Intel Kaby Lake processors, with which the microprocessor giant “pleased” users of personal computers earlier this year, already received a couple of materials on our site. In them we got acquainted with the older LGA1151-chips in the updated series of Core i7 and Core i5. And the main conclusion is this: to call Kaby Lake a fresh processor design can be a stretch. In fact, this is just a new stepping of the core, which Intel released due to the introduction in the old 14-nm process technology series of improvements. Three-dimensional tri-gate transistors in Kaby Lake received, on the one hand, higher silicon ribs of channels, and on the other – increased gaps between the gates, which actually means less density of semiconductor devices on the crystal. As a result, Kaby Lake has become more energy efficient than Skylake and has acquired the ability to operate at higher frequencies, but no changes at the microarchitecture level have occurred.
However, given the pace at which x86 processors are developing in recent years, even optimizations in the “power supply-frequency” dependencies, which are made in Kaby Lake, can become a sufficient reason for a positive attitude towards them. In the end, this in fact means that the novelties have improved the overclocking potential, and Intel did not please us with such gifts for a very long time. For example, if you remember the main milestones in the development of desktop processors: Conroe, Sandy Bridge, Ivy Bridge, Haswell and Skylake, it turns out that the best overclocking properties in this sequence were the Sandy Bridge generation processors that were introduced in 2011. At that time, with a certain luck, you could find a copy of the processor, capable of operating at 5-gigahertz frequency with air cooling, but no chips that Intel subsequently released on such exploits were no longer capable. At least until the arena appeared Kaby Lake.
Now, it seems, the holiday again returns to the street of overclockers. The optimized 14 + nm process technology, which Intel adapted to the production of Kaby Lake, returned hope for 5 GHz in the air. And although we did not get a single instance of Core i7-7700K in 3DNews that could take this frequency out of the box (and we’ve already verified five processors from different lots), the confidence in the possibility of this is given by the reports of colleagues in the shop , some of which still got samples of Kaby Lake, accelerating to a 5-gigahertz mark.
In general, according to employees from motherboard manufacturers’ laboratories, about 10-15 percent of all Kaby Lake processors related to the unlocked K-series are capable of accelerating to 5 GHz with air cooling. And the frequency of about half of the CPU can be brought to the desired level with the use of the possibility in Kaby Lake of automatic reduction of the multiplier in the performance of AVX-instructions.
However, if you rely not only on luck alone, then the chance of success in overclocking can be significantly increased. The fact is that the first problem that one has to face when raising the Kaby Lake frequency above the nominal is a sharp increase in the temperature of the processor cores. But this phenomenon is due not to the excessive heat release of the semiconductor crystal, but to the fact that the heat released by it is very poorly assigned. The bottleneck is between the copper nickel-plated processor cover (IHS) and the chip hidden under it. A small gap between them is filled with a special polymer thermal interface (TIM), the heat-conducting qualities of which have already raised serious questions for enthusiasts for several years.
With the release of Kaby Lake, the situation has become noticeably aggravated. The fact is that the updated technology process of 14+ nm has reduced the leakage currents and the mutual influence of transistors. And in theory, the frequency can now be raised much more seriously than before. But a higher frequency means a higher heat dissipation, so in order to remove heat from a small Kaby Lake crystal, the area of which is about 125 mm 2 it is required that the thermopaste bridge under the lid let through a very significant heat flow. And he, as practice shows, is not capable of this. As a result, the overclocking of Kaby Lake is much more likely to result in a banal overheating of the processor cores than in the frequency potential of a semiconductor crystal.
Fortunately, the way to solve this problem has long been known – this is the so-called scalping. Enthusiasts have long learned to separate the cover from the processors and change the Intel thermal interface to more effective thermal grease. Especially popular in such cases is using liquid metal – its thermal conductivity is approaching that provided by flux-less soldering of the processor chip to the lid. This method was used by Intel when assembling Sandy Bridge processors, and they were accelerated better than followers, where soldering was not used anymore. Therefore it seems quite logical that if inside Kaby Lake is replaced with a thermal interface, then its overclocking qualities should improve noticeably. Let’s check?
⇡ # Description of the test system
All overclocking tests conducted for this article were performed in a system built from the following components:
- The processor: Intel Core i7-7700K (Kaby Lake, 4 cores + HT, 4.2-4.5 GHz, 8 MB L3).
- The processor cooler: Noctua NH-U14S.
- Thermo-paste: Arctic MX-4.
- Motherboard: ASUS Maximus IX Hero (LGA1151, Intel Z270).
- Memory: 2 × 8 GB DDR4-2666 SDRAM, 15-15-15-35 (Corsair Vengeance LPX CMK16GX4M2A2666C16R).
- Video card: NVIDIA GeForce GTX 1080 (8 GB / 256-bit GDDR5X, 1607-1733 / 10,000 MHz).
- Disk subsystem: Kingston HyperX Savage 480 GB (SHSS37A / 480G).
- Power Supply: Corsair RM850i (80 Plus Gold, 850W).
⇡ # Test processor: Core i7-7700K
For tests in one of the retail stores, we purchased a completely ordinary serial processor Core i7-7700K. This is the flagship Intel quad-core Kaby Lake generation for the LGA1151 platform. Such a processor has a 4.2 GHz passport frequency, but thanks to Turbo Boost it can increase it to 4.5 GHz, supports Hyper-Threading technology and has a third-level cache memory of 8 MB.
Butch (FPO) of the copy we got is L650D048, that is, this CPU was produced in Malaysia from December 12 to December 18 last year. The processor is quite fresh: reviews on most sites usually feature CPUs that date to the end of summer or the beginning of autumn.
The FPO (Finished Process Order) marking line on the processor’s cover stands for:
- The first character is the identifier of the place of production: 0 – San Jose, Costa Rica; 1 – Cavite, Philippines; 3 – Costa Rica; 6 – Chandler, Arizona; 7 – Philippines; 8 – Leixlip, Ireland; 9 – Penang, Malaysia; J – Malaysia; L – Malaysia; Q – Malaysia; R – Manila, Philippines; X – Vietnam; Y – Leixlip, Ireland.
- The second figure is the year of production: 5 – 2015, 6 – 2016; 7 – 2017.
- The third and fourth digits are the production week number.
- The symbols from the fifth to the eighth – the identifier of the party.
The nominal voltage (VID) of this instance turned out to be set at 1.184V, that is, at first glance we got a fairly average processor from the point of view of the overclocking potential.
However, as practice shows, the serial processors of the Kaby Lake generation differ in the VID parameter not so much as Skylake. However, the quality of semiconductor chips, the accuracy of assembly and the parameters of the thermal interface can diverge very noticeably. The difference in the limiting frequency with air cooling of two externally similar CPUs can reach 500-600 MHz. Therefore, you can select the overclocked Kaby Lake only in an experimental way. Unfortunately.
The first test of the processor, which we conducted to evaluate its temperature regime when operating at passport frequencies, showed that it is heated not weak. To remove heat in the test, a powerful tower cooler was used for a 140 mm Noctua NH-U14S fan and Arctic MX-4 thermal paste, but still the maximum temperatures during testing in LinX 0.7.0 reached 80 degrees.
There is a 20-degree reserve before throttling starts, but for a serious overclocking this is very small. It is quite obvious that with an increase in the frequency and voltage V CORE the temperature limit will be reached very quickly.
The following table shows the parameters that were necessary for the operation of this processor at a particular frequency, as well as the maximum temperature recorded during stability tests in the event of any overclocking. In addition, for reference, the total peak consumption of the system measured at the output from the power supply unit is also given.
The good news is that the VID of our Core i7-7700K was chosen with a large margin. As it turned out, for trouble-free operation at the nominal frequency, its voltage can be reduced to 1.1 V. In the limiting temperature, it is possible to win back 6 additional degrees.
The bad thing is that this still does not make the processor successful in terms of overclocking capacity. Its overclocking to at least 4.6 GHz requires an increase in the supply voltage to 1.24 V, and such, in general, not a very significant increase in the frequency causes the processor to warm up to temperatures close to the limiting ones. All this is clearly shown on the graph.
Acceleration to 4.7 GHz for obvious reasons was impossible. And this means that the copy of Core i7-7700K, which fell into our hands, should be classified as unsuccessful. The maximum overclock to 4.6 GHz is a typical level of Skylake, made with the help of an optimized technical process Kaby Lake should, in theory, accelerate more. However, everything rests on high temperatures, which is clearly seen from the screenshot taken during the testing of the stability of the Core i7-7700K in its maximum acceleration.
If it were not for the proximity of the temperature limit, the Core i7-7700K certainly could be overclocked much more. The 4.6 GHz frequency was taken with an increase in the supply voltage by only 4 percent above the nominal VID. Obviously, a more significant increase in voltage would be absolutely safe for the processor chip. This can be judged at least because 14-nm Skylake processors for many users work for a long time at voltages of about 1.35-1.4 V without any problems. However, in order to bring our copy of Kaby Lake to such a level of stress, it is necessary to do something with the heat sink, and it is obviously necessary to begin with the replacement of the internal thermal interface.
It should be noted that such a situation with high temperatures is not a single case, peculiar only to a specific instance of the CPU, but a systematic phenomenon. When overclocking Kaby Lake, we come across time and again that temperatures can go off scale without a serious increase in voltage. Therefore, if you plan to uncover the entire frequency potential of Kaby Lake, scalping seems to be one of the necessary stages of the overclocking process.
⇡ # Scalping Kaby Lake
Previously, two main methods were used to remove the heat distribution cover from Intel processors: the lid was either cut with a thin blade, or it was detached from the processor board by shearing it in the vice. With the release of Skylake, the blade had to be discarded – the board to which the metal heat dissipator was glued was much thinner, the upper layer of the tracks was very shallow in it, and the likelihood of damaging the CPU with an inaccurate movement of the cutting edge became too high.
In the new CPU generation Kaby Lake no fundamental design changes in the processor assembly compared to Skylake did not happen. The metal cover has found characteristic outflows from above and from below and slightly increased the area of contact with the bottom of the cooler, but the processor board is again too thin and fragile, and we would not advise to risk the cutting of the heat distributor. On the photo below you can estimate the thickness of the board this time. Left – Haswell, in the middle – Skylake, on the right – Kaby Lake.
It seems that Kaby Lake has become even slightly thinner than its predecessor. However, this is an illusion, in fact, the thickness of the processor board, as before, is about 0.8 mm.
Therefore, the method of force shift of the lid in the vice remains safer. The traditional procedure is elementary. The processor board rests on one edge of the vise in one jaw, the edge of the lid in the other, and the jaw is gently clamped until the processor breaks up into two parts.
However, even such a method is not without problems . The effort that has to be developed to break the glue joint securing the heat dissipating cover and the processor board can be quite significant. And since the force vector due to the design of the vise is not completely parallel to the plane of the textolite, there is a non-zero probability of damage to the processor board. In the network, you can find a lot of tragic stories about how users broke both Skylake and Kaby Lake in the process of scalping with claws.
Fortunately, such an outcome can be insured. The community of enthusiasts has developed a number of devices that allow you to organize a scalping procedure so that the force that should rip off the lid was strictly parallel to the processor board. The most “untwisted” in this respect is the special tool Delid-Die-Mate 2, developed by the German overclocker der8auer .
This piece is a specialized screw clamp, which in its construction is fitted precisely for scalping. The moving parts in it have recesses for the processor board and the cover, and during operation they are displaced along the rails in different directions parallel to each other under the action of the screw. There is no space for incidents with such a statement of the process. It only upsets the high cost of the device – it costs € 30 and, obviously, can pay off only if you are scalping on an industrial scale.
Fortunately, there is another, much more budgetary option. For safe scalping in the vice, you can use forms printed on a 3D printer. Enthusiasts prepared a number of freely distributed projects of such forms that can be reproduced without any preliminary preparation. For example, we can recommend this option – it is successfully used by many overclockers, experimenters around the world.
Actually, we decided to use this device to facilitate the scalping of the Core i7-7700K. Here it is important to mention that the editorial board of 3DNews, like many of our readers, does not have its own 3D printers. But it’s not a problem. There is a mass of services (including fully online) that can print out to order any model for a ready-made project, and it’s easy to take advantage of any of these offers. As a result, the manufacture of a scalpening device from conventional ABS plastic cost us about 650 rubles.
The principle of operation of this device is very simple. The main part of the form is stacked processor – it is made in the shape of the processor cover.
The edge of the processor board extends beyond the edge of the mold, and on this side a “runner” is put on the device, which rests against the textolite with its bottom. It will be subsequently applied to it, which will have to “shift” the board from the processor cover.
This is how the assembly looks like. Note that the details are adjusted to each other in size so that the processor enclosed in such a “box” does not have any possibility for backlash or bending.
The device is then inserted into the vise and clamped.
Since the edge of the processor board rests against the slider, and the heat-conducting lid is rigidly fixed in the mold, when the jaws are compressed, the same shift is obtained. The device here acts as a guide. It provides the application of effort clearly on the desired line and protects the CPU from any deformation.
Well and further – on rolled down. If you gradually build up the effort, twisting the vice, the glue-sealant, which keeps the metal cover on the PCB, sooner or later succumbs. This moment can not be missed – usually it is accompanied by a distinct attack.
We disassemble the device and see that the heat dissipator is torn off and moved off the processor.
Now you can appreciate the inner world of Kaby Lake. Здесь нас встречает пресловутая фирменная интеловская термопаста, из-за которой нам пришлось идти на весь этот эксперимент.
Выглядит термоинтерфейс не слишком здорово. В лучшем случае он похож на обычную термопасту, однако в нашем процессоре Core i7-7700K теплопроводящий состав был явно пересушен и не размазывался, а крошился. Хотя Intel утверждает, что формула термопасты по сравнению с Skylake не менялась, в этом возникают определённые сомнения.
Дальше следует утомительная процедура по отчистке процессора от пасты и остатков клеевого состава. Немного усердия, и процессор приобретает весьма презентабельный вид.
Остаётся поместить между крышкой и процессорным кристаллом новый эффективный термоинтерфейс и собрать процессор обратно. Конечно, в теории можно было бы оставить процессор и без теплорассеивателя, однако его эксплуатация в таком состоянии сразу вызовет целый ряд затруднений. Его будет невозможно зафиксировать стандартным механизмом LGA1151, при установке кулеров потребуется модифицировать крепление из-за изменившейся высоты CPU, да и к тому же повредить открытый процессорный кристалл – проще простого. Проблемы возникнут и из-за толщины и хлипкости процессорной платы, которую обязательно нужно прижимать в процессорном разъёме не только в центре, но и по контуру, иначе она деформируется и не обеспечит соединения с контактами LGA1151, расположенными по периметру разъёма. Поэтому крышку рекомендуется возвратить на место – она решает целый букет проблем, а при использовании хорошего термоинтерфейса, её влияние на эффективность теплоотвода почти незаметно.
Лучше всего подходит для использования в качестве внутрипроцессорного термоинтерфейса жидкий металл. Мы обычно используем состав Coollaboratory Liquid Pro или Ultra с теплопроводностью около 82 Вт/(м·К), но сейчас на рынке доступны и другие подобные продукты.
Однако именно жидкий металл марки Coollaboratory Liquid Pro и Ultra доказал свою долгосрочную совместимость с процессорными внутренностями. Интеловские никелированные процессорные крышки с ним в реакцию не вступают и не корродируют по крайней мере в течение двух-трёх лет, за которые у нас уже накопилась порядочная статистика.
Пара хитростей: отчищать интеловский клей-герметик лучше всего каким-нибудь деревянным или пластиковым скребком вроде скульптурной стеки. А непосредственно перед нанесением на кристалл и теплорассеиватель жидкого металла не забудьте их тщательно обезжирить. Для этого хорошо подходит не только ацетон или уайт-спирит, но и, например, банальная жидкость для снятия лака.
После нанесения слоя жидкого металла остаётся собрать процессор обратно. В последнее время для этой цели мы используем суперклей на базе цианоакрилата. Помимо очевидных плюсов вроде быстроты образования прочного соединения, он имеет и ещё одно важное преимущество – при нагревании до 70-80 градусов клеевой шов становится пластичным. Это гарантирует ровную в горизонтальной плоскости усадку крышки после установки процессора в гнездо и его включения, а также лёгкость обратной разборки процессора, если такая необходимость возникнет. Утрата же соединением прочности в то время, когда процессор находится под нагрузкой, волновать не должна – в разъёме LGA1151 его крышка надёжно удерживается механизмом сокета.
Остаётся лишь напомнить о том, что важно не забыть оставить в клеевом шве разрыв для отвода горячего воздуха при нагревании процессора во время работы.
Весь описанный в этом разделе опыт подталкивает к выводу, что распечатанное на 3D-принтере приспособление облегчает процесс скальпирования и делает его более безопасным. Однако опытные и аккуратные пользователи вполне могут обойтись и одними тисками без каких-либо дополнительных механизмов. Буквально месяц назад мы скальпировали другой образец Kaby Lake по старинке и в очередной раз смогли убедиться, что подогрев процессора техническим феном с горячим воздухом температурой порядка 350 °С приводит к тому, что крышка отделяется от текстолита с приложением минимальных усилий.
If you notice an error – select it with the mouse and press CTRL + ENTER.