breakthrough-of-the-year-processors-and-memory.jpg

breakthrough of the year / processors and memory

In March, Intel vice president, Lisa Graff (Lisa Graff), publicly promised that during this year the company's attitude to desktop personal computers will be rethought. Among other measures, Intel was going to pay more attention, including high-performance desktops, Which market, as it turns out, does not show a decline, despite all the trends of recent years. In light of the decision, in recent months, we have received several new interesting products for enthusiasts: two processors of the Pentium. But a series of tempting announcements did not end there. Today, for those who consider themselves to be the vanguard of the computer community, Intel presents a high-performance platform with new DDR4 SDRAM, motherboards with a LGA2011-v3 connector and a brand new DDR4 SDRAM.

Yeah, it's about the long-awaited update of the top-of-the-line desktop processor family, which is now finally being transferred to the Haswell-E design, using the most modern version of the Core microarchitecture. And, along with this, the apologists of desktop systems. Firstly, the older Haswell-E has eight processing cores, while all the fastest desktop CPUs for desktops that were released before that day were limited to a six-core design. Secondly, the new processors got an improved four-channel memory controller, which supports faster and more economical DDR4 SDRAM. And thirdly, which is especially important for overclockers, Haswell-E are the only processors of the latest generation, the lid of which Thus, today's announcement has every chance to become the most important event in the market for desktop processors this year. The debut of the Broadwell microarchitecture and the 14-nm process in the CPU for desktops will take place only in a year, so it's likely that nothing will outshine the Haswell-E output.

I must say that the LGA2011 has been updated for a long time. It appeared in 2011 and has now become perceived as obsolete. For it, there are no processors that have a modern microarchitecture, and the existing set of system. It turned out that Intel SX 6 Gb / s and USB 3.0 ports to the motherboard manufacturers had to be added, resorting to additional controllers. To top it all, the recent release of 4-gigahertz Devices has a clock speed of 3.6 GHz, which is based on the Ivy Bridge, with a very serious doubt.

The LGA2011 platform LGA 2011-v3 (or as it is sometimes called, LGA2011-3) coming to replace the LGA2011, however, The processors for it, too, can not boast a higher clock speed than Devil's Canyon. Delivered Core i7-4790K record of the nominal frequency – 4.0 GHz – and remains unconcerned. Haswell-E are in the forefront of the development of a multi-core: the GGS, the number of computing cores is carried to eight. But in this connection, a quite reasonable question arises: which category of users is Intel oriented to such an offer? After all, in the game systems, such a high number of cores is the most likely not required, and creative work with a variety of high-definition content. However, we will not run ahead, and try to evaluate the attractiveness of Haswell-E in practice. Fortunately, our lab received a new eight-core flagship Core i7-5960X, as well as an average model Core i7-5930K with six cores.

⇡ # Haswell-E processors

As with the LGA2011, the processors of the Sandy Bridge-E and Ivy Bridge-E generations, the new Haswell-E family and the entire LGA2011-v3 platform represents an adaptation for the desktop server products market. Server-oriented processors Xeon E5-2600 v3, which are siblings considered in this Haswell-E material, should appear on the market in the very near future. However, while for the server market Intel will offer a very wide selection of Haswell-based products, the Haswell-E line includes only three processors. All of them have higher clock speed and can work with conventional unbuffered DDR4 SDRAM.

The desktop processors of the Haswell-E family are based on an eight-core semiconductor chip manufactured by long-debugged 22nm process technology with three-dimensional transistors. The main building blocks in these CPUs are the kernels with the well-known Haswell microarchitecture. Given the positioning of high-performance Haswell-E processors, their semiconductor chips lacks an integrated graphics core, but it has a four-channel memory controller and a more advanced PCI Express bus controller that supports up to 40 lines of the 3.0 standard and allows them to be divided into different combinations of three, four or five slots. Increased compared to conventional Haswell and the volume of the L3 cache: in Haswell-E, each core has its 2.5 megabyte share.

As a result, the full eight-core Haswell-E crystal has an area of ​​356 mm 2 and the number of transistors it contains has reached 2.6 billion. For comparison: the usual four-core. Haswell processors are based on an almost half smaller crystal 177 mm 2 consisting of 1.4 billion transistors. However, the CPU has still not not deliver. He managed to outperform the eight-core Vishera of AMD, but the six-core Sandy Bridge-E had a crystal with a noticeably larger area – 435 mm 2 .

The Haswell-E processors remain structurally similar to the usual Haswell processors. To connect the in-processor blocks (cores, cache, memory controller and PCI Express controller) into a single whole, a common ring bus is used, and the processor and chipset switching is provided via the DMI 2.0. GB / s. All server features, the QPI bus controller, which is necessary for the interaction of processors in multiprocessor configurations, are cut from Haswell-E. In other words, the CPU Core i7-5000 series is suitable only for single-processor high-performance workstations; for dual-processor servers, dedicated Xeon E5-2600 v3 will be targeted, which, by the way, can contain a much larger number of computing cores – up to 16 or even 18 pieces.

The strongest curiosity among Haswell-E, naturally, is carried by the older version of Core i7-5960X, which carries the status title Extreme Edition. It is the first Intel desktop processor with eight processing cores. Until now, a similar number of cores on the desktop of the AMD Bulldozer family, but there the cores have a greatly simplified design and are paired into modules in pairs, sharing part of the resources with each other. In the case of Haswell-E, we are talking about the high-grade kernels with Haswell microarchitecture, which are also endowed with Hyper-Threading technology support. Thus, the Core i7-5960X is recognized by the operating system as a processor with sixteen cores. In addition to the impressive number of cores, the Core i7-5960X stands out and a large amount of cache memory of the third level. This CPU is the only desktop CPU that has a 20 megabyte L3-cache.

It is obvious that Core i7-5960X Extreme Edition will be very popular in systems. However, it is installed on the traditional ( and therefore not too shocking) for all Extreme Edition-processors price – $ 999. True, the increase in the number of cores has also a negative side reduce the clock speed to keep the processor within the thermal package, which for the desktop platform LGA2011-v3 is limited to 140 W. Therefore, the i7-5960X is only 3.0 GHz.

However, in reality, thanks to Turbo Boost technology, the typical frequency for this processor is 3.3 GHz even at high load. This is the copy of the Core i7-5960X we received for tests.

It should be mentioned that we were much surprised by the very low supply voltage used by this processor. Its nominal level was only 0.97 V, which is significantly lower than that of conventional Haswell. Comparable voltage is used in energy-efficient Intel processors for the LGA1150 platform, but as we have seen, it was required even with the release of an eight-core high-performance CPU.

The middle version of Haswell-E, Core i7-5930K, stands as the ideological heiress of older processors Ivy Bridge-E, as it has six cores with Hyper-Threading technology support and 15-megabyte cache memory of the third level . The Core i7-4960X and Core i7-4930K are in the more advanced microarchitecture of the cores, . At the same time, at its nominal clock frequency – 3.5 GHz – Core i7-5930K falls in the gap between Core i7-4960X and Core i7-4930K. That is, in terms of performance, it should be similar to the carrier of the Extreme Edition of the previous generation.

Our laboratory was provided with a copy of the Core i7-5930K, and under high load, taking into account the operation of the turbo mode, this processor operates at a frequency of 3.6 GHz.

Note that the power supply used in the six-core Core i7-5930K is significantly higher than that of the older eight-core model. In our case, it was 1.11 V.

The younger processor in the Haswell-E family, the Core i7-5820K, in contrast to its quad-core predecessor Core i7-4820K, is equipped with six cores supporting Hyper-Threading What is the Core i7-5930K – 3.3 GHz. In other words, at a cost of about $ 390, this LGA2011-v3 has a LGA1150-quad core. However, it has a specific drawback – the number of supported PCI Express 3.0 lines in this processor is reduced to 28 pieces, which makes it impossible to use multi-GPU configurations working in the maximum performance mode x16 + x16 together with this chip. Owners of the Core i7-5820K will have the SLA or CrossfireX x16 + x8 mode, which, however, is not a significant flaw for many.

All of the resources to the Haswell-E family have unlocked multipliers, opening the way to their easy overclocking by changing the multiplier. In addition, as with Haswell's LGA1150 processors, overclocking and bus frequency are allowed: the processors support its base values ​​not only 100, but also 125, 167 and 250 MHz. Along the way, the user gets at his disposal and complete freedom in managing memory frequencies and L3-cache.

Externally, the Haswell-E processors look a bit like their LGA2011 predecessors. At least it concerns physical dimensions. However, the shape and thickness of the processor cover changed.

On the left – Core i7-5960X Extreme Edition (LGA 2011-v3), right – Core i7-3970X Extreme Edition (LGA 2011)

On the left – Core i7-5960X Extreme Edition (LGA 2011-v3), right – Core i7-3970X Extreme Edition (LGA 2011)

The processor socket also changed. Although the new LGA2011-v3 connector has the same number of contacts as before, mechanically with older processors it is not compatible. There is no compatibility between Haswell-E and their predecessors and at the electrical level. In Haswell-E, as in LGA 1150 processors, and built-in power converter has appeared, so that "outside" new CPUs receive only one base voltage of 1.8 V. In addition, the DDR4 memory interface is used by new processors differs.

It should be noted that the cooling cooling systems. Older coolers that could be used in the LGA 2011 platform are also compatible with LGA 2011-v3. The only thing that should be kept in mind is the heat release. All LGA 2011 processors, with the exception of the old Core i7-3970X, had a TDP of 130 watts, while the new Haswell-E had a typical heat dissipation of 140 W.

The following table, in which Haswell-E is contrasted to their predecessors Ivy Bridge-E and overclocking Core i7 processors for the LGA 1150 platform, should allow to completely evaluate the characteristics of the novelties:

Core i7-5960X Core i7-5930K Core i7-5820K Core i7-4960X Core i7-4930K Core i7-4820K Core i7-4790K
The code name Haswell-E Haswell-E Haswell-E Ivy Bridge-E Ivy Bridge-E Ivy Bridge-E Devil's Canyon
Nuclei / flows 8/16 6/12 6/12 6/12 6/12 4/8 4/8
Hyper-Threading Technology There are There are There are There are There are There are There are
The clock speed 3.0 GHz 3.5 GHz 3.3 GHz 3.6 GHz 3.4 GHz 3.7 GHz 4.0 GHz
The maximum frequency in the turbo mode 3.5 GHz 3.7 GHz 3.6 GHz 4.0 GHz 3.9 GHz 3.9 GHz 4.4 GHz
The unlocked multiplier There are There are There are There are There are There are There are
TDP 140 W 140 W 140 W 130 W 130 W 130 W 88 W
The lines of PCI Express 3.0 40 40 28 40 40 40 16
HD Graphics No No No No No No 4600
The frequency of the graphic core 1250 MHz
L3-cache 20MB 15 MB 15 MB 15 MB 12 MB 10 MB 8 MB
Memory support 4 channels
DDR4-2133
4 channels
DDR4-2133
4 channels
DDR4-2133
4 channels
DDR3-1866
4 channels
DDR3-1866
4 channels
DDR3-1866
2 channels
DDR3-1600
Extensions of the instruction set AVX2 AVX2 AVX2 AVX AVX AVX AVX 2.0
The packaging LGA 2011-v3 LGA 2011-v3 LGA 2011-v3 LGA 2011 LGA 2011 LGA 2011 LGA 1150
The price of the $ 999 $ 583 $ 389 $ 999 $ 583 $ 323 $ 339

⇡ # DDR4 SDRAM

The second most important innovation in the LGA 2011-v3 platform is the release of the DDR4 SDRAM for the new microarchitecture and the increase in the number of processing cores. Such memory is just beginning its journey to the market, and Haswell-E processors are the first site where this technology will be rolled out. However, we do not think that the appearance of the figure "four" in the name of memory will change something fundamentally. DDR4 memory is a simple evolutionary update of the familiar DDR3 SDRAM, and the Haswell-E's built-in four-channel memory controller is very similar to the controller.

In the DDR4 SDRAM specification adopted by JEDEC, it does not hide the new standard of memory with minimal changes. The cost of developing and implementing new controllers should be small, and this should be the key to a successful and ubiquitous transition to DDR4 SDRAM, which should gain full strength in the middle of next year. However, to say that DDR4-memory is absolutely similar to DDR3, it is still impossible. The new type of memory allows to increase the bandwidth and capacity of the modules, reduce their power consumption, and improve the overall reliability of the memory subsystem.

DDR4 SDRAM received a modified internal organization and thus became better suited for working with high-performance multi-core processors. A typical 8-gigabit DDR4 SDRAM device with a four-bit data interface consists of four groups of banks with four banks in each group. At the same time, there are 2 17 2 rows of 512 bytes in each bank of such a device. A similarly capacious DDR3 device consists of only eight banks, in which 2 16 two-kilobyte lines are used. In other words, there are more banks in DDR4 devices, but lines of much smaller size are used, which allows you to parallelize and quickly process incoming requests.

It is the increase in internal parallelism at the bank level that is the key to the increase in the frequency of DDR4 SDRAM. In fact, the core of any SDRAM-memory, and DDR4 including, continue to operate at a frequency of 100 to 266 MHz, which has not undergone significant changes in the history of SDRAM, starting from the last century. The increase in the frequency of the external interface, which DDR4 can be from 2133 to (in the distant future), 4266 MHz, is realized due to the internal pairwise multiplexing of queries to different groups of banks. This technique is superimposed on the eight-fold pre-selection of data embedded in DDR3 SDRAM, resulting in that each internal data access in DDR4 generates 16 transmissions on the external interface. That is why, in theory, the DDR4 SDRAM frequencies can be up to two times higher than DDR3, however, at the cost of increasing latencies.

Another important advantage of DDR4, in addition to increasing the speed, should be an increase in the volume of modules, the capacity of which can be brought to 128 GB. But among the usual non-registered memory slots, the maximum volume will be limited to 16 GB. At the same time, this option remains unrealized: all the sets of DDR4 SDRAM for Haswell-E come on sale either from 4-gigabyte or from 8-gigabyte modules.

The claimed efficiency of DDR4 SDRAM is achieved mainly due to lowering the supply voltage to 1.2 V. According to preliminary estimates, this allows to reduce the consumption by approximately 35 percent compared to DDR3L SDRAM. However, many DDR4-memory modules aimed at working in overclocking systems will require a higher voltage than specified in the specification – at 1.35 V.

There is no compatibility between DDR3 and DDR4 memory. The new modules, though similar to their predecessors in physical dimensions, have a different number of contacts – 288 pieces – and another location of the mechanical "key" in the connector. The contacts on the module themselves are closer to each other. This is clearly seen in the photo, on which we compared the memory slots in our laboratory from the sets of Corsair Vengeance Pro DDR3 and Corsair Vengeance LPX DDR4.

Top – DDR4 SDRAM module, bottom – DDR3 SDRAM module

At the time of the announcement of the LGA 2011-v3 platform, the main memory manufacturers for enthusiasts prepared their new product lines having nominal modes from DDR4-2133 to DDR4-3000. Although Haswell-E formally supports only DDR4-2133, there are no problems with overclocking the memory interface in the new platform, so the use of high-speed DDR4 modules does not cause any problems.

However, we are interested in comparing the speed of four-channel single-frequency DDR4-2133 and DDR3-2133 in similar systems with Core i7-4960X and Core i7-5960X processors running at 4.0 GHz. For DDR3-2133, the timings 11-11-11-31 typical for such memory were applied, and for DDR4-2133 timings are 15-15-15-35, which currently have an overwhelming majority of such modules.

Ivy Bridge-E, four-channel DDR3-2133, 11-11-11-31

Haswell-E, four-channel DDR4-2133, 15-15-15-35

According to the test results in AIDA64, the new DDR4 is indeed a little slower than DDR3 with the same frequency, which is due to its higher delays. However, a noticeable difference is manifested only in latency and writing speed, in performance on read operations DDR4 almost does not lag behind DDR3. In addition, the four-channel Haswell-E memory controller allows to develop a gigantic throughput that almost doubles this characteristic of the LGA 1150 platform. In other words, LGA 2011-v3 fully confirms its status as an elite high-performance platform, and the transition to DDR4 only slightly worsens the practical latency subsystem of memory, which in real problems is not of decisive importance.

In addition, we should not forget that DDR4 can have a significantly higher frequency than the memory of the previous standard, which can compensate for higher delays. In the following diagrams, we have shown the scalability of the speed of the Haswell-E memory subsystem. In each mode, the timings were set to the most typical values ​​for a given frequency.

It's easy to see that it's already possible to compete fully with DDR3-2133 DDR4-2400, and DDR4-2666 SDRAM is faster than the memory of the old standard. In particular, DDR4-2666 with timings 15-17-17-35 shows no worse latency than DDR3-2133, and it outperforms it by practical bandwidth. Unfortunately, when switching over the 2666-MHz frontier, modern timers begin to increase seriously in modern DDR4, and faster memory types like DDR4-2800 or DDR4-3000 do not provide a significant increase in practical bandwidth, only acting out on latency. Скорость же операций записи памяти у DDR4 оказывается ниже, чем у DDR3, при любых настройках, но это связано не столько с особенностями её архитектуры, сколько с более низкой частотой L3 кэша и контроллера памяти в процессоре Haswell-E. Впрочем, не следует забывать, что всё сказанное относится лишь к DDR4 первого поколения, и с течением времени на рынке наверняка появятся модули DDR4 SDRAM с лучшими характеристиками.

Правда, со стоимостью модулей DDR4 SDRAM всё это стыкуется не слишком гладко. Судя по отпускным ценам на DDR4, установленным производителями, DDR4-2133 первое время будет продаваться примерно на 35-40 процентов дороже распространённой DDR3-памяти с такой же скоростью. Комплекты же DDR4-2666 и более скоростные при этом относятся производителями к числу элитарных продуктов, и их приобретение обойдется в 2-2,5 раза дороже, чем покупка аналогичного объёма DDR3 SDRAM. Впрочем, совершенно очевидно, что DDR4 по мере своего распространения будет дешеветь. Как предсказывают аналитики, к концу следующего года «наценка за новизну» полностью исчезнет, и более прогрессивная память сравняется по цене с DDR3.

⇡#Набор системной логики Intel X99

Помимо внедрения в высокопроизводительные многоядерные процессоры микроархитектуры Haswell и поддержки новой DDR4-памяти, платформа LGA 2011-v3 примечательна и использованием в её составе нового чипсета-концентратора Intel X99 (кодовое имя Wellsburg). Этот набор системной логики пришёл на смену устаревшему X79 и привнёс в старшую интеловскую платформу полную поддержку современных интерфейсов. Причём, по богатству возможностей X99 обошёл даже недавно выпущенный Z97 для процессоров в LGA 1150 исполнении.

В первую очередь следует отметить появление в новом наборе системной логики высокоскоростного контроллера шины USB. Теперь из 14 имеющихся портов этого типа шесть могут работать не только в режиме USB 2.0, но и как USB 3.0, обеспечивая пропускную способность до 5 Гбит/с.

Ещё более значительное обновление затронуло интерфейс SATA. Все имеющиеся порты теперь поддерживают режим 6 Гбит/с, плюс контроллеров, их обслуживающих, стало два. Это позволило увеличить общее число SATA 3.0 портов до 10, и кроме того, обеспечить поддержку новых интерфейсов SATA Express и M.2, для реализации которых используется комбинации из портов SATA 3.0 и линий PCI Express. Для управления дисковой подсистемой Intel внедрил новую версию Rapid Storage Technology 13.1, она позволяет на равных правах работать как с M.2 и Serial Express, так и с SATA-накопителями. Однако следует иметь в виду, что RAID-конфигурации из большего, чем шесть, количества дисков не поддерживаются, а все диски, объединяемые в массив, должны подключаться к одному и тому же контроллеру.

Встроенный в концентратор контроллер PCI Express 2.0 по сравнению с прошлой его версией не изменился и поддерживает всего лишь 8 линий. Однако, учитывая мощность внутрипроцессорного контроллера PCI Express 3.0, этого кажется вполне достаточно. Так, средствами набора системной логики X99 процессорные 40 линий PCI Express 3.0 для поддержки SLI или CrissfireX конфигураций могут делиться на три слота x16, x16 и x8, либо на четыре или пять слотов x8. Правда, последний вариант требует от производителя материнской платы использования дополнительных схемотехнических решений. Что же касается процессора Core i7-5820K с 28 линиями PCI Express 3.0, то у него линии PCI Express могут быть распределены либо на два слота x16 и x8, либо на три слота x8. Естественно, процессорные линии PCI Express 3.0 допускают и более мелкую декомпозицию, поэтому их допустимо подводить к слотам M.2, портам SATA Express и даже к каким-либо дополнительным контроллерам, например Thunderbolt.

Кстати, говоря о X99, Intel обещает, что среди основанных на нём плат появятся и продукты, поддерживающие интерфейс Thunderbolt 2 с пропускной способностью до 20 Гбит/с. А это — в четыре раза более высокая скорость, чем обеспечивается шиной USB 3.0. Кроме того, Thunderbolt 2 предлагает широкий спектр разнообразных возможностей, включая формирование «цепочек» из устройств, подключение 4K-дисплеев по протоколу DisplayPort 1.2, соединение двух компьютеров и прочее.

Из других нововведений следует отметить, что в новом чипсете, также как и в наборе логики Z97 для LGA 1150 систем, появился встроенный генератор тактовой частоты. И это избавляет производителей системных плат от необходимости использования дополнительных электронных компонентов и упрощает дизайн платформы.

Набор системной логики Intel X99 имеет расчётное тепловыделение на уровне 6,5 Вт, то есть, он экономичнее, чем X79. Следовательно, как и раньше, LGA 2011-v3 материнские платы наверняка смогут обходиться пассивным охлаждением чипсета.

К выходу процессоров Haswell-E ведущие производители материнских плат подготовили большой набор разнообразных LGA 2011-v3 платформ. Например, при подготовке этого обзора мы сумели познакомиться с материнской платой ASUS X99 Deluxe.

Эта флагманская платформа для процессоров Haswell-E одного из ведущих производителей материнских плат может похвастать наличием восьми слотов DDR4 DIMM и пятью слотами PCI Express 3.0 x16. Благодаря этому ASUS X99 Deluxe поддерживает до 64 Гбайт четырёхканальной DDR4 SDRAM со скоростью вплоть до DDR4-3000. Слоты же PCI Express дают возможность собирать сколь угодно сложные графические конфигурации, включающие до трёх GPU, работающих в режиме SLI или CrossfireX. Процессорные 40 линий PCI Express при этом гибко распределяются по слотам в зависимости от конкретных нужд пользователя. Кроме того, на плате есть и один дополнительный слот PCI Express 2.0 x4, работающий через чипсет.

На плате в общей сложности имеется 12 портов SATA 6 Гбит/с (к чипсетным 10 добавляется пара портов, реализованных через контроллер ASM106SE), четыре из которых собраны в два порта SATA Express. Кроме того, реализован и слот M.2, причём на него заведены 4 процессорные линии PCI Express 3.0, что обеспечивает поддержку наиболее скоростных вариантов твердотельных накопителей соответствующего форм-фактора.

Но и это ещё не всё. В комплект поставки с ASUS X99 Deluxe входит дополнительный PCI Express → M.2 переходник, который позволяет установить в систему и второй M.2-накопитель.

Общее число портов USB 3.0 доведено до 14, для этого к чипсетному контроллеру добавлено два дополнительных, разработки ASMedia. Также на ASUS X99 Deluxe есть два гигабитных сетевых интерфейса и модуль Wi-Fi 802.11 a/b/g/n/ac. Интегрированный звук построен на базе качественного кодека ALC1150, причём в звуковом тракте применена новая схема с качественными компонентами, экранированием и разводкой каналов по разным слоям материнской платы.

Процессорный разъём запитан по цифровой 8-канальной схеме с 9-ступенчатым противодействием падению напряжения под нагрузкой, память же получает питание от двух двухканальных стабилизаторов. Отдельно следует отметить, что схема питания по сравнению с обычными платами усилена, так как потребление восьмиядерных Haswell-E при экстремальном разгоне может легко переваливать за 1 кВт. Для подключения вентиляторов на ASUS X99 Deluxe предусматривается шесть коннекторов: четыре — системных и два — для кулера CPU. В дополнение к этому в комплект поставки входит специальный модуль, позволяющий подсоединить ещё три дополнительных вентилятора.

Разгон системы и конфигурирование схемы питания осуществляется через UEFI BIOS с новым «графитовым» интерфейсом, а кроме того, всё это может быть выполнено и из среды операционной системы посредством фирменного комплекса программного обеспечения.

Примерно такой же набор возможностей предлагают и флагманские LGA 2011-v3 материнские платы других производителей, с обзорами которых вы, несомненно, сможете ознакомиться на нашем сайте позднее.

В целом же можно констатировать, что материнки, основанные на новом наборе логики X99, выглядят существенно лучше своих предшественниц в части возможностей подключения внешних устройств и носителей информации. И это закономерно, так как основное совершенствование набора логики для Haswell-E произошло именно в этом направлении.

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

Tags: , , ,
Previous Post
new-kaveri-cheaper-and-more-economical-processors-and-memory.jpg
Uncategorized

new Kaveri – cheaper and more economical / processors and memory

Next Post
to-remember-processors-and-memory.png
Uncategorized

to remember … / Processors and memory

Leave a Reply

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