I hope it is not against the rules to post this, also it's not a full review like TPU! does, but on the overclocking of the RAM. Review by HWBox.gr Translation by MrAlex This was originally intended for the English version of the website, but current plans for it are unknown, therefore the review will be posted here before it becomes too out-dated. Please excuse any mistakes, this took a long time to translate and there might be some gramatical or spelling mistakes. Introduction The market for buying RAM has always been interesting. There are numerous companies, a plethora of high quality components, tough competition and competitive pricing. It is very difficult to stay on top in this area, because the needs of prospective buyers will vary, especially nowadays that overclocking has spread everywhere. It is hard to find a computer user who has not heard the name Kingston, and it's not only computer users – but the average person as well. Established in1987, Kingston became a preference of consumers, which has led it to success. You only have to name a few examples: In August 2001, Kingston took fifth place out of the best Manufacturing Companies around the world. In 2002, 2003 and 2004, Kingston was designated Memory Manufacturer of the Year in England. In 2005-2007, Kingston was declared the world's first DRAM Module House Memory Maker! Today it now offers more than 2,000 products that utilise its memory; from computers, servers and printers to MP3 Players, digital cameras and mobile phones! In November 2002, Kingston made the decision to launch a new range of high performance RAM, which was dubbed HyperX . The HyperX series was designed to satisfy even the most demanding PC enthusiasts. The first RAM modules to be released were the HyperX DDR 333MHz (PC2700) and DDR 400MHz (PC3200) which were given to the Intel Socket 478 and AMD Socket 939 platforms. Then in 2004, AMD recognized the tremendous effort of Kingston with the Athlon 64 and Opteron processors, so AMD dubbed it as the best partner! The big jump in the series was but a few years later with the release of DDR2 RAM. Who could forget the incredible 2GB DDR2 1150MHz 5-5-5-15 kit, which utilised the unbeatable overclocking potential of the Micron D9GKX chips. With the arrival of 2007, the transition to DDR3 began, offering unattainable speeds, lower latencies, higher bandwidth and lower power consumption compared to DDR2. The first arrival of DDR3 kits were for the P35/X38/P45/X48 platforms. At that time, the really monstrous chips were the Micron D9GTS, D9GTR and towards the end of the 775 era, the D9JNL. The Kingston HyperX PC3-14400 was singled out again! Unfortunately, the new platforms from Intel, X58 & P55, require the use of memory with lower voltage, so all the above ICs had to be scrapped, to which we entered the era of the 1.65V Elpida MNH-E Hyper, MGH-E Hyper & BBSE. Using the above as encyclopaedic knowledge, we slowly "get to the chase". Kingston sent one of the fastest DDR3 dual-channel Intel P55 HyperX RAM kits to HWBOX in order to test thoroughly and to then present in detail. We would not do it any other way. We therefore present the Kingston HyperX 2x2GB 2000MHz CL9 kit! Specifications – Pictures Before we look at the pictures, here are the specifications of the new HyperX DDR3 kit: - Part Number: KHX2000C9D3T1K2/4GX - 4GB (2x2GB) Dual-Channel Optimized kit - PC3-16000 - DDR3-2000MHz - Timings: 9-9-9 - Voltage: 1.65V - Intel XMP-ready for 2000MHz @ 9-9-9-27-110 1T@1.65V - On the Intel platforms which were released before the X58 Chipset and current AMD systems, the RAM operates voltages from 1.75V up to 1.95V - The Intel X58 platform can operate on arrays of single / dual / triple-channel with a voltage of 1.65V - The Intel P55 platform can operate on arrays of single / dual-channel with a voltage of 1.65V We continue now with some pictures to get an idea regarding the appearance of HyperX 2000MHz CL9. They arrived in our hands in a simple package, without any external protection in a carton box. Unfortunately, a few points have been deducted from their final score, simply because they are prone to damage and very exposed. [Packaging depends on the store though?] The first things that will surely grab your attention as soon as you remove it from the packaging are the tall fins and the blue colouring. If you own a MSI or Gigabyte motherboard, the colours will go in harmony with those of the motherboard, offering a pleasant sight! Taking a closer look, the tall fans perform very well, so low temperatures are expected. Better yet if you want to overclock even further and not be afraid to go above the recommended voltage limit, it can be used in tandem with a RAM cooler or better yet a single 120mm fan, which will disperse the heat released quickly. All the above is great, but what about applying an aftermarket heat sink on the processor? Is the height of the heat spreaders preventing it? The only case we can imagine that you have such an issue is the use of two fans arranged in a push –pull setup. For example, if you choose the monstrous Prolimatech Megahalem and buy a second fan for the push-pull configuration, the fan located near the memory will not fit. Simply place the RAM in slots 2 and 4 and the problem is solved! (At least in this case - literally). In a more extreme situation, what about using the HyperX and the colossal Noctua D14 cooler? With the RAM in lots 1 and 3, there is not even the tiniest chance of it fitting, unless you don’t mind a crooked DIMM slot. Moving on to slots 2 and 4, although everything fits like packed sardines, we can’t but discern how much easier things would be if the heat spreaders were a bit shorter. The aluminium fins of the Noctua heat sink is literally a hair away from the HyperX DIMM in slot 2. Overall we are pleased with the cooling solution that Kingston has chosen. This guarantees the best performance, along with a quality and stylish cooler which means longer life and much more overclocking potential. In quite a few cases, the utilisation of a second fan on the heat sink will prove problematic, which can be viewed as a disadvantage for some and this definitely stops out potential buyers. It really would have been better if Kingston had used shorter heat spreaders. Still, since this RAM is directed at P55 users and if you encounter any problems, you can just put the RAM in slots 2 and 4, and there shouldn’t be much of a problem. It is worth mentioning that after discussions with some… "special" people, we’ve managed to identify the chips on the RAM kit. We’ve found out that this specific RAM kit could possibly use the Elpida BBSE chips. They are of course no match for the incredible Elpida Hyper chips, but maybe we have a pleasant surprise in store… Part Number Decoder The part numbers of Kingston’s products are not just a random combination of numbers and letters. Each one means something, not just for our own purposes but also for the proper registration from Kingston. Especially when talking about RAM modules such as the HyperX, which have even more importance as it is designed for enthusiasts and gamers, in which useful information can be hidden in the part number. Test System – Methodology Before proceeding to the testing, here are the system specifications that will be used to test the HyperX RAM, the programs used and a complete description of the methodology. Our goal was not only to achieve the maximum frequency where the RAM is just about "benchable", but where the frequency is 100% stable for a system that would be used 24/7. So let's see exactly what we will use and how: CPU: Core i5 750 / Core i7 870 (Stepping 5, Revision B1) Cooler: AC LGA1366 M/B: ASUS Maximus Formula III RAM: Kingston HyperX 4GB 2000 MHz CL9 HDD: Western Digital 74GB Velociraptor GPU: Gigabyte 8800GT 512MB DDR3 PSU: Xilence 750W OS: Windows 7 Ultimate x64 Programs that were used CPUID CPU-Z v1.54 HCI MemTest 4.0 SuperPi / mod1.5 XS MaxxMEM2 - PreView Lavalys EVEREST 5.5 (Cache & Memory Benchmark) CPU-Tweaker v1.4 BIOS Settings The main aim here is to get the speeds that are feasible and achievable by anyone, or at least similar to those that a normal user can achieve. For this reason we tried to keep the processor speed around 4GHz with 4 cores enabled and PLL Voltage at 1.85V. The VTT (IMC Voltage) barely had to rise above 1.378V, when utilising speeds below 2150-2200MHz. However with most speeds above 2200MHz, the integrated memory controller of the CPU began to find it difficult to keep the 1.378V set, so the voltage was increased to between 1.43-1.48V. For this reason it is not recommended to run above 2200MHz unless very good cooling is used. Methodology The HCI Memtest 4.0 will be used for 24/7 Stability tests and the famous SuperPi 32M to test the stability of the RAM to emulate real scenarios. There wasn’t the slightest reason to use the SuperPi 1M, PiFast and wPrime tests due to the reason that they barely stress the RAM. The RAM has been tested thoroughly with CAS 6, CAS 7, CAS 8 and CAS 9 timings and operated at voltages of 1.65, 1.7, 1.75 and 1.8V. Turning to the HCI MemTest, the developers recommended at least testing the memory for 20 minutes to see if they are supposedly 100% stable. Instead, we ran four instances of MemTest and we allocated 800MB for each RAM module (total 3200MB) to be tested. Knowing that we used Windows 7 64-Bit, this test used up almost the entire RAM available to the system. We stopped the test after it had completed 300%, which is basically 50 minutes. Moreover, because most Core i5 / Core i7 system owners run RAM at speeds of 1600, 1800 or 2000MHz, we ran MemTest at those frequencies, using 1.68V and the "tightest" timings possible. This is so that you can make a direct comparison of Kingston HyperX 2000MHz C9 with its competitors. CAS Latency and Voltage Scaling It is time to move on to a more important part of our review, which is none other than the CAS Latency and Voltage scaling of the Kingston HyperX KHX2000C9D3T1K2/4GX. These results will show not only the how the RAM clocks at default settings, but at higher voltages and tighter timings, to see if the RAM kit gives the Overclocker and the Power User, what they really need. So first let's take a quick look at the chart and we will then look at the CAS Latencies in much more detail: What the Kingston HyperX RAM has just shown us has left us with our mouths open! What can we say? The RAM is good for any use, even if the owner is a super-demanding Overclocker, with demanding needs - it is indeed unusual to see RAM rated at 2000 MHz with a CAS Latency of 9 to be able to go all the way down to CL6, and with CL9 being able to reach 2400 MHz! But behind all this lies a little "secret." What is it? The use of looser tRCD, which essentially "unlocks" the RAM, allowing the ability to reach much higher clock speeds, which give a clear advantage in all the benchmarks and programs tested. Let's look in more detail on each separate CAS Latency: 6-9-6-X 1T: Intel P55 - ASUS MAXIMUS III Formula: 1874 MHz :: 1.65V | 1926 MHz :: 1.70V | 1986 MHz :: 1.75V | 2034 MHz :: 1.80V Initial testing began with timings 6-7-6. But after the stress testing, we’ve seen that it is better to use timings in which the RAM can deliver. So we tried tRCD from 7 to 10 to see where the most scaling occurs, and we’ve been quite surprised. While scaling with 6-7-6 was virtually zero, with looser tRCD exceeded all expectations! As you see with 1.8V we managed to get stable at 2034MHz for SuperPi 32M run. We then ran 4x800MB 300% MemTest runs (50 minutes of continuous stressing) and the results were less than satisfying: 1874MHz using 1.65V and 1926MHz using 1.7V. 7-9-7-X 1T: Intel P55 - ASUS MAXIMUS III Formula: 2074 MHz :: 1.65V | 2140 MHz :: 1.70V | 2190 MHz :: 1.75V Considering the fact that tRCD 9 and 10 proved salutary, we continued our tests. While timings of 6-9-6 stopped the memory at 2034MHz with 1.8V, when loosened to 7-9-7, we were able to reach 2074MHz at 1.65V! The scaling was satisfactory (every 0.05V, an increment of 50MHz was applied) up to 2190MHz. From there, things started to get difficult, not because of the modules, but because of BCLK which was at 219MHz. See for the CL6 and CL7 timing tests we used an Intel Core i5 750 which doesn’t have a ratio of 2:12 for the FSB / DRAM Multiplier. But no problem, we didn’t give up. We stopped for two days testing and came back more focused. While waiting for the Intel Core i7 860, 4 instances of MemTest were run, stable at 300%, at the following speeds: 2074MHz at 1.65V and 2094MHz at 1.7V. The small difference between the i7 860 and the i5 750 is that the IMC of the i7 860 is able to work with more than 210MHz BCLK. 7-10-7-X 1T: Intel P55 - ASUS MAXIMUS III Formula: 2274 MHz :: 1.80V Finally, the Core i7 860 arrived. The end of the 2:10 divisor of the i5 750! With a ratio of 2:12, and just 190MHz BCLK the modules managed to reach up to 2274MHz! Don’t forget that we are still at CAS 7 and we’ve already approached 2300MHz. Can this RAM go any higher? 8-9-8-X 1T: Intel P55 - ASUS MAXIMUS III Formula: 2120 MHz :: 1.65V | 2176 MHz :: 1.70V In SuperPi 32M using timings of 8-9-8 and utilising the 2:12 divider, we were still not satisfied running at speeds of 2120MHz and 2176MHz with voltages of 1.65V and 1.7V, respectively. Certainly though in MemTest, the Kingston HyperX again showed its prowess! They were rock stable at the same speeds and timing shown in the SuperPi 32M test. 8-10-8-X 1T: Intel P55 - ASUS MAXIMUS III Formula: 2250 MHz :: 1.75V | 2320 MHz :: 1.80V Raising the tRCD 10, we did not believe our eyes ... we were running SuperPi 32M at speeds of 2320MHz, with 1.8V! We also think it appropriate to mention that to see a difference between tRCD 9 and tRCD 10, the integrated memory controller (IMC) of the processor has to be able to handle it. 9-9-9-X 1T: Intel P55 - ASUS MAXIMUS III Formula: 2188 MHz :: 1.65V At this point in the review we would like to remind you that the memories have a factory operating frequency of 2000MHz with timings of 9-9-9-27-110 1T (1.65V). Keeping the voltage at the same level and provided that the IMC of your CPU can hold out, you will be witnessing the amazing results that follow. SuperPi 32M was bench able at 2188MHz with 1.65V, and the tRFC at 88, as has been in the whole review. With a tRCD of 9, the modules were unable to go above ~ 2200MHz stable, regardless of the voltage. 9-10-9-X 1T: Intel P55 - ASUS MAXIMUS III Formula: 2274 MHz :: 1.70V | 2360 MHz :: 1.75V | 2395 MHz :: 1.80V Thus, reaching the end of testing, we increased the tRCD to 10. We were literally left with our mouths open! Using voltages of 1.8, the modules manages to reach 2395MHz benchable in SuperPi 32M, right within walking distance of the 2400MHz barrier! Note: Although mentioned in the methodology, it is good to mention it here as well. When the frequency exceeds 2200MHz, the 1.378V VTT (IMC) Voltage is no longer enough. Because no-one will be running this high for 24/7 use, we increased the VTT from 1.43 up to 1.48V. The limit, I would say, would be anything over a VTT of 1.4V. Highlights In this section you will have the opportunity to look at the impressive frequencies which the Kingston HyperX managed to reach, as well as three stability tests which you will leave you shocked! But before I go there, I’ll remind you exactly what the default setting are from the factory, which is a frequency of 2000MHz with timings of 9-9-9-27-110 1T. This was done with the assistance of two particularly useful programs, MaxxMEM2 - PreView and the popular Lavalys EVEREST Cache and Memory Benchmark. If you’re wondering how to access the specific benchmark, open Everest, go to the Tools menu and then to the corresponding benchmark. It is apparent though, that something isn’t right with MaxxMEM2. Look at the difference between the copy and read tests and you will see. In MaxxMEM2 the speed for the RAM’s copies/read/write are 19663/17965/15613 MB per second respectively, while in Everest they are 21406/18761/15613 MB per second. The graph shows that the Everest benchmark is far more reliable. Of course what is important for MaxxMEM2 is not the bandwidth measurements but by the score in which is gives 1238, exactly what Kingston says it should be! In any circumstance do not confuse and compare the bandwidth between triple channel memory and dual chancel memory as the difference will be admittedly massive, but unfair. This is due to the triple channel’s memory bandwidth being sufficiently greater than compared to dual channel memory, for reasons which will not be mentioned now. We will continue our review with our favourite tests consisting of two parts: the memtest 300% stability test and SuperPi 32m at speeds exceeding 2400 MHz. Yes, you read that correctly, 2400 MHz. For the memtest we did something very simple. We put the memory to do 4x800 MB tests up to 300%, at speeds of 1600 MHz, 1800 MHz and 2000 MHz. These frequencies were not achieved by luck. Bearing in mind that the majority of people with an LGA 1156 processor will run this RAM in frequencies similar to the ones we have shown you, we have found the tightest timings which work stable. The results are more than surprisingly, making us proud! We managed to get 1600 MHz at 6-7-6-18-88 1T, 1800 MHz at 6-8-6-18-88 1T and 2000 MHz at 7-9-6-18-88 1T. Don’t forget that at the time of writing of this review, the RAM has a retail price of €168 on the Greek market. MemTest 300% Stable @ 1.68V 1600 MHz @ 6-7-6-181800 MHz @ 6-8-6-18 2000 MHz @ 7-9-6-21 The 2400+ MHz Zone @ CAS Latency 9 2408 MHz @ 9-12-9 @ 1.81V2442 MHz @ 9-12-9 @ 1.81V 2480 MHz @ 9-15-9 @ 1.82V 2400+ MHz zone at CAS' 9. What’s this? We’ve saved the best for last. The first run managed to get 2408 MHz with timings of 10-10-10-31, simply to see the effect at such high frequencies. The scene changed quickly though! Congratulations are deserved for Kingston for producing such an amazing kit, because after exhaustive testing and sufficient memory tweaking we arrived at frequencies of 2442 MHz at 9-12-9 and 2480 MHz at 9-15-9, SuperPi 32m bench able. The use of such high voltage though is not recommended for 24/7 use, having in all cases exceeded 1.8v to reach frequencies above 2400 Mhz. This Kingston HyperX kit can proudly say it is part of the 2400+ MHz club. At this point we would like to make a necessary clarification, which requires special attention. The new processors by Intel, as we all know, are equipped with an integrated memory controller (IMC). This directly affects the overclocking potential of the RAM. So watch out, because once you break the 2200 MHz barrier, the IMC will have difficulty withstanding the overclock, especially if the VTT voltage is at a low level. So how much is needed for 2400 MHz? If you are going to play with such high frequencies it is good to know that the IMC plays a large role. In our case the VTT (IMC) voltage was almost 1.55v, the RAM’s voltage at 1.815v and the PLL voltage at 1.95v, while the vast majority of timings/sub-timings were given manually. Temperatures Regarding temperatures, the HyperX heat sinks will cover your requirements. Although the heat sink is very tall, it does its job. And believe me, it does it very well. To measure the temperatures we used the GEIL Evo Cyclone’s LED thermometer. The MAXIMUM temperature at 1.65v, even after one hour stressing, was 24°C. Even going up to 1.7v, it never exceeded 25°C. At 1.75v we were feeling ambitious - with right as well as the temperatures only reached 27°C! Up to 1.8v, the temperature ranged between 28 to 30°C not even hot to the touch! Does it matter though? If you are simple user, you will never need to give more than 1.7v, but if you are an overclocker you will certainly use some form of RAM cooling. But even at 1.8v, with temperatures of 28°C, there is almost nothing to worry about! Conclusion Here we are at the last part of the review, the conclusion. The truth is when I first started testing the RAM, I never expected it to reach such high frequencies, firstly because of the factory settings of 2000 MHz at 9-9-9-27 and secondly because it cost just €168. Nowadays, usually companies which have aggressive marketing strategies choose to charge a premium for such high quality modules. Although they offer high clocks and characteristics which raise an eyebrow, the price is still too much for the average user. This review is proof that amidst the depths of cheap RAM, there are a few gems to be found! I personally believe that the Kingston HyperX 2000 MHz C9 deserve every single cent of what it is worth. You should always know that in the market of high performance memory, there will be both unpleasant and pleasant surprises. This is due to companies not using standard chips on their RAM; therefore each memory kit differs from one another. But you should know something else: Never run modules more than they are designed for, such as high frequencies and tight timing. If such is the case you are currently looking for, then this memory kit is not for you. For what reason do we say this? Quite simply, because these Elpida Hyper Memory ICs are binned for low CAS/trcd/reversion latenices. These times are difficult and there are VERY few people who are willing to spend an extensive premium for RAM with tight timing and high frequencies. Costing just €168, quite frankly there is nothing more to be asking for.