The Inquirer-Home

More RAM for less power is forecast in 2010

Analysis DDR3 hits 1.2V at 1333 and 1.35V at 1600
Mon Nov 30 2009, 15:23

WHEN LOOKING AT PC system power consumption, the usual focus is on the CPU and GPU, followed by the chipset, disk drives and such. Memory isn't considered that important in the power equation until you start putting in three or more modules and start overclocking them.

However, memory should not be ignored as the memory subsystem on a modern PC can draw 20 to 50 Watts, depending on the capacity, speed and chip generation used. In fact, 12-DIMM 96GB DDR3-1333 Nehalem Xeon workstation or server ECC memory arrays will draw upwards of 200 Watts, with the corresponding heat generation as well. At least that isn't as bad as FB-DIMMs where, according to a 2008 Microsoft server power budget survey, 128GB of FB-DIMM DDR2 memory in older Xeon systems draw nearly 350 Watts of power.

Not to mention notebook PCs, where tightly confined dense DIMMs without fan airflow are often known to reach dangerous temperatures in excess of 70C during operation. That hasn't improved with the appearance of DDR3 either, at least not yet, though battery life in notebooks will of course gain a little from the use of low voltage memory.

So, even though default memory module voltages dropped over time from 2.5V for DDR1 to 1.8V for DDR2 and 1.5V for DDR3, the current 'green' energy saving push, combined with the need for cool silent fan-free operation, has accelerated the migration to low voltage memory somewhat. Even server DDR2 memory has had 1.5V parts for two years now, with even lower voltage levels coming through the most recent 1.2V LPDDR2 specification. But the real focus is on the current cream of the crop memory standard, DDR3.

As mentioned, even at the default 1.5V and 1333 speed, DDR3 can turn quite hot in the large arrays of registered ECC modules used in workstations and servers. At the same time, DRAM process technology, with 40nm dies coming out now from firms like Hynix and Samsung, has enabled that performance to be reached with far lower voltages - 1.35V, 1.3V and even 1.2V now. When you consider that, everything else being the same, a 1.5V module will consume over half more power than a 1.2V module, with the additional heat generation thrown in as well, the benefits are obvious.

piMany vendors are offering low-voltage DIMM kits now, for markets as varied as notebooks, Intel P55 home theatre silent PCs, all the way up to high-end servers. Geil in Taipei was among the first to have 1.3V certified, 1.2V capable DDR3-1333 modules, which we tested several months ago. G.Skill, also in Taipei, has come up with dual-channel and three-channel kits supporting 1.35V at up to DDR3-1600 CL7-8-7 speed, the first on desktop platforms. Many new Intel-based P55 and X58 mainboards do support lower voltages in BIOS as well.

The big vendors, like Hynix or Samsung - whose DRAM production dwarfs all of the Taiwan DRAM vendors combined - have taken the low voltage push seriously, as you can see from their slides at the Denali MEMCON conference this past June:

samsunglpddr2

The cumulative savings from higher densities, better processes and lower voltages do help a lot, as you can see:

samsunglpddr3savings

And, according to Denali Systems, a memory design expert company, the total power consumption per chip for different DRAM technologies can differ a lot. Multiply it by the hundreds of chips often seen in high end systems - compared to a maximum of 96 chips in a six DIMM three channel desktop memory setup on a Core i7 - and you can get an idea how much power saving is at stake.

denalimemusage

What you can also notice is that the 1.2V capable DDR3 RAM, expected only a year from now, is creeping in right now. More mainboards are supporting low DDR3 voltages, too.

Keep in mind that the similar lower voltage push will happen in the graphics memory field as well, where GDDR5 can proceed to evolve both at the speed front towards 6 Gbps and towards lower power versions for single slot cards as well as in mobile GPU applications.

In summary, as most PCs will soon ship with 4GB or more memory, and most workstations and servers with upwards of 12GB, memory power consumption and cool operation will become more important. It will be acheved not just by using newer smaller process geometries, but also denser designs and, of course, lower voltage levels.

Coupled with these improved processes and densities, low-voltage DDR3 memory will help save a lot of power while still delivering up to DDR3-1600 performance, helping to improve battery life and heat dissipation across all platform types. As for the higher speeds, well the usual 1.5V and 1.65V modules will still be there for the 'enthusiast' performance realms above DDR3-2000, until DDR4, running at 1.2V by default, starts appearing sometime in 2011. µ

Share this:

Comments
Racetrack memory

The future of RAM.

http://www-03.ibm.com/press/us/en/pressrelease/23859.wss

posted by : Mike, 01 December 2009 Complain about this comment
ha!

its called stringing the piggies along for years before anything remotely similar to whats being promised. screw gddr3 im waiting for 5 or 6 ^ ^.gddr 3 prices have fallen too quickly wink* wink*

posted by : super dude, 01 December 2009 Complain about this comment
Piffle.... mere baby steps

So where is all this great, non-volatile ultrafast memory we've been promised for years? Seems to me that would cut the power budget a bit if done right.

posted by : rich wargo, 30 November 2009 Complain about this comment
@ "..don't quote me"

Jason Goatcher said "Of course, that's assuming we get something called DDR5 at all. I've heard ruminations that we may move on to something else entirely. Haven't really researched it though, so don't quote me".

Seriously, that was one of the most unintentionally funny posts I've read in quite some time. Actual out loud laughing. Cheers.

posted by : satis-factual, 30 November 2009 Complain about this comment
gddr != ddr

@memory hungry. GDDR isn't the same as DDR. The problems with GDDR4 won't affect DDR4 at all.

posted by : bw, 30 November 2009 Complain about this comment
have to agree

i'd never considered memory to be any factor in system power usage, until recently. 4 sticks of ddr3 that i installed on my am3 mb were getting just as hot as my video card. had to shell some $ for a dedicated memory module fan. chips are not overclocked beyond manufacturer specks (i think that i actually run them lower but with tight timings)

posted by : joed, 30 November 2009 Complain about this comment
Learn the lingo

The first memory went at a certain speed(I forget what) DDR(Double data rate) goes at twice that speed. DDR2, doubled again, DDR3, doubled a third time. 2^3 or 8. Are you beginning to see a pattern here? DDR5 is, wait for it...32 times as fast as the stuff we started with.

Of course, that's assuming we get something called DDR5 at all. I've heard ruminations that we may move on to something else entirely. Haven't really researched it though, so don't quote me.

posted by : Jason Goatcher, 30 November 2009 Complain about this comment
Cut Power by Doubling Bit Width....

Another Drashek WaterMellon Pie? NO.

Today Elphedria Samples Hit for NEW DDR2 with 32 bit. twice old & less power,now. although at DDR2 voltages. Past was 16 bit & used assemble/disassemble that took two lines in & Out for 16bitx2. Now whole shebangun does it all in one simple 32bit step.

sample shipments of a 2-gigabit x32-bit I/O DDR2 SDRAM that operates up to a high speed of 1066 Mbps. The new memory device is capable of meeting the need for 1.8V x32-bit

posted by : Memorite, 30 November 2009 Complain about this comment
Why DDR4?

Why not just skip DDR4 directly to DDR5? Everyone knows that DDR4 didn't show any improvement over DDR3, it was shortly available in graphic cards market before it was dumped completely (just in some high end ATI cards).
DDR2 was not necessarily as well, but at least it brought a significant frequency boost.

posted by : Memory Hungry, 30 November 2009 Complain about this comment
aboutus
Advertisement
Subscribe to INQ newsletters
Advertisement
INQ Poll

Authorities in several countries raided Megaupload recently, shut down all of its services, seized hundreds of servers and arrested several of its executives on criminal charges.

Do you think the move was justified?