AS IT becomes ever more affordable, DDR3 PC memory is diverging into two module mark(et)ing streams, just like DDR2 did two years ago.

On one side, you've got the usual "gigahurtz freak" push up to the maximum frequency without much care for the latency - or, for that matter, voltage.

On the other hand, there's the focus on minimising the latency for a given frequency. For instance, in the DDR2 world, a FSB1600 CPU on an Nvidia 680i chipset would have better Sandra bandwidth and latency results with dual-channel DDR2-800 CL3 memory kept in sync with the FSB, than with hotter dual-channel DDR2-1200 CL5 modules.

Product: OCZ ReaperX ES 2 x 2 GB DDR3-1333
Website: www.ocztechnology.com

The X38 and later chipsets from Intel changed the equation to focus more on using the spare bandwidth and depend a little less on latency, but even there the in-sync approach, up to now, gives the best results. And, if you're keeping the FSB low, say at 1333, then a dual-channel low-latency DDR3-1333 may be the choice for you.

Here's an interesting entry from OCZ Technology: their brand new ReaperX ES not only is the first in this series to support very low latency (CL 6-5-5 by default), but also incorporates a huge heat pipe assembly with aluminium fins standing on top of each module. Come to think of it, such contraption would be even more useful on their DDR3-2000 class EPP2 modules for the Nvidia Nforce 790i, as the voltages there may be a little higher still.

The modules aren't exactly light, and you should take care to press them at the module, not heat pipe, side when performing the mobo socket insertion. It also might be pretty hard, or impossible, to populate all four sockets with these modules, unless your board has a bit wider than usual DIMM socket spacing. Otherwise, the concept seems to work - the DIMMs weren't more than lukewarm to the touch even at 32 C or 90 F temperature at my home, and yes it's like that every day of the year in Singapore.

I ran the DIMMs on the equally new Intel D48BT2 "Bonetrail 2" mainboard based on the X48 chipset. While I couldn't manage to run DDR3-1333 CL5 yet (I didn't go above 1.8 volts here), I did get DDR3-1500 CL7-7-7 when running a QX6850 3 GHz quad core CPU with FSB 1500 setting. A new BIOS update from Intel for this board should have a bit more flexible memory O/C options. Even then, these are pretty good numbers for a start knowing these are 2 GB modules. We'll be doing more on this mobo as its BIOS improves - the goal is to get DDR3-1600 CL7 stable operation.

I also used the modules on the Asus Striker II Extreme Nforce 790i based board, and it had stability problems with the same DDR3-1500 setting. Again, a BIOS is being updated there, let's see if it solves the problems - Nforce does give a little higher Sandra bandwidth scores than the X48 provided all settings are equal, but the intermittent crashes and occasional refusals to boot are too high a price too pay for that bit of extra result ego boost.

In summary, a good approach. However, I'd like to see the modules - and, even more, myself - get a little thinner first. Fitting four of these effortlessly in the board sockets should be a goal. Also, think of this as very interesting memory for three-channel Bloomfield and Gainestown Nehalem platforms this autumn: DDR3-1333 may be enough frequency there due to 3 channel setup and integrated memory controller, but the low module latency would match well with the low CPU mem con latency.

Then, since two or four of these would provide a pretty decent flat top supporting surface, some hooks to fit a pair of small chipset-class fans on top of the heat pipes would make sense, just in case. And, as mentioned, use the sa me approach more aggressively on the upcoming XMP and EPP2 DDR3-2000++ module kits too.

The Good: heat pipes, low latency, high capacity
The Bad: more frequency options would be needed
The Ugly: trouble fitting all four in there if required.

Bartender's verdict: