THE COMBINATION of early low power die availability and eager Taiwanese memory module and mainboard manufacturers means we're looking at 1.3V DDR3 now instead of in 2010.
When we've looked at manufacturers' efforts to make their PCs greener, that is, with less power-hungry memory, we were predicting 1.35V DDR3 in 2010, 1.2V DDR3 in 2011 and 1.2V DDR4 a year later.
But here we already have a pair of these babies from Geil, the Taiwanese high-end memory brand that showed the modules running at Computex seven weeks ago.
Most likely the world's first actual sale kits if the kind, the two modules are even packed in a recycled cardboard material not unlike the packaging used for supermarket eggs.
The DIMMs, intended to be paired up with the upcoming P55-based Lynnfield mainboards, can actually run at as low as 1.2V on their advertised DDR3-1333 CL9 speed, even though the official designation is 1.3V. There are no heat spreaders, and no wonder - there was literally no heat emanating from these modules during the tests.
For the low voltage DDR3 to work in your system, besides the memory module using the suitable dies, you also need a mainboard supporting the given voltage electrically and in the BIOS. Since there are no official P55-based boards launched, I used the one X58 based Core i7 board that does - the Gigabyte EX58 Extreme with Intel Core i7 975XE processor. This ultra-high-end mainboard with huge trademark heatsink fins for the chipset offers the 1.3V memory option in the BIOS.
We ran the system with the usual Windows Vista64 setup using Sandra 2009 SP3 benchmark to check the speed impact, versus the usual GEIL 1333 CL9 1.5V DIMMs. Both were tested as dual channel only, to simulate the upcoming P55 benchmark more closely.
When using 'default' both DIMMs went to CL 9-9-9-24 at the DDR3-1333. Nothing special except for the voltage and, in fact, identical memory bandwidth benchmark results in both cases. Unfortunately, Sandra's power usage estimator couldn't differentiate between the 1.5V and 1.3V DIMM power consumption, but we guess that can be fixed.
We didn't bother with overclocking, since obviously the target here is to save power at the default clock. However, we did try some latency adjustment. While we could easily get CL 7-7-7-18 on the 1.5V modules, the 1.3V modules did CL 8-8-8-20 at their best.
Still, that's not bad, given the estimated one-third (or thereabouts) power saving by going to 1.3V.
In Sandra, reducing the extra cycle of the three key latency points, when all else is the same, will give around five per cent performance jump on the synthetic bandwidth benchmark.
Is that a fair trade? Thirty per cent lower power consumption for five per cent lower tuned memory bandwidth benchmark? But remember, the 30 per cent lower memory power consumption may only result in two per cent lower total system consumption.
But then, a five per cent synthetic benchmark delta may only mean a one per cent or even less real application performance hit.
So if you want an absolutely low-heat, low-power setup for a totally silent fanless home theatre P55 PC later this year, then this memory is for you. µ
Lowest voltage modules available, yet still at decent speed.
There's never enough speed.
More board BIOS support needed while we avait the P55.