overclocker Skylake from China / Processors and memory


The progress that Intel processors undergo during the change of generations of microarchitecture has recently slowed noticeably. Indeed, if we compare the processors of the next few years of release similar to each other in positioning, it turns out that their computing performance differs at best by 3-7 percent, despite the fact that the developers are constantly talking about huge steps (“ticks” and ” takah “) in the development of microarchitecture. Therefore, it is not surprising that many system owners even five years ago simply do not see the point in upgrading their computers and continue to stay with Sandy Bridge, sincerely believing that in the absence of real competition the development of desktop processors has seriously slowed down.

However, in reality, everything is not so unambiguous. With the thesis of the absence of a noticeable increase in productivity, we will not argue. But with the fact that nothing interesting has happened in the development of desktop systems for a long time, it is impossible to agree. And it’s not about the graphics core built into the processors and not about the introduction of new interfaces into modern platforms, but about the fact that over the last couple of years Intel has gone a long way towards returning to overclocking its original meaning.

With the transfer of processors to the Core architecture, Intel tried to overturn the idea of ​​overclocking processors, making it not a means of saving the budget, but, on the contrary, a very costly sport. And partly Intel (with the active assistance of the rest of the industry) it was possible: processors, capable of overclocking, became noticeably smaller, and they became much more expensive. However, the massive practical overclocking, fortunately, still did not disappear. In part, it was fueled by the efforts of AMD, and sometimes some gifts (for example, the Pentium Anniversary Edition) were presented by Intel itself. But the most important event for an overclocker’s renaissance occurred recently – with the advent of the LGA1151 platform and the sixth-generation Core processors.

The fact is that, as it turned out, in the LGA1151 ecosystem, any processors can be overclocked in general: as specially designed for this model with an index K in the name, and all other Skylake with the blocked multiplication factor. Yes, overclocking of ordinary processors is not so straightforward as for expensive K-versions. It is performed not by changing the multiplier, but by increasing the frequency of the basic clock. Therefore, not all boards are suitable for this overclocking, special BIOS versions are needed, and AVX / AVX2 instructions are disabled as a side effect. Nevertheless, the results are quite positive: in a number of applications – primarily of a game nature – one can get a noticeable increase in productivity.

Of course, we can say that Intel has no direct relationship to the revival of mass overclocking, because the company still does not give free, but tries to impose restrictions on overclocking the Skylake models without the K index by organizational measures. But first, the struggle with the overclocking of mass processors is not particularly successful, and open loopholes have not disappeared anyway to this day. And secondly, who, if not Intel, laid the hardware basis for such dispersal? This whole story became a reality only due to the fact that in the LGA1151 platform there appeared the possibility of using two clock generators at once: one for the processor frequency, and the second for the frequency of the interfaces. Prior to the advent of Skylake, this separation of frequencies in the design of the platforms was not laid, and therefore it was simply impossible to overclock the processors in some other way than changing their multipliers. In other words, Intel is involved in the return of overclocking inexpensive processors in the most direct way, and to deny its role in this would be unfair.

But the most convincing proof that Intel was seriously considering the possibility to open Skolake’s full freedom of overclocking, not so long ago came from China. On numerous Chinese trading platforms, pre-release processors of the Skylake generation began to be sold actively, which are freely dispersed by the frequency of the clock generator in general, without any tricks and limitations. That is, the blocking of the overclocking of non-K processors was added by the microprocessor giant just before the Skylake was brought to the market, and it is not surprising that such artificial inhibitions could eventually be bypassed.

This whole story is not only about the fact that Intel wanted to add overclocking to Skylake, and then changed its mind, but the crafty motherboard manufacturers were able to implement the original Intel plan without the involvement of the microprocessor giant. She adds one more important touch to the already well-known picture of Skylake overclocking: in nature, there are pre-release LGA1151 processors, which, with the help of increasing the frequency of the clock, are able to accelerate out of the box, without any additional conditions. And this fact is of great practical value, because these processors can be obtained if desired, and at a very attractive price.

Actually, this article is devoted to the story about testing such an easily overclocked pre-recovery processor. Recently, computer forums are flooding avalanche of enthusiastic reviews from users who purchase on Chinese Internet sites low-frequency engineering samples of four-core Skylake and raise their frequency to the level of Core i7-6700K.

To believe that this is possible is not too simple, since the Chinese sell such processors about two to two and a half times cheaper than a full-fledged modern quad core Core i7. That’s why we decided to work it out ourselves and ordered one of the processors with Aliexpress with a mysterious and Intel Core i7-6400T name that does not exist in Intel’s understanding, for which the Chinese vendors promise the possibility of easy overclocking and operation at a frequency of about 4.0 GHz.

⇡ # Chinese Core i7-6400T: from where the feet grow

First of all, let’s try to find out what is really hidden under the name Core i7-6400T. In the understanding of Intel, such a processor does not exist at all, and Chinese vendors use this name solely for their own convenience – it is necessary to somehow designate the product being sold. The marking applied on the processor cover also does not add clarity. No clear name on the covers of such CPUs is given, and the only available text string “Intel Confidential” serves as a reminder that this processor is an engineering model and is not a serial product.

From a formal point of view, the processors labeled “Intel Confidential” are preliminary chips that the microprocessor giant uses for distribution to its partners: in the ranks of manufacturers of motherboards, computer builders, and among software developers. This is done so that the process of bringing new chips to the market goes more smoothly. Preliminary samples are needed by the company’s partners for understandable reasons – for debugging and optimizing their own future products. Therefore, it is not surprising that the dispatch of Intel Confidential samples long before the official release of processors analogous to the microarchitecture is a rather massive phenomenon. However, such processors on sale, of course, should not fall, and even more so on such a scale as it happened with the conditional Core i7-6400T.

Where did the number of “extra” engineering samples of Skylake come from, which now leaked into the market? The answer to this question is quite simple, if we recall that in the process of developing this processor design, three of its stepping (versions) were successively replaced. Typically, Intel, after it switched to the “tick-to-tak” development model, managed to manage one or two steppings while preparing new processors, but the Skylake project was not very smooth. The new, 14-nm process technology, which was introduced in Broadwell processors, initially could provide only a fairly limited frequency potential. For example, the 14-nm Core i7-5775C and Core i5-5675C received even lower passport frequencies than their 22-nm Haswell generation predecessors, and the results of their overclocking turned out to be very bad. But then it was not so critical, because the desktop Broadwell was positioned as a completely pass-through product with a very narrow scope. But in Skylake, Intel specialists had to deal with the struggle for gigahertz already with full force. And the ability to work at frequencies no worse than Haswell’s was achieved far from immediately, but only after a twofold redesign of the first version of the kernel.

As a result, the final version of the company Skylake was obtained only by the beginning of 2015, that is, about six months before the official announcement. Intel partners needed a new generation of test processors much earlier, and the company began to send out Skylake engineering samples even before the final versions of the CPU appeared at its disposal. Therefore, later, when Intel finally managed to design and launch the final stepping of the new core in mass production, it had to distribute samples, which this time completely corresponded to the design of future serial products, on a new one.

According to the rules, all the preliminary engineering processors had to be destroyed, but what about the Chinese before the rules? And it is exactly those saved from the recycling of the Skylake prefinal version and finally ended up on the trading floors. This means that inexpensive Core i7-6400T, which are sold on Aliexpress and Taobao, are not exactly the same products as the serial processors, but processors with an earlier version of the design.

It’s easy to verify this – just look at the four-letter marking S-Spec on the cover of the processors: it encodes both the characteristics of the sample and its belonging to one or another stepping.

In the table below we provide information on the most common markings of engineering samples of four-core Skylake.

S-Spec Nuclei / flows Rated frequency, GHz TDP, W Free multiplier Stepping of the nucleus
QH8F 4/8 2.2 65 A0
QH8E 4/8 2.2 80
QH8G 4/8 2.2 80
QH73 4/8 2,3 95 +
QHQJ 4/8 1.6 35 Q0
QHQG 4/8 2.2 65
QHJE 4/8 2.6 65
QHQF 4/8 2.6 95 +
QJEB 4/8 2.8 35 R0
QJE9 4/8 3,4 65
QJE6 4/8 4.0 95 +

The most “right” engineering samples of processors have stepping R0 – they are full analogues of serial products, since this stepping is used in those Skylake variants that are sold in stores. The Chinese sellers – for obvious reasons – only have processors of the past steppings: A0 and Q0. However, although this reduces their practical value, but not much. For earlier steppings, a lower frequency potential and some limitations in the operation of the processor memory controller are characteristic, and in stepping A0 there are also errors in the implementation of the embedded graphics core. However, if we take into account the fact that such CPUs are sold at a much lower price than serial quad-core R0 stepping, their acquisition acquires quite obvious meaning. Especially if we take into account the fact that Skylake steppings A0 and Q0, in contrast to R0, do not have any protection against overclocking by the clock generator and therefore freely freely increase the frequency above the value set by Intel, even in spite of the blocked factors.

In other words, the Skylake engineering of the early steppings offered at Chinese trading floors is a good and moreover a truly overclocking way to save money. Having paid for this processor twice or even three times less than the current Core i7 in the store, from the engineering sample of stepping A0 or Q0 after overclocking it is possible to squeeze a comparable level of performance.

A cursory analysis of the Aliexpress and Taobao range shows that the bulk of the low-priced pre-production samples of Skylake have S-Spec QHQG, QHQJ, QH8G and QH8F. Of particular interest are the QHQG and QHQJ processors, because they relate not to the original, but to the later, second stepping Skylake, in which the frequency potential is slightly improved and a number of problems are fixed, for example instability of the built-in graphics. Of the pair QHQG and QHQJ, the first option seems more attractive, since it is designed for a higher clock speed and has a higher nominal multiplier, but in fact, if we take into account overclocking, then there should not be a big difference between such processors. Yes, to overclock QHQJ, the frequency of the clock generator will have to be picked up more strongly than in the case of QHQG, but this can hardly become a serious problem anyway.

In addition, engineering processors are also differentiated according to the last line of marking, having the form L501C679. It coded information about the place and date of production, as well as the batch number. The first symbol is the factory code (L – Malaysia, C – USA); the second is the year of production (4-2014, 5-2015); the next two digits – the week number of the year, during which the processor was manufactured; and all that follows is the party code. It is believed that processors with a later release date are accelerated better, but if such a correlation takes place, then it is not too noticeable. Nevertheless, the engineering processors of the beginning of 2015 are in great demand, and therefore usually stand somewhat more expensive than their earlier counterparts.

⇡ # Core i7-6400T QHQG: detailed dating

For practical experiments, we purchased an engineering sample of the Core i7-6400T QHQG release of the first week of 2015 (L501). It cost about 10 thousand rubles, that is, somewhere around twice cheaper than the most affordable Skylake class Core i7. Does he discard the funds invested in him?

As mentioned above, there really is no processor with the name Core i7-6400T in Intel’s assortment. This name was invented by the sellers on the basis that the rated frequency of 2.2 GHz, on which such samples operate, corresponds to the nominal frequency of the Core i5-6400T. However, for all other parameters Core i7-6400T is exactly Core i7: it has four cores supporting Hyper-Threading, and its cache memory of the third level has a capacity of 8 MB. But if you compare the Chinese Core i7-6400T with a full Core i7, then the serial processor will seriously benefit by clock speed.

Core i7-6400T Core i7-6700T Core i7-6700 Core i7-6700K
The code name Skylake Skylake Skylake Skylake
Stepping of the nucleus Q0 R0 R0 R0
Kernels / streams 4/8 4/8 4/8 4/8
Technology Hyper-Threading There are There are There are There are
The clock the frequency of 2.2 GHz 2.8 GHz 3,4 GHz 4.0 GHz
Maximum frequency in the turbo mode 2.6 GHz 3.6 GHz 4.0 GHz 4.2 GHz
Unlocked the multiplier of the No No No There are
TDP 65 W 35 W 65 W 91 W
HD Graphics HD Graphics 530 HD Graphics 530 HD Graphics 530 HD Graphics 530
L3-cache 8 MB 8 MB 8 MB 8 MB
Support for the DDR4 Two channels of DDR4-2133 Two channels of DDR4-2133 Two channels of DDR4-2133 Two channels of DDR4-2133
Support for the DDR3 Two channels of DDR3L-1600 Two channels of DDR3L-1600 Two channels of DDR3L-1600 Два канала DDR3L-1600
Линии PCI Express 3.0 16 16 16 16
Расширения набора инструкций SSE4.1/4.2, AVX 2.0 SSE4.1/4.2, AVX 2.0 SSE4.1/4.2, AVX 2.0 SSE4.1/4.2, AVX 2.0
Упаковка LGA1151 LGA1151 LGA1151 LGA1151
Цена $130-$155 $303 $303 $339

В низких частотах инженерного образца нет ничего удивительного. Во-первых, перед нами – степпинг Q0, частотный потенциал которого гораздо ниже, чем у последовавшей за ним серийной версии ядра. Во-вторых, инженерные процессоры, выпущенные задолго до официального анонса, всегда работают на более низких частотах. Ведь они предназначаются для знакомства партнёров Intel с новой архитектурой, а не для покорения высот производительности.

Несмотря на то, что Core i7-6400T – процессор не серийный и даже отличается от нормальных Skylake по степпингу, большинство LGA1151-материнских плат может работать с ним без каких-либо проблем. Необходимая для этого версия микрокода обычно встроена в BIOS по умолчанию, по крайней мере в платах ведущих производителей. Поэтому относительно поддержки инженерного образца переживать не стоит.

Вместе с тем детектируется он большинством материнок как «Genuine Intel CPU 0000» – ни о каком Core i7 в названии, естественно, здесь речь не идёт. Однако никаких претензий к функциональности нет: в этом процессоре нормально работают все фирменные интеловские технологии: Hyper-Threading, SpeedStep, VT-x, TXT и так далее. Работает и Turbo Boost: при нагрузке на все ядра процессор может разгоняться до 2,4 ГГц, а при однопоточных вычислениях частота может возрастать до 2,6 ГГц. Впрочем, в любом случае при эксплуатации в номинальном режиме китайский Core i7-6400T не интересен – его частоты слишком низки.

Диагностическая утилита определяет рассматриваемый инженерник следующим образом:

Встроенный бенчмарк подсказывает, что даже многопоточная производительность этого процессора в номинальном режиме находится на уровне ниже Core i5 поколения Sandy Bridge. А значит, для того, чтобы Core i7-6400T мог раскрыть свои сильные стороны, нужен разгон.

⇡ # Dispersal

Так как множитель подавляющего большинства инженерных процессоров заблокирован, разгонять их приходится базовой частотой. В случае со Skylake это вполне возможно на материнских платах, которые имеют внешний тактовый генератор, а это – практически любые платы на базе набора логики Intel Z170. Если бы речь шла о разгоне серийных процессоров, то для успеха потребовался бы ещё и особый, подправленный BIOS. Но для разгона предсерийных образцов Skylake степпингов A0 и Q0 он не нужен – такие процессоры свободно разгоняются по частоте и на обычных версиях прошивок. А это значит, что поднять частоту Core i7-6400T не так уж и сложно. И даже более того, инженерные версии Skylake степпинга Q0 выгодно отличаются от своих более поздних серийных собратьев тем, что при разгоне частотой тактового генератора в них не отключаются термодатчики и AVX/AVX2-инструкции. То есть разгон получается даже более полноценным.

При разгоне Skylake по частоте есть две тонкости: технология Enhanced Intel SpeedStep должна быть принудительно отключена, а параметр Boot Performance Mode требуется перевести в значение Turbo Performance. Эти настройки в большинстве случаев избавляют от проблем с холодным стартом системы. В остальном же алгоритм очень простой: процессорный множитель фиксируется на максимально допустимом значении, после чего наращивается частота базового тактового генератора BCLK и при необходимости для обеспечения стабильности прибавляется процессорное напряжение. Не нужно лишь забывать, что с BCLK связана и частота работы памяти, поэтому при разгоне базовым тактовым генератором попутно требуется корректировать множители, отвечающие за формирование частоты DDR4 SDRAM.

Оверклокинг инженерных Skylake степпинга Q0 происходит именно по этому алгоритму, но не следует ожидать от них таких же результатов разгона, какие получаются на серийных процессорах. Степпинг Q0 – предварительный, и максимальная частота, при которой способны работать его носители, лежит в окрестности 4-гигагерцевой отметки.

В процессе изучения возможностей нашего экземпляра Core i7-6400T QHQG мы построили зависимость его предельной частоты от уровня подаваемого напряжения. На приведённом ниже графике хорошо видно, что частотный потенциал инженерных Skylake действительно хуже, чем у серийных процессоров, но дело не в каких-то искусственных ограничениях, а в самом дизайне ядра степпинга Q0.

Номинальное напряжение тестировавшегося экземпляра Skylake – 1,12 В, однако эксперименты по разгону мы начали с более низкой величины 1,0 В. И как показали практические испытания, даже в этом случае инженерный Core i7-6400T спокойно берёт 3-гигагерцевую планку. Однако особенно обольщаться не стоит. Увеличение напряжения отодвигает предел стабильного разгона не слишком сильно, зато рабочие температуры при этом прирастают очень круто.

В результате прибавка к стартовому напряжению в размере 0,4 В позволила отодвинуть предельную частоту лишь на 850 МГц. И максимальным результатом, который удалось выдавить из нашего экземпляра Core i7-6400T QHQG, оказалось лишь 3,9 ГГц. Попытки же дальнейшего приращения частоты пришлось отмести из-за чрезмерного нагрева процессорного кристалла в тестах стабильности в LinX 0.7.0, даже несмотря на то, что все эксперименты проводились с достаточно неплохим кулером Noctua NH-U14S.

Как видно на приведённом скриншоте, разгон проводился с установкой коэффициента умножения 24х. Это максимальный турбомножитель, при котором исследуемый процессор может работать с нагрузкой сразу на все ядра. Частоту BCLK удалось поднять до значения 162,5 ГГц, что в итоге вывело Core i7-6400T на 3,9-гигагерцевую отметку. Однако для того, чтобы в таком состоянии можно было провести полный цикл тестов стабильности, напряжение Vcore пришлось увеличить до 1,425 В. А это приводило почти к критическому нагреву – троттлинг у инженерных Skylake активируется при 100 градусах, как и у обычных Core i7.

Очевидно, что, как и в серийных процессорах, в инженерных образцах тоже используется полимерный термоинтерфейс, причём явно не лучшего качества. Но для процессора со штатной частотой 2,2 ГГц это и неудивительно. Тем не менее следует иметь в виду, что лучшего разгона Core i7-6400T можно добиться с помощью скальпирования.

На первый взгляд, разгон Skylake до 3,9 ГГц не кажется особенным оверклокерским успехом. Однако не забывайте, речь идёт о предварительном степпинге и о процессоре, номинальная частота которого была в 1,8 раза ниже достигнутого разгона. Поэтому полученный результат на самом деле не так уж и плох. В конце концов, взяв для эксперимента процессор стоимостью порядка $130-$150, в итоге мы пришли к той частоте, которую обеспечивают 300-долларовые CPU. А это, вообще говоря, совсем не иллюзорный выигрыш.

Кроме того, любой оверклокинг – это всегда лотерея. И с другим экземпляром Core i7-6400T результат мог бы оказаться совсем иным. Например, в Сети можно обнаружить немало отзывов, свидетельствующих о покорении инженерными CPU 4-гигагерцевой отметки или даже о возможности стабильной работы при частотах порядка 4,2 ГГц. Иными словами, Core i7-6400T – это вполне достойный объект приложения сил для экономного оверклокера.

Что же касается конкретно нашего случая, то разгон удалось провести без каких-либо особых усилий и без утомительного подбора второстепенных параметров. Изменению подвергалось лишь единственное напряжение VCOREа в целом с набором применённых настроек можно ознакомиться на следующем скриншоте.

Разгон выполнялся на плате ASUS Maximus VIII Ranger, но это не имеет большого значения. Примерно аналогичным образом китайские инженерные процессоры разгоняются и на других материнских платах. Правда, в отдельных случаях можно столкнуться с отключением AVX/AVX2-инструкций и температурного мониторинга, например такое происходит на некоторых материнских платах компании ASRock. Но в этом случае на помощь могут прийти модифицированные энтузиастами версии BIOS, которые регулярно выкладываются на корейском сайте Hardware Tips.

Если вы заметили ошибку — выделите ее мышью и нажмите CTRL+ENTER.

What's Your Reaction?

hate hate
0
hate
confused confused
0
confused
Cute
1
Cute
LOL
0
LOL
Love
1
Love
OMG
0
OMG
Win
0
Win
WTF
0
WTF
Angry
0
Angry
Cry
0
Cry
Geeky
0
Geeky
fail fail
0
fail
fun fun
0
fun

Comments 0

Your email address will not be published. Required fields are marked *

overclocker Skylake from China / Processors and memory

log in

reset password

Back to
log in
Choose A Format
Gif
GIF format