DDR 4 sets the pace for fast memory
Ramping up the Ram
Extreme Hardware Reducing the size of chips has benefits beyond those seen in CPUs and GPUs. One other key component to gain an advantage is memory.
The specification of the next generation of memory module, DDR 4 (DDR stands for 'double data rate'), is currently being finalised ahead of its anticipated launch later this year. It will use less power than today's DDR 3 and is based on a 32-36nm fabrication process, the numbers referring to the width of transistors used in the chip.
The first generation of DDR4 modules will run on a voltage of 1.2V, compared with DDR3's 1.5V requirement. That is expected to fall to 1.05V in later generations.
Samsung's first DDR 4 Dimm
In addition to lower power consumption, the other key advantages of DDR4 over its predecessor are higher clock frequencies and a wider range of transfer speeds.
While DDR3 was limited to transfer rates of 800-2133 million transfers per second (MT/s), the DDR 4 spec will allow for transfer rates of between 2133MT/s and 4266MT/s.
The topology, or the way components are connected to each other, of DDR4 chips is different too. While DDR 3 and its predecessors used a multi-channel setup to maximise data throughput, DDR 4 uses a point-to-point topology: each memory module has its own dedicated line to the memory controller.
One of the consequences of a point-to-point design is that every DIMM slot needs to be filled to maximise performance.
The first DDR 4 module was developed by Samsung, which announced its completion in January 2011. The company claims the new DDR 4 Dram module can achieve data transfer rates of 2.133Gb/s at 1.2V, compared with 1.35V and 1.5V DDR 3 Dram at an equivalent 30nm-class process technology, with speeds of up to 1.6Gbps.
When applied to a notebook, it reduces power consumption by 40 per cent compared with a 1.5V DDR 3 module, Samsung adds.
The potential gains for a huge data centre are vast
Samsung also said that its initial DDR 4 modules would be able to run at speeds up to 3.2Gb/s, compared with a maximum of 1.6Gb/s for DDR 3.
Reduced voltage is not the only way that DDR 4 reduces power consumption. It uses an improved version of the pseudo-open drain interface seen in graphics-specific DDR 3, which draws current only when it is connected to ground and thus uses relatively little power.
Exceeding the limits
There are two significant problems that need to be overcome in the design of DDR 4 systems, both caused by the point-to-point architecture.
The first involves the handling of large amounts of memory: all that extra data being processed per cycle makes handling high memory volumes particularly challenging. This is a problem for servers which may have to use switches to reduce the number of direct channels between Dimms and the memory controller.
Hynix does DDR4
The second problem is space. Point-to-point connections take up more room than multi-channel interfaces and so the memory chips need to be denser to fit in the same space. Samsung’s 30nm-class processers will alleviate that problem.
While speedy, low-power memory is important for desktop and laptop computers, it is crucial for servers. The potential gains for a huge data centre such as those run by Google, Facebook, Apple and Amazon are vast.
Mike Howard, an analyst with iSuppli, says: "Server platforms are the ones really screaming for this stuff because they need the bandwidth and the lower voltage to reduce their power consumption."
It is the potential for generating revenue in the enterprise market that is driving Samsung and its competitors. One of those competitors is Micron.
The company is behind both Samsung and Hynix in the market but announced in May that it was in talks to acquire bust Japanese memory maker Elpida. That deal, assuming it is completed successfully, would make Micron the second biggest manufacturer of computer memory in the world, behind Samsung.
Coming sooner... or later?
Micron has also partnered with Taiwanese memory manufacturer Nanya in a bid to compete with Korea's Samsung and Hynix. Micron’s first DDR4 module was shipped to key partners in May and the company says it expects to ship in volume by the beginning of Q4.
Like Samsung, Micron is offering a DDR 4 module built using a 30nm-class fab process and will initially achieve speeds of up to 2400MT/s before ramping up to 3200MT/s.
Micron predicts that the first DDR 4-based products will hit the shelves in 2013. But one hindrance to mass adoption is the CPU manufacturers: until they integrate DDR4 into their processors' memory controllers, we won’t see the full benefits.
Micron predicts that the first
DDR4-based products will hit the shelves in 2013
Intel’s current roadmap adds support for DDR 4 in its Xeon Haswell-EX processor, due only in 2014. Support for desktops and notebooks is likely to be further away than that.
Howard, however, is optimistic. "While Intel is only supporting DDR 4 on its server platforms in 2014, I have a feeling it's going to push it to its compute platforms as well in 2014," he says.
But his optimism may be misplaced. There are already rumours that Intel will stick with DDR 3 for Haswell, and rival chip maker AMD has yet to commit to DDR 4 at all.
It could be a while yet before we see DDR 4 in our notebooks and desktops. ®