bit-tech.net

4GB DDR3 Memory Roundup - Part 1

Introduction

For the most part we've avoided DDR3 reviews to date, favouring DDR2 until now simply because of the price of 4GB kits. Why would you want 2GB of extremely fast memory when 4GB of DDR2 for a fraction of the price will stop you from hitting the wall or a slow hard drive pagefile that much sooner?

With the inevitable arrival of Intel's Core i7 and the AMD socket AM3 CPUs in the near future, both of which will be DDR3-based, we decided to take a look at some 4GB kits to get you off the ground - if you were to buy now, they'd still be an investment for future upgrades.

We intend this to be a first part of, well, a few. You see we started with just three kits then as things happen an article grows and more and more of our memory partners wanted to get involved. At the time of writing the current count is seven including already tested, en-route or interested. It's important to note that we're not just concentrating on performance and simply comparing apples to apples - this isn't a test of who is fastest = better. We are covering all bases: latency differences, price and specifically its relation to performance, aesthetics and cooling, warranty and support - everything that matters.

With the huge amount of testing involved we've attacked the first couple of kits that have come our way: From Corsair we have a 4GB (2x2GB) kit with DHX heatspreaders, rated to 1,600MHz (PC3-12800) at a typical latency set of 9-9-9-24 at 1.8V. From G.Skill we have two kits: the first is a pre-release 4GB (2x2GB) kit with Thermaltake heatspreaders and heatpipes, rated to 1,800MHz (PC3-14400) at a typical latency set of just 8-8-8-21-1T at 1.9V. The second is already on the shop shelves - it's a 4GB (2x2GB) kit with G.Skill Pi-heatspreaders that's rated to 1,600MHz (PC3-12800) at a typical latency set of 7-7-7-21 at 1.8-1.9V.

4GB DDR3 Memory Roundup - Part 1 Introduction to bit-tech DDR3 Memory Testing 4GB DDR3 Memory Roundup - Part 1 Introduction to bit-tech DDR3 Memory Testing
Ignore the bubbles - we'd just filled it up! Click to enlarge

We've tested on two of the most popular (and only) performance chipsets out there right now - the nForce 790i Ultra SLI and Intel X48. EVGA supplied us with its quite frankly awesome looking fully watercooled BlackPearl motherboard that uses specially designed Innovatek blocks on the MOSFETs, northbridge, southbridge and CPU to keep it cool, and we paired this with a repaired kit from Aqua-PCs in the UK.

Back when we tested the OCZ Flex II PC2-9200 4GB kit that could be watercooled, we found the Aqua-PC kit sub-par, however after discussing it with Aqua-PC who were very understanding despite our negative review, the pump and reservoir were immediately replaced and we have to say, for the price it now works fantastically to keep even a whole motherboard and CPU cool. Aqua-PCs was even kind enough to ship us the correct 10mm barbs within a day - we can't rate the service and the guys behind it highly enough.

4GB DDR3 Memory Roundup - Part 1 Introduction to bit-tech DDR3 Memory Testing

Along with the nForce board we've also got an Asus Rampage Extreme built on Intel's X48 chipset - this board is the beans when it comes to features and potential performance, and if we thought the Maximus II Formula had everything the Rampage Extreme takes it to a whole mindbogglingly new level. While not watercooled, it's got more heatsinks than you can shake an aluminium factory at and a BIOS that required a few Ibuprofen and some tears when I realised I'd have to donate a large chunk of time to the blind art of clock skews.

To cover as many scenarios as possible, we've taken into consideration performance at the lowest achievable latencies at 1,600MHz, stock speed performance and highest overclock at a locked 9-9-9-24-2T to affix some uniform boundaries to our testing procedure.

Finally, in order to quell the calls of "well how does it compare to DDR2?" we've taken the same hardware we use in every recent motherboard article and used it here too - this way the results should be as directly comparable as possible.