Life At 2 GT/s: 6 GB DDR3-2000 Memory Kits Compared

Test Settings

Swipe to scroll horizontally
Test System Configuration
CPUIntel Core i7 920 (Bloomfield)-2.66 GHz, 8.0 MB Cache
CPU CoolerSwiftech Apogee GTZ Liquid Cooling
MotherboardDFI LANPartyUT X58-T3eH8, BIOS 217 (02/17/2009) Intel X58/ICH10R Chipset, LGA-1366
GraphicsZotac GeForce GTX260² 896MB 576/999MHz GPU/Shader, GDDR3-2484
Hard DrivesWestern Digital VelociRaptor WD30000HLFS 0.3 TB, 10,000 RPM, 16 MB Cache
SoundIntegrated HD Audio
NetworkIntegrated Gigabit Networking
PowerCoolermaster RS850-EMBA 850W, ATX12V v2.2, EPS12V
OpticalLite-On LH-20A1L, 20X DVD±R
Software
Operating SystemMicrosoft Windows Vista Ultimate x64 SP1
GraphicsNVidia Forceware 182.08
ChipsetIntel INF 9.1.0.1007

The DFI LANParty UT X58-T3eH8 was chosen for today’s test platform by comparing test results from our own reviews to those of various memory manufacturers.

Winner of our previous High-End DDR3 comparison, Kingston sent a newer set of its KHX16000D3ULT1K3/6GX modules to represent the latest memory ICs. The new modules have larger identification labels but otherwise appear identical to the originals seen below.

The retail-boxed Dominator GT DDR3-2000 CAS 8 memory kit includes a fan, but we also wanted to try out Corsair's new thermoelectric cooling device. Though Corsair has its own external liquid cooler capable of handling this small thermal load, we stepped up to a custom component-cooling configuration from Swiftech, which will hopefully improve overall performance even further.

Swiftech’s MCP350 pump provides excellent coolant flow for 3/8” flexible tubing. The company’s MCR220-QP radiator provides around twice the surface area compared to typical external liquid coolers, while its integrated reservoir reduces clutter. Two MCB-120 Radbox Revision 2 spacers provided space to mount the water pump and wall-hang the assembly, while maintaining unrestricted airflow to and from the top fan.

Swipe to scroll horizontally
Benchmark Configuration
Stability TestMemtest86+ v1.70, single pass (~45 minutes) Max Speed at CAS 9; Min Latency at DDR3-2000, 1866, 1600
Bandwidth TestSiSoftware Sandra XII SP2 Version 2008.5.14.24 Memory Test = Bandwidth Benchmark
Thomas Soderstrom
Thomas Soderstrom is a Senior Staff Editor at Tom's Hardware US. He tests and reviews cases, cooling, memory and motherboards.
  • philologos
    Can I make a suggestion? I would like to see a trial, in which the i7's integrated memory controller is overvolted incrementally to death.

    Much has been made since the inception of Nehalem about the offical "safe" vDimm limit of 1.65v. There have been no lack of intrepid overclockers who've pushed this spec to 1.7, even 1.75 volts. I want to see the actual limitation demonstrated in spectacular fashion with pictures. Also, I highly suspect memory voltage on the X58 platform is capable of increasing the CPU heat output; it would be satisfying to see confirmation of this. Give this idea some thought.
    Reply
  • dragonsprayer
    Hey guys, THG is the best! People you need free tech support THG is #1! The article is nice, I wish we had a big comparison to really show - 3 cpu's, different rans....well anyways!!! Good info as always, fact is you do get significant ranges of mem bandwidth at the same settings. You can take different types of ram , set them at the same settings all 100% stable and yes huge swings in membandidth.

    Yes, in in many cases high end ram give slight increases, those ranges of 10-15% are common for bandwidth. Yet the i7 system is so advanced and so far a head of software, the software rarely knows the difference or the results i should say.

    WSZ says: Skip 2000 ram, go for 1600 ram, a 920 at 1500-1700 fsb is untouchable for the pricer to performance provided. THG has show muskin 7-7-7 or slightly higher latecy ram to really blow everyone away.

    We live in the best times, for the advancement of computer parts, we an not shirink for ever!!! We are at 32nm? Then what 3 atoms thickness?

    Yes, AMD really has a good system now with the south bridge 750 oc, but nobody cares - i mean my customers. 7-8-7-20 at 1.65v at 1550 to 1650 is the sweet spot if you can squeeze it out - you may just go with 8-8-8-24 with 6GB of $100 ram. Sweet spot on the 20 multipier is 3.8ghz air cooled. We getting 4.4ghz with water - i toss a photo in for fun - this is mid level system with minmual rads but runs 70-75c at 4.3ghz all day with quad sli: http://s329.photobucket.com/albums/l373/warpedsystems/?action=view&current=WSZ-WCcm-T31.jpg
    Reply
  • dragonsprayer
    philologos -- does not help at all, in fact it just makes the system unstable, skip the stroy why we set thte QPI to 1.65v instead of 1.33v as was to be done. The reuslts:
    system was find, after 3 weeks starting have issues - crashes
    bottom line is do not to it - 1.33-1.35v is fine for most QPI
    Reply
  • doomtomb
    1600MHz DDR3 still provides the best price/performance. Not just that, but there is about 0.5% real world performance gain from higher clocked memory.
    Reply
  • dragonsprayer
    exactly doomtomb
    in fact this has not change for 3 years

    around 1600-1700 is the sweet spot, in highend and exact system, make just one lateny setting lower - we are build 3 systems. one, is a $4000 air cooled system, one is a $5000 water cooled and one is a $6000+ water - these are not gaming sytems.

    Getting the fsb is right is the most important thing in building a system, while i can it fsb in a i7 - it still the memory bus evern if its not fsb or front side bus

    2-2-2-5 ram for ddr400 ddr
    3-3-4-8 ram in ddr800 ddr2
    7-7-7-20 with ddr3
    Reply
  • Crashman
    philologosCan I make a suggestion? I would like to see a trial, in which the i7's integrated memory controller is overvolted incrementally to death.Much has been made since the inception of Nehalem about the offical "safe" vDimm limit of 1.65v. There have been no lack of intrepid overclockers who've pushed this spec to 1.7, even 1.75 volts. I want to see the actual limitation demonstrated in spectacular fashion with pictures. Also, I highly suspect memory voltage on the X58 platform is capable of increasing the CPU heat output; it would be satisfying to see confirmation of this. Give this idea some thought.
    Unfortunately that's impossible. They don't die instantly, they die over a matter of weeks. Intel published a bunch of data on how long the memory controller last at various voltage levels, we're talking several weeks at 1.70v and at least a few days at 1.75v.

    So, when the memory controller finally did die, nobody woul know whether the voltage or time was the most significant factor.
    Reply
  • Kill@dor
    Awesome review Thomas. Thank you for making a great comparison... I still think Kingston's DDR3-2000 is the best value over all. Pushing the CPU past 1.7v core is not an option in my opinion, and anyone with good clocking skills can push a core i7 920+ to 4.0GHz @ 1.25v with this kind of RAM speed and bandwidth. I'm running an (1066 strap)E6600 @ 3.6GHz, 2:8 ratio and 20MB/s dual-channel throughput on DDR2-1200, the 8800GT Superclocked Version and it runs crysis like silk on 2ms 1920x1080p. Super Talent has shown better performance and value in this line and i'm glad to see it as well. This is the kind of performance i have been waiting for and its a great confirmation of what i have been suspecting about these RAM's. The only reason why i haven't changed the graphics is because Nvidia/ATI have not had a significant jump in hardware besides the 4870x2 and GTX295 (too expensive/too much power needed), but i know the good stuff is coming Q4. Time to build another 5yr build with a quad this time ^_^
    Reply
  • sublifer
    Great review. At the pricing chart I was eyeing the Kingston with the low price and decent rated timings. I was surprised to see the cheapest kit, the Super Talent, to match up with the best in timings, that'd definitely be a great buy.
    Reply
  • dragonsprayer
    Ya, check that supertalent review on newegg - oops thats me! we used that in many systems!
    if you want to pot for the extra bandwidth get muskin

    corasir gave us 10%+ more bandwidth at the same settings a supertalent - then again it was double the price
    Reply
  • avatar_raq
    Nice article..But I thought it'll include real time programs and games benchmarks. Honestly, we all know the difference netween them is so small, that's negligible.
    Reply