. Back when I had a 52-megabyte hard drive, I had a drawer full of floppies that contained whatever data I didn’t need instant access to. When all I had was a 10-gig drive, I had a drawer full of recordable CDs.

An SSD plus a big, cheap hard drive is preferable to either of those options. It’s much faster, not to mention cheaper.

Any time the cost per gigabyte starts bothering me too much, I just have to think back to the $350 I paid for a 3.2 GB Quantum hard drive in the late 1990s, or the $200 I paid for an 850 MB Western Digital drive in the mid 1990s. And then I still had to shell out for a desk drawer full of floppies or CD-Rs to hold whatever I couldn’t fit on my HDD.

7. Windows XP doesn’t benefit from SSDs.

This doesn’t matter anymore since Windows XP is out of production, but even when it did matter, XP was able to benefit just fine from the fast seek times and fast throughput. Even without advanced TRIM support, the drive was faster than a conventional HDD. I used SSDs in the XP era and was happy with them.

8. SSDs don’t perform well with full-drive encryption.

There was a time when there was some truth to this, but modern encryption software is SSD-aware and encryption software is something you need to be keeping up to date anyway.

Under conditions such as these cited by Anthony Vance, they can still run about three times as fast as a good conventional HDD. And with some drives there’s virtually no difference.

And that’s it for the SSD myths. I hope this will help you make a more informed decision as you purchase new hardware.

One thought on “SSD myths

  • October 12, 2010 at 11:37 pm
    Permalink

    “Any time the cost per gigabyte starts bothering me too much, I just have to think back to the $350 I paid for a 3.2 GB Quantum hard drive in the late 1990s …”

    or the $500 my friends paid for their first 180 Kb single-sided floppy drive ca 1977 or 1978.

Comments are closed.

%d bloggers like this:
WordPress Appliance - Powered by TurnKey Linux