is there life on the AMD platform? / Processors and memory

Such a topic as the importance of CPU speed for modern games, we last touched last May. Then our sample of candidates for the role of a game processor was limited to Intel chips, and we left them for AMD, because, as it seemed to us, they are of little interest for the construction of a gaming PC.

The processor division of Advanced Micro Devices has completely disappeared from the competition with Intel in the segment of high-performance CPUs for home PCs and workstations (well, or, with more optimistic intonations, retreated to regroup to return with the Zen architecture). Even reviews of more affordable processors with a built-in graphics core, which seem to be to the liking of an economical buyer, we have to end time and again with a lot of reservations, despite the attractiveness of the concept itself and the successful technical solutions embodied in the “red” APU.

 AMD A10-7870K "height =" 534 "width =" 800 "/>

<p> However, what we learned about the processor-dependent games and GPUs last time on the example of the Intel processor line gave rise to certain hopes for AMD chips, primarily for the older models of the FX family for the AM3 + platform. A rather unexpected discovery was that the games most demanding for the speed of the central processor, in the first place, need a certain number of x86 cores, and clock speeds are not decisive for quad-core processors. Some of the dual-core chips, in turn, quite successfully compensated for their limitations due to a) Hyper-threading technology; b) high clock speeds. </p>
<p> As you know, the cornerstone of AMD's marketing strategy for the Bulldozer / Piledriver / Steamroller microarchitecture processors is the larger number of cores that CPU data can offer to the user, compared to similar Intel products. True, in the strict sense, two-, four-, six- and eight-core AMD crystals are such only in integer calculations. Each so-called module in this architecture has only one block of floating-point operations per pair of integer execution clusters. With this load, you can consider the AMD module as an analog of the single-core Intel processor equipped with Hyper-threading technology. </p>
<p> In itself, this feature of AMD processors does not yet put a cross on them, as, as we already found out earlier, games do not actually require more than four full-bodied x86 cores. And the senior representatives of the FX line are just quad-core CPUs even on a strict, Intel-based account. Also, as the tests of Intel chips showed, simultaneous multi-threading (SMT) is a great help for dual core Core i3, which at high clock speeds compete with younger representatives of the Core i5 family. Hence, there is hope for AMD's dual-module processors (which include some FX models and most APUs of the A8 / A10 family). </p>
<div align=

 AMD FX-9590 "height =" 534 "width =" 800 "/>

<p> In general, in the architecture of AMD processors in theory there are no signs that forbid them to have sufficient performance to match the demands of a powerful discrete GPU. </p>
<p> The main defect in the development of the Bulldozer / Piledriver / Steamroller family is the relatively low performance per clock, even in tasks that are favorable for these chips. On the other hand, AMD products boast rather high clock speeds crossing the 4 GHz boundary, provided that the nature of the computational load and the number of cores involved allow the power to be kept in the target range for a particular chip. Which of the two factors will be stronger in a particular task – modern games? </p>
<h2 class= ⇡ # The results of the first part of “Processorependence”

In the last article, we posed the following questions, which made it possible to narrow the scope of the study – otherwise colossal – to several convenient and demonstrative tests:

  1. How modern games are sensitive to CPU performance?
  2. At what frame rate (and, correspondingly, at what GPUs) does processor independence manifest itself?

Recall what we got in the end. First, of the nine AAA games we used, some do not rely very much on CPU performance (Alien: Isolation, Far Cry 4, Tomb Raider). Others depend on this parameter to an enormous extent (Thief, Company of Heroes 2, Metro: Last Light).

However, even in the latter category, the speed of the CPU is not a factor directly determining the frame rate. The value has a balance between CPU, GPU and graphics settings in the game. It is easy to focus on the following empirically obtained rule: if a pair with a sufficiently powerful CPU GPU is able to provide 50-60 frames per second and higher with these graphics quality settings (see the table with test descriptions), then the CPU performance matters. Choosing the components of the PC for iron reviews, it is better to repulse it from the GPU, as we test the video card on a platform that obviously overlaps their needs in a powerful CPU.

As shown by more detailed analysis, three of the nine mentioned games (Battlefield 4, Thief, Metro: Last Light) first of all require the presence of four cores of the CPU, and to the frequency at which they work, they are practically indifferent. From a practical point of view, this reduces the choice to absolutely any kind of Core i5. Neither the Hyper-threading quad-Core i7 quad-core, nor the six- and eight-core CPUs for the LGA2011 platform in these games have any practical advantages. And thanks to the support of Hyper-threading, a good game processor can be obtained even from dual-core Core i3.

⇡ # Participants of the testing

In the first part of the article, the test subject was representatives of the Haswell Refresh line for the LGA1150 connector and Haswell-E processors for the LGA2011-v3 platform. It was about nine months ago – formally a considerable time for the computer industry, however, in the aspect of interest to us (gaming performance in a pair with a discrete GPU), the next two iterations of the Intel architecture – Broadwell and Skylake – advanced little. And six-and eight-core desktop-class CPUs are still based on the Haswell core, Intel has not yet introduced anything new in this category.

This means that the data obtained, with which we will then compare the results of AMD’s “stones,” are quite relevant in modern conditions and can be extrapolated to the Skylake lineup – adjusted for a slightly different frequency grid of the latter.

The number of threads The size of the cache memory L3, MByte Base frequency, GHz Max. Frequency of the Turbo, GHz Operative memory
LGA2011-v3 Core i7-5960X 8 16 20 3.0 3,5 4 × DDR4 SDRAM, 2133 MHz
Core i7-5830K 6 12 15 3,5 3.7
Core i7-5820K 3.3 3.6
LGA 1150 Core i7-4790K 4 8 8 4.0 4.4 2 × DDR3 SDRAM, 1600 MHz
Core i7-4790 3.6 4.0
Core i7-4790S 3,2 4.0
Core i7-4790T 2.7 3.9
Core i7-4785T 2.2 3,2
Core i5-4690K 4 4 6 3,5 3.9
Core i5-4690 3,5 3.9
Core i5-4690S 3,2 3.9
Core i5-4590 3.3 3.7
Core i5-4590S 3.0 3.7
Core i5-4690T 2.5 3,5
Core i5-4460 3,2 3,4
Core i5-4460S 2.9 3,4
Core i5-4590T 2.0 3.0
Core i5-4460T 1.9 2.7
Core i3-4370 2 4 4 3.8
Core i3-4360 3.7
Core i3-4350 3.6
Core i3-4360T 3,2
Core i3-4350T 3.1
Core i3-4340TE 2.6
Core i3-4160 2 4 3 3.6
Core i3-4150 3,5
Core i3-4160T 3.1
Core i3-4150T 3.0
Pentium G3460 2 2 3 3,5 2 × DDR3 SDRAM, 1600 MHz
Pentium G3450 3,4
Pentium G3440 3.3
Pentium G3258 3,2 2 × DDR3 SDRAM, 1333 MHz
Pentium G3250 3,2
Pentium G3240 3.1
Pentium G3450T 2.9 2 × DDR3 SDRAM, 1600 MHz
Pentium G3440T 2.8
Pentium G3250T 2.8 2 × DDR3 SDRAM, 1333 MHz
Pentium G3240T 2.7
Celeron G1850 2 2 2 2.9 2 × DDR3 SDRAM, 1333 MHz
Celeron G1840 2.8
Celeron G1840T 2.5

Recall how we selected the test participants from the Intel team and with what settings they were tested. From each group we took either an older model, whose frequency could be lowered below the nominal one, or one of the younger ones, having an unlocked multiplier (which, if necessary, was accelerated). In the table, these CPUs are in bold.

The three younger Intel series on Haswell core do not have Turbo Boost technology and work under constant load at a constant frequency, which allows one processor to accurately simulate the performance of all the others in its group. However, the performance of Core i5 and i7 chips equipped with Turbo Boost technology can not be accurately displayed with the help of the older model in each family, reducing its frequency, because the base frequency multiplier, unlike the maximum, is not adjustable. For this reason, we tested the top chips on the top Turbo-frequency of the respective models. Benefit in practice Turbo Boost acts very aggressively, and, except for a heavy multithreaded load, the Intel CPU for the most part operates at frequencies close to its upper limit.

The clock speed, GHz

The clock speed, GHz

The Celeron G1850

2,5 2.8 2.9
The Pentium G3258 The clock speed, GHz 2,7 2.8 2.9 3.1 3,2 3.3 3,4 3,5 Core i3-4360 The clock speed, GHz 2,6 3,1 3,2 3,6 3.7 3.8 Core i5-4690K The clock speed, GHz 2,7 3.0 3,4 3,5 3,7 3.9 The clock speed, GHz

The clock speed, GHz

3,2 3.9 4 4.4 The clock speed, GHz

The clock speed, GHz

3,6 3.7 Core i7-5960X The clock speed, GHz 3,5

Since Intel offers a huge variety of CPUs in Haswell’s desktop line, we have reduced the array of Turbo-frequency values, eliminating the coinciding positions or those that are only 100 MHz away from neighboring ones. The table above shows the frequencies available to each core according to Intel specifications: tests were performed on selected frequencies.

Note. We could not get a chip of the Core i3-41XX series (from i3-43XX featuring a L3 cache size) for tests, and the Pentium G3258 processor, formally “unlocked”, refused to overclock with a multiplier on the SABERTOOTH Z97 test platform MARK 1, therefore frequencies above 3.2 GHz on this core have not been tried.

AMD processors selected for the second part of “Processor Dependence” are members of the Caveri / Godavari and Vishera families. The first is more correct to call APU in accordance with the manufacturer’s nomenclature, and it’s no accident: the lion’s share of the crystal area is occupied by an integrated graphics engine, and the central processor is equipped with two or one module of the micro-architecture Steamroller. The more modern and probably the last iteration of this architecture – Excavator (the core of Carrizo) – is reserved for mobile AMD chips. Despite all the variety of APU models, from the CPU point of view the Caveri / Godavari line is described by only four values ​​of the upper frequency. The remaining differences between the chips are mostly reduced to the configuration of integrated graphics.

The Vishera core is the pinnacle of the development of the AMD FX line in its current form, AMD has been focusing its efforts on hybrid processors for the time being (fans admiring Zen architecture brand name), desktop CPUs with two or four modules and without built-in GPU stopped on the Piledriver architecture, the successors of which are Steamroller and Excavator. Accordingly, AMD FX chips are deprived of the later optimizations of the x86 pipeline, not to mention other signs of platform obsolescence. In particular, FX does not support the PCIe bus version 3.0.

The CPU connector

The model The number of modules (integer cores) The number of threads memory L3, Mbyte Base frequency, GHz Max. Frequency of the Turbo, GHz Operative memory
AM3 + FX-9590 4 (8) 8 4 x 2 4.7 5 2 × DDR3 SDRAM, 1866 MHz
FX-9370 4.4 4.7
FX-8370 4 4.3
FX-8350 4 4.2
FX-8320 3,5 4
FX-8310 3,4 4.3
FX-8300 3.3 4.2
FX-8370E 3.3 4.3
FX-8320E 3,2 4
FX-6350 3 (6) 6 3 x 2 3.9 4.2
FX-6300 3,5 4.1
FX-4350 2 (4) 4 2 x 2 4.2 4.3
FX-4320 4 4.2
FX-4300 3.8 4
FM2 + A10-7870K 2 (4) 4 2 × 2 3.9 4.1 2 × DDR3 SDRAM, 2133 MHz
A10-7860K 3.6 4
A10 PRO-7850B 3.7 4
A10 PRO-7800B 3,5 3.9
A10-7700K 3,4 3.8
A8-7670K 3.6 3.9
A8-7650K 3.3 3.8
A8 PRO-7600B 3.1 3.8
A6-7470K 1 (2) 2 1 3,7 4
A6 PRO-7400B 3,5 3,9 2 × DDR3 SDRAM, 1866 МГц
A4 PRO-7350B 3,4 3,8
Athlon X4 870K 2(4) 4 3,9 4,1 2 × DDR3 SDRAM, 1866 МГц
Athlon X4 860K 3,7 4
Athlon X4 840 3,1 3,8
Athlon X2 450 1(2) 2 3,5 3,9

Тестирование процессоров AMD мы проводили по методике, отработанной на чипах Intel. Были выбраны четыре чипа, венчающие каждую из представленных категорий CPU. Тактовая частота каждого из них была зафиксирована и приводилась в соответствие с верхней границей частоты, свойственной интересующим моделям в соответствующей линейке. Пункты, отстоящие от соседних на 100 МГц, были отброшены (за исключением A10-7870K, который был случайно протестирован на 4,0 и 4,1 ГГц – лишние данные решили сохранить).

Мы, впрочем, так и не смогли запустить чип FX-9590 на частоте 5 ГГц по причине его неуемного энергопотребления. В этом плане еще были какие-то надежды на башенный кулер Thermalright Archon, но материнская плата AsRock Fatal1ty 990FX Professional не смогла обеспечить требуемое напряжение питания CPU, в результате чего нам пришлось удовлетвориться частотой 4,7 ГГц.

Планы по тестированию бюджетных APU семейства Kabini были зарублены на корню потому, что платформа Socket AM1 поддерживает только слот PCIe 2.0 x4 для подключения дискретного GPU. Кроме того, частоты в диапазоне 1,3-2,2 ГГц и так не позволят этим чипам конкурировать с чем-либо более мощным, чем Intel Atom.

Тактовая частота, ГГц 3,8 3,9 4
Тактовая частота, ГГц 3,8 3,9 4 4,1
Тактовая частота, ГГц 4 4,2 4,3
Тактовая частота, ГГц 3,5 3,9
Тактовая частота, ГГц 4 4,2 4,3 4,7 5

⇡#Тестовые стенды

Конфигурация тестовых стендов
Разъем процессора LGA1150 LGA2011-v3 Socket FM2+ Socket AM3+
Материнская плата ASUS SABERTOOTH Z97 MARK 1 ASUS RAMPAGE V EXTREME ASUS A88X-PRO AsRock Fatal1ty 990FX Professional
Оперативная память AMD Radeon R9 Gamer Series, 1333/1600 МГц, 2 × 8 Гбайт Corsair Vengeance LPX, 2133 МГц, 4 × 4 Гбайт AMD Radeon R9 Gamer Series, 1866/2133 МГц, 2 × 8 Гбайт AMD Radeon R9 Gamer Series, 1866 МГц, 2 × 8 Гбайт
ПЗУ Intel SSD 520 240 Гбайт + Crucial M550 512 Гбайт
Графическая карта NVIDIA GeForce GTX 980
Блок питания Corsair AX1200i, 1200 Вт
Охлаждение CPU Thermalright Archon
Корпус CoolerMaster Test Bench V1.0
Операционная система Windows 8.1 Pro X64

AsRock Fatal1ty 990FX Professional



В предыдущей части статьи мы выбрали четыре игры из тестового пакета, который используем для тестирования видеоадаптеров, наиболее чувствительные к производительности центрального процессора: Battlefield 4, Thief, Company of Heroes 2 и Metro: Last Light. Разница в результатах, полученных на Celeron и Core i7, в них варьируется от 47 до 107 %. К слову, как мы еще увидим, одномодульные APU производства AMD в этом тесте опустятся даже ниже, чем двухъядерный Celeron на ядре Haswell.

Настройки игр были выбраны с таким расчетом, чтобы при довольно мощном тестовом GPU (GeForce GTX 980) частота смены кадров при разрешении 1920 × 1080 оказалась в диапазоне 60-80 FPS, а при относительно слабом (GeForce GTX 660) не опустилась ниже 30 FPS.

В целях совместимости с результатами, полученными в предыдущей части «Процессорозависимости», платформа AMD была протестирована с драйверами NVIDIA, актуальными на тот момент.

Бенчмарки: игры
Программа Настройки Полноэкранное сглаживание Разрешение
Metro: Last Light, встроенный бенчмарк Макс. качество Нет 1920 × 1080
Company of Heroes 2, встроенный бенчмарк Макс. качество Нет
Battlefield 4 + FRAPS Макс. качество. Начало миссии Tashgar MSAA 4x + FXAA
Thief, встроенный бенчмарк Макс. качество SSAA 4x + FXAA

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

What's Your Reaction?

hate hate
confused confused
fail fail
fun fun

Comments 0

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

is there life on the AMD platform? / Processors and memory

log in

reset password

Back to
log in
Choose A Format
GIF format