Size matters. And when it comes to SSDs, size costs too. At least, it did. Back in 2008, Intel launched its first solid state drive aimed at mere punters rather than enterprise customers. You forked out £400 and you received 80GB in return. Yeah, just 80GB.
Okay, conventional hard disks were also pricier per gigabyte in 2008, but it was still roughly a case of swapping the 'GB' and '£' symbols around. Less than £80 bought you more than 400GB of spinning magnetic platters even then.
So size has definitely been a source of insecurity for SSDs. Sure, you can combine a tiny SSD with a fat old magnetic drive and theoretically have the best of both worlds: speed and performance. Even the tiniest solid state drive will soak up a Windows installation. Simply chuck the rest in the big old data bin that is a cheap conventional hard disk.
The problem with that theory is applications. They like storage performance too - especially games - and a decent games library needs some fairly serious space. For us, that means an absolute minimum of 250GB, and ideally closer to the 500GB mark.
The good news is that's now becoming much more realistic. SSDs in the 500GB region can now be had for under £250. Okay, you'll have to pay more to get the latest and fastest models, but we're no longer talking about silly money for truly usable solid state storage.
If the size problem is nearly solved, there's plenty more keeping solid state storage interesting. While CPU and GPU development seems to be tailing off, SSDs are still a fairly early phase technology. Part of their development path involves things like controller chipsets and the flash memory chips themselves. That's good old Moore's law in action regarding the latter, and it's probably going to have more impact on SSD prices and performance than it will on processors or graphics chips in the near future.
Then there's the question of storage interfaces. SSDs have been developing much faster than the interfaces that hook them up to PCs. Already, SSDs have too much game for SATA 6Gbps. An alternative is needed, but what?
First things first. We need to deal with the elephant in our group test. Or rather, the elephant that's not in our group test, but which you might have expected to amble along and make an appearance. Whatever. We've got a problematic pachyderm, and its name is PCI Express.
According to some, it's the next big thing in solid state storage and that's because we've hit the wall with regard to some aspects of the SATA storage interface. Already, pretty much all high-performance SATA SSDs are limited by the 6Gbps maximum throughput of the current top-end SATA interface. That's why drives tend to bunch up at a little over 550MB/s for sequential reads and writes. That's all a SATA 6Gbps connection can manage.
Enter PCI Express. The latest 3.0 spec delivers just under a GB/s of bandwidth per lane, and even the old PCI Express 2.0 does nearly 500MB/s per lane. Of course, you can string together as many as 16 lanes and end up with some pretty epic theoretical bandwidth, but it's really only graphics cards that go for the full 16-lane monty. For storage, we're typically talking two to four lanes. Whether you're talking PCI Express 2.0 or 3.0, that's enough to give SATA a good, firm spanking.
A handy recent example is the latest MacBook Air from Apple. Okay, it's not exactly a PC, but the hardware is all PC-derived and it's a nice example of just what can be achieved thanks to PCI Express, even in an ultraportable form factor. For the record, the MacBook uses PCI Express in 2.0 trim with two links available to its SSD, and therefore has a peak theoretical throughput of around 1GB/s.
In early tests, the actual throughput is about 750MB/s for reads and writes, and is very likely limited by the drive, not the interface. Anyway, that's serious bandwidth by any measure. It's getting on for 50 per cent quicker than anything you can by in SATA format, and we're talking about one of the slimmest ultraportable notebooks you can buy. Incredible.
But it's only part of the story. We haven't tested a MacBook, but it will almost definitely fall foul of the same limitations as any other PCI Express SSD. And that's random access. Currently, even the very fastest SATA SSDs fail to get into three figures when it comes to MB/s for random access. That's nowhere near the limit for SATA 6Gbps.
The same goes for the PCI Express alternative. It's not the interface putting a cap on random access performance, it's the drive itself. And here's the thing: when it comes to the subjective feel of your PC, random drive performance is at least as important as sequential throughput. Maybe more so.
In fact, random access performance is also the metric by which SSDs really separate themselves from the fastest conventional hard drives. An SSD might be two or three times faster for sequential work loads, but it might be 20, even 50 times faster for random access tasks. And that's why I've just blown the last 500 words banging on about it.
PCI Express, or the posited SATA Express hybrid of the two interfaces, is very probably the future of storage, but it's likely not the panacea you might have thought. It's important to understand that.
What else is new in solid state, then? Capacity, that's what. Ever since SSDs first came on the scene, perhaps the biggest single drawback has been lack of capacity. At first we tried to kid ourselves that you could use an SSD as small as 40GB as a boot drive and have a conventional magnetic drive for the mass storage duties, and that does kind of work, but it's not ideal.
No, 'ideal' would be an SSD big enough for everything that's performance critical, including a mahoosive games library. That thinking has lead to some pretty spectacular mental gymnastics.
First, we pretended 128GB might just be workable. Then 250GB-odd became the realistic option. For some people, that might just be workable, but once you hit around 500GB or more, then you are really gaming with gas. There will be some for whom even that isn't enough, but for us, it's the point at which the storage limitation drawback truly fades into the background.
The good news is that prices for such drives are now approaching attainability for mere mortals. As I write these very words, the good people at www.scan.co.uk will do you a 480GB OCZ Agility drive based on the SandForce SF-2281 controller for just £200. That's not exactly chump change - and things have moved on from the days when the SandForce SF-2281 was the daddy of all SSD controllers - but it's not crazy money either, and it would be a pretty darn nice drive to use in the real world.
So we come to SSD controllers in general - ever the hot topic for solid state storage. Things have slowed down a bit lately. It's probably the quiet before the gathering storm that will eventually be SATA Express.
In any case, there hasn't really been a flurry of hot new controller chipsets quite yet. One of the newer offerings is the LAMD LM87800, as seen in Corsair's Neutron GTX drives. Not all that much is known about it, other than it's a dual-core ARM chip. Other than that, it's largely the usual suspects.
Then there's the Barefoot 3 from OCZ-owned Indilinx, which pairs an ARM core with OCZ's somewhat mysterious Aragon co-processor. Apart from that, SandForce's once world-beating SF-2281 soldiers on, while Marvell's 9174 and Samsung's MDX remain pretty competitive.
As for other technical advances, well, you can read about the details in our solid state drive reviews, but the usual progress applies, including the ever-shrinking size of NAND memory cells.
That said, one thing that hasn't really taken off is TLC, or triple-level memory cells. It looks like it will be a few years yet before that hits the mainstream. If that's disappointing, remember that memory prices have still been dropping in the meantime. The widespread adoption of TLC and the increased memory density it delivers will only make big SSDs even cheaper.
Faster, bigger, cheaper: What's next for SSDs?
Now we know big SSDs are getting cheaper, what else can we expect to see in the near future? As we've discussed, PCI Express is going to help lift the lid on peak sequential performance, and Apple's MacBook Air offers around 750MB/s for both reads and writes, but that's just the beginning.
ADATA has announced the SX2000, and the numbers it delivers are truly spectacular. For starters, it weighs in at an enormous 1.6TB. Then there are read and write speeds of 1.8GB/s. And here's the clincher: ADATA is claiming it's also good for 200,000IOPS. So random performance should be mighty impressive, too.
For the record, that's made possible by the new SSF-8639 interface, which is essentially a four-lane PCI Express solution. It won't be cheap, though; we expect pricing somewhere north of £1,500. It's really an enterprise class drive, but it does hint at the future of drives for desktop PCs.
Another interesting development is Thunderbolt - Intel's superfast general purpose connection. The second generation of Thunderbolt is good for 20Gbps, which puts it in similar territory to a four-lane PCI Express interface. Intel has recently been showing off a 128GB Thunderbolt 2.0 thumb drive prototype. The idea of a thumb drive that outperforms today's SSDs is intriguing.
As for the development of the memory chips themselves, things keep getting smaller. Toshiba has just announced its latest 64Gb NAND chips, claimed to be the smallest yet. And as these chips get smaller, they should hopefully get cheaper too.
Even better, Toshiba says it's tooling up to produce triple-level-cell (TLC) memory on the same process, later this year. So far, only Samsung has sold TLC memory into the mainstream. Toshiba says its TLC chips are initially targeted at smartphones and tablets, where memory density is super-critical. But the new chips are also destined for PC SSDs. 500GB drives for well under £200 next year? That looks very likely indeed.