The state of DDR4
Do almost anything often enough, and the individual steps become routine. It certainly feels that way when I’m building computers: There are days I feel like I could do it in my sleep (and, truth be told, there are days I probably have). But when you’re selecting parts, most of the components require at least a little analysis. The CPU and the motherboard are major, and have to be considered together. Of course storage is a big deal — you always want enough and you always want it to be fast, and optimizing those qualities with affordability can be a headache (especially at the moment). For a gaming system, the right video card is essential. And in addition to being your system’s face to the world, the case will affect both the way you build the computer and how much you can expand it later. And… that’s it. Right? Am I forgetting anything?
Oh yeah, RAM. Sorry about that. But I admit it: When I’m shopping for memory, I pretty much turn off my brain. It’s not intentional, and it’s not that I think RAM isn’t important. It’s just that it’s the easiest kind of hardware to shrug off. Most of us, even when we’re in our most rabid enthusiast mode, roll our eyes at the arcane industry of RAM timings. Provided the DIMMs’ voltages and speeds are compatible with the motherboard, everything else usually seems academic. Oh, if I’m being super picky about components, I might pay a bit more attention, but in reality that’s rarely necessary. These days, you can take it for granted that basic DDR3 is going to be what you want.
It occurred to me last week, however, that we’re rapidly approaching another time when that might not be the case. As processors and other hardware get more powerful, it becomes more difficult — and important — for RAM to keep up. Over the last 20 years or so we’ve seen single data rate (SDR) memory give way to double data rate (DDR) memory, which transfers data on both the rising and falling edges of the clock signal. And over the last decade, we’ve slowly progressed from DDR to DDR2 (with an internal clock running at half DDR’s rate) to DDR3 (operating at half DDR2’s rate), constantly upping performance while reducing power usage along the way. But what about the next logical step in this particular cycle: DDR4?
It’s coming, though when — and whether — you’ll see it for use in your home computer is still somewhat up in the air.
Samsung announced a year ago that it had developed the first 2GB DDR4 modules using a 30nm process technology, achieving with them transfer rates of 2.133Gbps at 1.2 volts. Three months later, Hynix came up with its own DDR4-2400 modules operating at the same voltage.
The website DDR4.org, which is (not surprisingly) devoted to information about the upcoming memory standard, is a bit light on substantive information but fleshes out a few of the details. Among them: Data transfer rates would start at 2,133MT/s (about where DDR3 is leaving off) and could eventually double to 4,266MT/s, and initial energy consumption is expected to be about 1.2 volts (though later chips might be able to use as little as 1.05 volts). The site goes on to say that “other sources have noted that the [power] consumption is 40 percent less for DDR4 than for an equivalent DDR3 chip.” Not bad. Another nugget of useful information on the site is the mention that DDR4 might make use of “pseudo open drain” technology, which was adapted from GDDR memory.
More interesting still is that you’ll be able to finally forget about dual-, triple-, and quad-channel memory configurations. DDR4 will use a point-to-point topology, so there will be only one DIMM per channel. As Bit-tech pointed out, this is a continuation of the simplification process that memory has undergone ever since the move to DDR. As what happened when the PCI bus became PCI Express, “DDR4 will become a point-to-point bus and the parallelism is being left with the memory controller itself with multiple memory channels,” the site explains.
I contacted Corsair, a company that knows just a little about memory, to find out what all this is likely to mean for system building. Robert Pearce, the company’s senior technical marketing specialist, gave me some good information.
The first DDR4 chips, he told me, will be coming out at the end of 2012 — but “the initial target platforms are servers, not desktops.” In fact, Pearce even acknowledged the rumors that DDR4 might never be seen on desktops at all, but went on to point out that “this is really something that will be dictated by Intel and AMD since they are the driving force behind the memory used in client computing platforms (based on the fact that the memory controller is integrated into the CPU).”
X-bit labs estimates that, if DDR4 does show up on desktops, it will probably be in 2014 and production ramp-up will be rapid, with DDR4 instantly garnering about 12% market share and increase to 56% in 2015. Given how long DDR3 took to catch on, I’m a bit skeptical about that, but it’s far too early to say for sure.
Something else that’s difficult to predict with any accuracy is what performance improvements users can expect compared with DDR3 — and on that matter Pearce is rather conservative. “[They] are not likely to be very significant,” he told me. “Over time, the DDR4 signaling structure will allow for higher bus frequencies, but the overall access times (clock period x latency) are not changing dramatically. This means that while there is higher peak bandwidth (which can provide performance improvements in a server environment where multiple accesses from completely independent threads can be queued up), the time needed to get data for a given request will not really change.”
Sure, what will happen with DDR4 remains to be seen, but put me in the “optimistic” column. I recall people saying a lot of similar things about DDR3, too, a few years ago, and that eventually earned wide-scale adoption. Even if DDR4′s changes on most fronts are minor, I think the industry will come around. Does it have any other choice?