Memory Kingston HyperX DDR2 750 Mhz и 675 Mhz
Progress is inexorable. On change to one technologies come others, so was replaced such operative memory as DDR SDRAM by memory DDR2 (for more detail information please look, for example, the review "DDRII SDRAM - the new standard "fast "memory"). And though DDR2 it is not so widely widespread yet as previous one, but there is no any doubt that will grow up steadily it’s popularity – that’s because of the prices on DDR2 have already decreased to comprehensible level and many recent chipsets of Intel corporation dominating in the market of DDR2 platforms (for example, chipset NVIDIA nForce4 SLI Intel Edition), do not support the operative memory DDR any more. Thus the time has come to look at this operative memory in details.
CAS and frequency - the basic competitors
As usual, it is necessary to distinguish memory for mainstream, working on regular frequencies and pressure with timings, recommended by JEDEC specifications, and memory "advanced" or overclockers, which, as a rule, despises JEDEC rules, also tries to squeeze out of silicon chips (both the memory, and chipsets/controllers, its serving) a maximum of high-speed opportunities. It is clear, that the majority, in the market, are memory for mainstream and that’s not so interesting to computer enthusiasts, because its role reduced to routine performance of the functions (stable work in regular modes) and to demand from it "star wars" - would be wrong.
Another matter – overclocking memory. Here are - open space for creativity of both users and manufacturers: everyone tries show himself from better side, and such products quite often serve as a "face" of the company (showing its potential and "steepness") though the turn is done due to more "greyish" and cheap products. This overclocking memory is interesting for enthusiasts (and notice for observers), though it’s mass application obviously restrains essentially by higher prices for memory and also for the products capable to work "correctly" with such memory.
The first generation of system chipsets with DDR2 support (basically, from Intel) was guided by clock frequencies of memory 400 and 533 MHz. In due time for them have been approved the specifications of memory DDR2 (for example, JESD79-2B.pdf and Intel_DDR2_400_533_Spec_Addendum_rev_0_9.pdf) according to which the pressure of a feed in a range from 1,7 up to 1,9 volt (usually 1,8) and work timings 5-5-5, 4-4-4 and 3-3-3 (CAS Latency - tRCD - tRP accordingly) for frequency of 533 MHz (for the clear reasons the 400 MHz are not in interest for us). Generally "regular" for DDR2-533 and timings were 4-4-4 (at a pressure 1,8V) – they are in SDP the majority of corresponding modules of memory. However the latest update of specifications has admitted an existence of DDR2-533 modules working at a pressure of a feed 1,9V and with timings 3-3-3, which before, were considered as overclocking. Later (in 2005, before Intel chipsets with DDR2-667 support) specifications were updated once again and appeared DDR2-667 memory with timings 4-4-4 and 5-5-5 and even DDR2-800 with timings 4-4-4, 5-5-5 and 6-6-6! By the way, to denote different memories timings is accepted by the corresponding letter on the end (please look page 74 of first of documents under the reference above):
Type of memory | Timings according to JEDEC ( CL- tRCD- tRP) |
DDR2-800C | 4-4-4 |
DDR2-800D | 5-5-5 |
DDR2-800E | 6-6-6 |
DDR2-667C | 4-4-4 |
DDR2-667D | 5-5-5 |
DDR2-533B | 3-3-3 |
DDR2-533C | 4-4-4 |
DDR2-533D | 5-5-5 |
DDR2-400B | 3-3-3 |
DDR2-400Ñ | 4-4-4 |
According to the same document minimal timing tRAS should be corresponded 45 nanoseconds (for all modules of memory, except for DDR2-400 V), and tRC varies from 55 up to 65 nanoseconds depending on type of the memory, please look to the following table (please notice, that optionaly here is the opportunity of use CL=2!).
However, to us it’s more convenient to operate in terms of quantity of steps of frequency of the bus of memory (they are specified in BIOS Setup motherboards for adjustment and tuning), therefore the final tablet of DDR2 types specified at present looks like this:
Table 1. Types of memory DDR2 according to JEDEC.
Type of memory (on frequency of data transmission ) | Alternative marking (on peak transfer rate Mb/sec.) | Length of frequency of the bus , n/sec. | Timings according to JEDEC (CL- tRCD- tRP) | tRAS, cycle | tRC, cycle |
DDR2-800 C | PC2-6400 | 2,5 | 4-4-4 | 18 | 22 |
DDR2-800 D | PC2-6400 | 2,5 | 5-5-5 | 18 | 23 |
DDR2-800 E | PC2-6400 | 2,5 | 6-6-6 | 18 | 24 |
DDR2-667 C | PC2-5300 | 3,0 | 4-4-4 | 15 | 19 |
DDR2-667 D | PC2-5300 | 3,0 | 5-5-5 | 15 | 20 |
DDR2-533 B | PC2-4300 | 3,75 | 3-3-3 | 12 | 15 |
DDR2-533 C | PC2-4300 | 3,75 | 4-4-4 | 12 | 16 |
DDR2-533 D | PC2-4300 | 3,75 | 5-5-5 | 12 | 17 |
DDR2-400 B | PC2-3200 | 5,0 | 3-3-3 | 8 | 11 |
DDR2-400Ñ | PC2-3200 | 5,0 | 4-4-4 | 9 | 13 |
Everything that "drops out" of this table to the less (on CAS) it’s better to consider as overclocking modules with timings.
However, despite of "promising" specifications, in practice everything looks more modestly - even after official appearance of chipsets for DDR2-667 overwhelming majority of Intel certificated for work with them modules of memory DDR2-667 have timings 5-5-5 (please look, for example, ddr2_667_dram_results) and only very rare modules DDR2-533 "are certificated" by the manufacturer of chipsets on timings 3-3-3 (please look ddr2_dram_results). Business with chipset NVIDIA nForce4 SLI Intel Edition goes a little bit better (please look nForce_memory_compatibility).
On this background many brands of overclocking memory (Corsair, Geil, Kingston HyperX, OCZ, etc.) compete in demonstrations of potential of their "non-standard" products (if to follow the JEDEC and Intel). For example, the Kingston even in autumn of 2004 has produced modules KHX5400 of HyperX series for work (I do not know, on what chipset they "counted" it) on clock frequency of 675 MHz (675, instead of 667) with timings 4-4-4-10-1 T at a pressure 1,85V (please look, khx_ddr2.asp). Later, during spring of 2005 these modules have been improved ("were improved"), which has allowed the company to produce new KHX6000 series with working frequency of 750 MHz and "guaranteed" timings 4-4-4-12-1 T on this frequency of memory voltage 1,9 V (please look the previous link). And the last, in particular, for demonstration of the highest potential of the top products have been used Kingston: in the beginning of March on the one of Kingston stands, within the limits of next Intel Developer Forum, was exposed the system with memory on 866 MHz frequency!
Was used serial and standard motherboard ASUS P5 AD2-E Premium on Intel 925 XE Express chipset, directed maximum to DDR2-533 (because i955/945 weren’t produced officially then).
In system was used processor Pentium 4 Extreme Edition 3,46 GHz (Gallatin) in unblocked by a multiplier that has allowed to work on the system bus of 1302 MHz (instead of regular 1067 MHz).
Confirming to the words and as arguments for the highest transfer rate of the memory working on frequency of 866 MHz, experts Kingston showed the "active" screen with test utilities SiSoftware Sandra 2005 Pro and CPU-Z.
You can see that on frequency timings works of memory make of 868 MHz 5-5-5-12. However a representative from Kingston has noted, that it is necessary to make working frequency a little lower - to 828 MHz - and modules KHX6000 will easy work with timings 4-4-4-12 (on which, actually, they "will be certificated" by the manufacturer, but for frequency of 750 MHz).
We managed to receive these Kingston KHX6000 modules for tests in our test laboratory. We shall compare them as to direct predecessors Kingston KHX5400 (DDR2-675), and with typical budgetary products of this manufacturer - ValueRAM DDR2-533 which we have briefly described earlier.
Participants of tests and a configuration of the stand
So, for our test we have chosen three complete sets of memory modules from Kingston (pairs 2 x512 Mb): "usual" ValueRAM PC2-4200 (DDR2-533 CL4); "advanced" HyperX PC2-5400 (DDR2-675 CL4) and, at last, "overclockers" HyperX PC2-6000 (DDR2-750 CL4). We’ll try to find out, what is possible "to squeeze out" of these modules and what productive increase of a computer due to it is possible we can receive. For us will be interesting the mutually contradictory aspect of overclocking :
- Overclocking up to a limit (to the detriment of timings)
- Decrease of timings of memory to the detriment of clock frequency to maximum (to find any compromise between them).
As we could see, there are the magnificent modules DDR400 from Kingston in series HyperX , for "regular" work with minimally possible timings 2-2-2-5 (please look, for example, item "Memory DDR400 with minimal CAS").Therefore to us it is double interesting, how this manufacturer managed "to overcome" CAS in memory of new architecture.
As the test stand we used motherboard MSI P4 N Diamond on chipset NVIDIA nForce 4 SLI Intel Edition ideally suitable for "overclocking" purposes. For our purposes better to use processor Pentium 4 Extreme Edition 3.73 GHz with the fast system bus of 1066 MHz, and as a video card was used ASUS AX X800 XT on the basis of GPU Radeon X800 XT. The given processor was successfully dispersed up to 300 MHz on the bus (up to 4,2 GHz on frequency of a core), therefore for DDR2 533, 566 and 600 parameters of productivity were removed at work of a chipset in synchronous mode FSB and memory. We couldn’t check up the synchronous modes for smaller frequencies - for not clear reasons, BIOS of a motherboard has not allowed to reduce FSB less than 266 (1066) MHz. For higher frequencies of memory it was used asynchronous (even - pseudo-synchronous) mode of chipset.
Kingston ValueRAM PC2-4200 CL4 2 x 512 Mb (KVR533D2E4K2/1G)
It is "standard" Upgrade Kit from Kingston 1 Gb (2 x 512 Mb) from two modules.
One-sided modules, it is used nine 512- Mb chips DRAM from Elpida.
The mode ECC of correction mistakes is supported (was not used in our tests, because the chipset nForce 4 SLI Intel Edition doesn’t support it).
Core voltage of these modules - 1,8 volt. The system with these modules has started without problems. The information registered in SPD modules:
SPD information | Kingston PC2-4200 CL4 2 x 512 Mb (KVR533D2E4K2/1G) | ||
Frequency | 200 (400) | 266 (533) | 266 (533) |
CAS# latency (CL) | 3 | 4 | 5 |
RAS# to CAS# (tRC) | 3 | 4 | 4 |
RAS# precharge (tRP) | 3 | 4 | 4 |
RAS# Active Delay (tRAS) | 9 | 12 | 12 |
That is declared in SPD characteristics in accuracy correspond to the name of modules: 4-4-4-12 on frequency of 533 MHz and 3-3-3-9 on 400 MHz. Lets try to disperse this memory slightly. To check of stability we’ll use run S&M (former testmem) in mode Quick (8 minutes).
The maximal frequencies at which modules of operative memory worked stably:
Timings | 3-3-3-9-1 T | 4-3-3-9-1 T | 4-4-4-12-1 T | 5-5-5-16-2 T |
1,85 V | 451 MHz | 480 MHz | 640 MHz | 692 MHz |
2,10 V | – | – | – | – |
We can say that it was a surprise that for the given modules of memory the increase of core voltage from standard up to 2,00 and even 2,10V has led to that modules began to work with failures. The system could start on slightly higher clock frequencies, but in any case already on first-third minute S&M started to fix failures even on those frequencies where modules earlier worked without problems. However, this effect isn’t so strange for modern microelectronic crystals where threshold pressure and an operating conditions "are checked up" to 100-th shares of volt and their excess (and not just underscore) can be a fault of an overheat (including, local sites of chips) and also displacement of the transfer characteristic of gates to edges of working area where a safety margin is essential below - especially, on such high frequencies of work. So we can’t take it as a shortage of modules.
Dependence CAS and transfer rate of memory from frequency of work of modules. Look to this schedule.
On these schedules dark blue line corresponds to timings 3-3-3-9, violet - 4-3-3-9, yellow - 4-4-4-12, blue - 5-5-5-16-2 T; the shaded line - at synchronous operating mode FSB: DRAM. Let's notice that using of 2 T command rate has been unprofitable: at small timings the advantage in frequency from 2 T-addressings has been tiny, and losses in transfer rate and CAS of memories - huge. So NVIDIA was right focusing that nForce 4 SLI Intel Edition has allocated bus addresses and commands for each of modules (1 T-addressing just and is the main advantage from the use of such scheme).
We can say that in the whole these modules are well suitable for budgetary and mainstream solutions. On declared timings they work with appreciable "safety factor", without a raising of a pressure the work was stable as DDR2-667 (but only on 5-5-5-15-2 T), support ECC. However, because of this support, they cost about 150 $ (against 110 $ for usual Gb DDR2-533). Taking into consideration that they don’t put records of speed, we can’t recommend them to overclocker (though 640 MHz on 4-4-4-12-1 T - is a quite good result). And, perhaps, the only thing, on what it is possible to complain here - that these modules are absolutely not capable to work with CAS 3-3-3 on frequency of 533 MHz (in contrast to some budgetary competitors). But in a workstation on the basis of chipsets Intel 925 X/955 X Express (these chipsets support ECC) they approach ideally. Additional information on modules KVR533 D2 E4 K2 and operative memory DDR2 you can find in article "Memory DDR2-533 from Kingston Technology: the first sight and is a little about prospects”.
Kingston HyperX PC2-5400 CL4 2 x 512 Mb (KHX5400D2K2/1G)
It is a "standard" complete set of overclocking memories under brand HyperX, which become the first with support of such high frequency for CAS 4-4-4-10-1 T during autumn of 2004. 1 Gb (2 x 512 Mb), one-side modules, are executed on the basis of eight 512-Mb DRAM chips and closed by an aluminium radiator of dark blue color, which is difficult to remove.
Voltage - 1,85 V (it is raised concerning 1,8 V). The manufacturer tests such modules (in pairs) for stable work with ò timings 4-4-4-10-1 T on frequency of 675 MHz. Our system also has started without problems. But the SPD of modules was unexpected:
SPD information | Kingston PC2-5400 CL4 2 x 512 Mb ( KHX5400 D2 K2/1 G) | ||
Frequency | 200 (400) | 266 (533) | 266 (533) |
CAS# latency (CL) | 3 | 4 | 5 |
RAS# to CAS# (tRC) | 3 | 4 | 4 |
RAS# precharge (tRP) | 3 | 4 | 4 |
RAS# Active Delay (tRAS) | 9 | 12 | 12 |
The matter is that in Kingston specifications to KHX5400D2K2modules specifies, that they should work on frequency of 675 MHz with timings 4-4-4-10, however there is no any hint that they are capable of it, in SPD we do not find. Simply, SPD modules KVR533D2E4K2/1G and KHX5400D2K2/1G are identical. It is made so in order modules could be started practically on any motherboard.
We shall try to arrange to modules one more check by means of S&M. The maximal frequencies at which modules worked stable:
Timings | 3-3-3-9-1 T | 4-4-4-12-1 T | 5-5-5-16-2 T |
1,85 V | 533 MHz | 640 MHz | 692 MHz |
2,10 V | – | 640 MHz | 720 MHz |
The first results of tests on minimal timings have been encouraging: modules have successfully worked on 533 MHz frequency. We tested these modules for a long time in our laboratory with various motherboards, result has shown, that they work stable on 533 MHz even with timings 3-3-3-7! But that’s all: to lift the frequency up to 560 MHz with same timings was impossible even with a raising of a pressure. Use of timings has not improved 4-3-3-9 the situation and maximal frequency with timings 4-4-4-12-1 T that was reached has made only 640 MHz (to improve this result by raising of pressure also was impossible). And only the use of absolutely "modest" timings 5-5-5-16-2 T has allowed us to achieve stable work of modules on the declared 675 MHz.
If render to justice we can mark, that in our laboratory was also tested other pair modules KHX5400D2K2/1G - with a label "Evaluation". Here these modules worked super on 675 MHz with timings 4-4-4-10-1 T at a pressure 1,85 V. on various motherboards and chipsets. So let’s not guess, why was observed such "disagreement" between a sample and a serial one, but for this article logically we have taken the last one.
The "problem" in work of these modules was 1 T-addressing … so, unfortunately, it means, that on nForce 4 SLI IE from reached 720 MHz is not so much; the schedules:
You can see that there are no any special advantages (both on transfer rate and CAS) mode 720 MHz 5-5-5-16-2 T in compare 640 MHz 4-4-4-12-1 T. So we can add that in speed, described above, twice cheaper modules ValueRAM, at work on high clock frequencies these modules in general don’t give advantages! Overclocking of Pentium 4 bus up to FSB 1300-1400 MHz can do only extreme users (for core Prescott with minimal multiplier 14 overcloking corresponds to frequencies of the processor 4,4 - 5,0 GHz!) so we don’t see any need in use of these frequencies. However, as modules DDR2-533 working with minimal timings 3-3-3-7, these modules will be good (though it’s a sum for cheaper modules KHX4300). If they also have a price (270 $) or less …
Kingston HyperX PC2-6000 CL4 2 x 512 Mb (KHX6000D2K2/1G)
One more "overclocking" complete set HyperX from two 512-Mb modules. We can say that it’s a “high end” among DDR2 modules at present moment, but probably in nearest future there will come changes to more faster modules (please look, for example, the beginning and the end of this article). Modules two-side, used sixteen 256 Mbit chips DRAM (in predecessors was used twice more capacious eight chips). They are closed by a firm aluminium radiator of dark blue color.
By the way, in work on regular pressure these and previous modules were cold or hardly warm (if, certainly, there is no close to them any "oven"). Nominal voltage for them is 1,9 V.
The system has started without any problems. And SPD modules, fortunately, any more does not make so strange impression:
SPD information | Kingston PC2-6000 CL4 2 x 512 Mb (KHX6000D2K2/1G) | ||
Frequency | 200 (400) | 266 (533) | 333 (667) |
CAS# latency (CL) | 3 | 4 | 5 |
RAS# to CAS# (tRC) | 3 | 4 | 5 |
RAS# precharge (tRP) | 3 | 4 | 5 |
RAS# Active Delay (tRAS) | 9 | 12 | 15 |
However, if to compare that Kingston promises the work of these modules on frequency of 750 MHz with timings 4-4-4-12-1 T (at a pressure 1,90 V), so SPD information, we can say, is extremely careful.
The maximal frequencies on which modules KHX6000D2K2/1G worked stable:
Timings | 3-3-3-9-1 T | 4-4-4-12-1 T | 5-5-5-16-2 T |
1,85 V | 533 MHz | 667 MHz | 750 MHz |
2,10 V | 566 MHz | 700 MHz | 750 MHz |
It is evident that these modules, unlike two previous pairs, obviously work at the raised voltage more successfully. In specifications 1,90 V is instead of 1,80 or 1,85, as at KVR533 or KHX5400. However, with increase of clock frequency the difference between results of tests with pressure 1,85 and 2,10 V is quickly reduced and to 750 MHz comes to zero. But it was impossible to receive even 780 MHz to having used a pressure 2,20 V and timings 5-5-5-16-2 T. But on smaller frequencies these modules provide magnificent CAS: 3-3-3-9-1 T to 566 MHz and 4-4-4-12 to 700 MHz! The excellent result though the delays declared by the manufacturer 4-4-4-12-1 T on 750 MHz we didn’t reach, despite of "correct" (like) test system. (Nevertheless Kingston offers guarantees of working with these timings on the declared frequency, but depending on system can be demanded some fine tunings BIOS of a motherboard. We do not know, what does it mean.
Let's look at schedules of CAS and transfer rate depending on frequency:
As well as in the previous situation "highest" frequencies with worsened timings have not given any advantage in transfer rate of operative memory, but "increased " it’s CAS. But KHX6000D2K2/1G works with timings 3-3-3-9-1 T on frequencies to 566 MHz (it is the best CAS result: 66,5-67,0 nanoseconds) and successfully works with timings 4-4-4-12-1 T on frequencies to 700 MHz (the best result on transfer rate and one of the best on CAS). So to overclock the memory above 700 MHz, probably, it is inexpedient, so, these modules we can mark as ideal choice for overclocker. But the price (370 $) is high.
At last we shall result the table of "base" frequencies and typical timings for the pairs modules of memory from Kingston investigated in the given review:
Frequency , MHz | 400 | 466 | 533 | 566 | 600 | 633 | 667 | 700 | 750 |
ValueRAM PC2-4200 | 3-3-3-9 | 4-3-3-9 | 4-4-4-12 | 4-4-4-12 | 4-4-4-12 | 4-4-4-12 | 5-5-5-16 | - | - |
HyperX PC2-5400 | 3-3-3-9 | 3-3-3-9 | 3-3-3-9 | 4-4-4-12 | 4-4-4-12 | 4-4-4-12 | 5-5-5-16 | 5-5-5-16 | - |
HyperX PC2-6000 | 3-3-3-9 | 3-3-3-9 | 3-3-3-9 | 3-3-3-9 | 4-4-4-12 | 4-4-4-12 | 4-4-4-12 | 4-4-4-12 | 5-5-5-16 |
Productivity in applications
So, we can say that modern "high-frequency" (667 MHz and above) operative memory doesn’t have any advantages over "low-frequency " (533 MHz): both key characteristics (TRand CAS) are at the same level or even worse. It is possible to ascertain and that " overclocking" modules of memory, as a rule, allow to use strongly lowered timings on "classical" frequencies of 533 and 667 MHz. And it is good. But for reduced on 8-9 % CAS you need to overpay 130, and even all 230 dollars! Does it cost?
Let's try to lead the express analysis of productivity and take for testing five most popular programs, using them for typical and "top" timings and frequencies. For described above test system with modules KHX6000 we have got following results:
Application | 533 MHz4-4-4-12 | 533 MHz3-3-3-9 | 667 MHz4-4-4-12 | 750 MHz5-5-5-15 |
CINEBENCH 2003, score | 395 | 395 | 395 | 391 |
ScienceMark 2.0, Primorida, sec | 27,01 | 26,69 | 27,08 | 26,84 |
WinRAR 3.30 Benchmark, Kb /sec | 487 | 535 | 535 | 481 |
Far Cry, fps | 79,7 | 81,8 | 82,0 | 80,5 |
Half-Life 2, fps | 94 ,4 | 96,1 | 96,8 | 95,2 |
It is obvious, that the fastest configurations are the second and the third - both with actually overclocking timings, though not provided by JEDEC. The advantage of the third system above the second truly doesn’t cost any efforts and additional expenses. The system with a memory on frequency of 750 MHz, but with "increased" timings obviously loses even 533 MHz! And the system with memory of 533 MHz andtimings 4-4-4-12 (typical at present moment) loses same with timings 3-3-3-9 about 1-2 %.
In these four application s the difference is in productivity nearby 3 % between "fast" and "standard" operative memory. By the way, sometimes the standard error makes this 3 %. However, probably, someone will interest in 10 % gain of productivity in WinRAR, extremely "sensitively" reacting on any changes connected with operative memory. But it is necessary to note that in another conditions "narrow" 800 MHz of the system bus the real gain of productivity will be smaller unfortunately. Our recent detailed researches of platforms on chipsets Intel 945 P Express and NVIDIA nForce4 SLI Intel Edition confirmed that: advantage of systems with DDR2-667 (timings 4-4-4-12) above DDR2-533 (timings 3-3-3-7) is insignificant (obviously less than 1 %), whereas timings 3-3-3-7 against 4-4-4-12 on frequency of memory of 533 MHz nevertheless allow to receive a certain advantage . And in comparison with the previous platforms which are not support memory DDR2-667 (and even DDR2 in general) high-speed advantage decisions is not especially felt.
The conclusion
Progress - is good certainly. But nevertheless we’ll try to be objective.
First, if you are not make overclocking the use of expensive "overclocking" modules DDR2 - money for a wind. Such modules cost much and the increase of productivity provided by them is insignificant.
Second, series ValueRAM from Kingston shows excellent parameters and at the moderate dispersal of the processor of these modules will be more than enough (only one disappointment – don’t work on 3-3-3 on 533 MHz).
Third, the majority of modules "overclocking" memories HyperX DDR2 is capable to work in "full force " (declared, by the way, in their official specifications) are not on all motherboards and chipsets.
Fourth, better to use 1T command rate: increase of frequencies for 2T doesn’t compensate the losses of productivity.
Five, we are not sure in necessity to increase of speed of work of operative memory on an existing platform further (processors) Intel. As it was already marked in our article about dual - cores CPU in spite of the fact that dual - cores CPU can not reach 170 % of productivity of CPU system, transition to more faster operative memory for CPU system gives in the same conditions only 11 % of speed gain. Fast memory today is necessary only for dual - cores and they in current solution from Intel are limited with insufficient throughput of the general system bus.
AMD dual - cores CPU don’t have such "problem of a bus" and could involve all power of new types of memory, but they are going to DDR2 not earlier then at the end of this year. Modules DDR2 working on frequencies to 1066 MHz are very interesting (they were actively promoted on the last Computex). But, unfortunately, to use these modules you can only after half a year. Though cleanly moral satisfaction and pleasure from working with such smart modules as Kingston KHX6000 you’ll certainly receive.