AMD FX-8320E in the test

After we recently examined AMD's new FX-8370E processor, today we take a look at the FX-8320E, which is located in the middle of the AMD desktop processors. 8 cores, each with a clock rate of 3,2 GHz, turbo mode, but also a lower TDP of 95 watts are the key data. However, this processor is available for prices starting at 140 euros. Our test explains how the FX-8320E performs in practice.

Intro

AMD tries to squeeze as much performance as possible out of the Vishera processors so that you don't lose touch with Intel in the performance segment. While the APUs (CPU and graphics unit combined) are generally well positioned, they lag behind the competition in the actual processor business. AMD's high-end models of the FX series are a prime example of the approach mentioned. With the FX-9000 models, the maximum possible performance has been teased out of the architecture. However, this is at the expense of the power consumption and leads to a TDP of 220 watts. A value that Intel does not achieve even on the LGA 2011 socket.

With lower TDPs, AMD has responded to the charge of excessive power consumption and recently put the FX-8370E into the race. Now we have the FX-8320E, which also only joins the 95 watt class and tries to re-represent the midfield among AMD's desktop processors.

AMD's motto: You can build high-performance PCs at low prices, and advertise primarily with the eight processor cores, from which very well parallelized software can benefit massively. In addition, the 8320E has a base clock of 3,2 GHz and a turbo clock of around 4 GHz. For comparison: the non-E version clocks at 3,5 GHz in the basic clock and also 4 GHz in turbo mode.

On the following pages we clarify where the FX-8320E is to be classified, and discuss the strengths and weaknesses of the relatively inexpensive CPU.

Test environment

Hardware: Intel systems

Intel socket LGA-1150

• Intel Core i7-4790K:
  Haswell architecture, C0 stepping, 4,0 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Intel Core i7-4770:
  Haswell architecture, C0 stepping, 3,5 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Intel Core i5-4670K:
  Haswell architecture, C0 stepping, 3,4 GHz, 4 cores, turbo mode active, 4 x DDR3-1600
• Intel Core i5-4670:
  Haswell architecture, C0 stepping, 3,4 GHz, 4 cores, turbo mode active, 4 x DDR3-1600

The new Haswell processors are operated on the MSI mainboard Z87-G43. We imported the latest beta BIOS V1.2B1 and activated all energy-saving mechanisms in the BIOS. The memory modules used come from G.Skill. It is a 4 x 4 GByte kit of the Ripjaws Z DDR3-1600 with the latencies CL9-9-9-24.

Intel socket LGA-1155

• Core i7 3770K:
  IB architecture, E1 stepping, 3,5 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Intel Core i5-3570K:
  IB architecture, E1 stepping, 3,4 GHz, 4 cores, turbo mode active, 4 x DDR3-1600
• Intel Core i7-2700K:
  Self-service architecture, D2 stepping, 3,5 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Core i7 2600K:
  Self-service architecture, D2 stepping, 3,4 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Core i5 2500K:
  SB architecture, D2 stepping, 3,3 GHz, 4 cores, turbo mode active, 4 x DDR3-1333

For Intel's “Sandy Bridge” and “Ivy Bridge” processors for the LGA-1155 socket, 4 x 4 GB G.Skill Ripjaws Z DDR3-1600 are used, operated on DDR3-1333. The memories are operated with the CL9-9-9-24 2T latencies. That comes as a mainboard MSI Z77A-GD65* used with BIOS version 7751vP0. All energy saving mechanisms are activated in the BIOS.

MSI Z87-G43Intel socket LGA-2011 and LGA-2011-3

• Core i7-5960X
  R2 stepping, 3,0 GHz, 8 cores, turbo mode active, HTT active, 4 x DDR4-2133
• Core i7-4960X
  S1 stepping, 3,6 GHz, 6 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Core i7-3960X
  C2 stepping, 3,3 GHz, 6 cores, turbo mode active, HTT active, 4 x DDR3-1600

For the processors of the LGA-2011 socket, 4 x 4 GB G.Skill Ripjaws Z DDR3-1600 and an ASUS Rampage IV Gene with BIOS 4901 are used. The socket LGA-2011-3 is measured with 4 x 4 GByte Corsair Vengeance LPX DDR4-2666, operated with DDR4-2133 and timings of 15-15-15-36. An MSI X99S Gaming 7 with BIOS V17.4 is used as the mainboard.

Hardware: AMD systems

AMD Socket FM2 +

• A10-7850K
  Steamroller architecture, A1 stepping, 3,7 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A8-7600
  Steamroller architecture, A1 stepping, 3,3 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A10-6800K
  Piledriver architecture, A1 stepping, 4,1 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A10-6700
  Piledriver architecture, A1 stepping, 3,7 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A10-6500T
  Piledriver architecture, A1 stepping, 2,1 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600

The processors for FM2 and FM2 + were measured on the MSI A85XA-G65. The Kaveri models and the A10-6800K also on the MSI A88XM-E45.

AMD Socket AM3 +

• FX-8370E:
  Piledriver architecture, C0 stepping, 3,3 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• FX-8350:
  Piledriver architecture, C0 stepping, 4,0 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• FX-8150:
  Bulldozer architecture, B2 stepping, 3,6 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600

The same G.Skill modules are used for AMD's Bulldozer processors for the AM3 + socket as on the Intel systems. That comes as a mainboard ASUS Crosshair V Formula* (990FX chipset) used with BIOS 1703. All energy saving mechanisms are activated in the BIOS.

ASUS Sabertooth 990FXDue to the lack of support for the new processor from our test board ASUS Crosshair V Formula, we had to change the board. The measurement results for the AM3 + socket were therefore completely created on the new ASUS Sabertooth 990FX.

AMD Socket FM2

• A10-5800K

Trinity architecture, A1 stepping, 3,8 GHz, 4 cores, 4 x DDR3-1600

The G.Skill DDR3 memory mentioned above works here as well. The Gigabyte GA-F2A85X-UP4 with BIOS F4 is used as the mainboard. All energy saving mechanisms are activated in the BIOS.

More hardware

Graphic card:

• MSI Radeon HD 7970 Lightning
• AMD Radeon HD 3450 (DDR3):
  only for measurements of power consumption

Memory:

• 16 GB (4 x 4 GB) G.Skill Ripjaws Z DDR3-1600
  SPD operation: DDR3-1600, 9-9-9-24 at 1,5 volts

Power adapter:

• be quiet! DARK POWER 550W R10

Hard disk:

• Seagate ST2000VX000

Cooler:

• Scythe Katana III
• Noctua NH-U12S

Measurement:

• Tenma 72-6185 clamp ammeter
• Voltcraft Energy Check 3000
• Tenma infrared thermometer 72-820
• Metex multimeter M-3640D
• Profitec KD 302 energy cost meter

Software and benchmarks

Operating system and driver

• Windows Ultimate 7 (64-bit) including all updates up to May 2013
• AMD / ATI Catalyst 13.4 WHQL
• Windows 7 supplied drivers for the chipset and network card

CPU benchmarks

• Synthetic benchmarks
  • PCMark05 v120_1901
  • PCMark 7 v1.4.0
  • Euler 3D
• Audio editing
  • iTunes 11.0.2
  • LAM 3.98.4 (compiled with Visual Studio 2008)
  • Nero AAC Encoder 1.5.1
  • OggEnc 2.87
• image editing
  • Gimp 2.8.4
  • Irfan View 4.35
• video editing
  • HandBrake 0.9.9 RC1 64bit
  • Avisynth 2.58 and x264 encoder 0.132.2310
• Packer
  • 7Zip 9.20 (64 bit)
  • WinRAR 4.20 (64 bit)
  • WinZip 17.5 Build 10480g (64bit)
• Rendering
  • Blender 2.67 (64 bit)
  • Cinebench 11.529 (64 bit)
  • Pov-Ray 3.7 RC7 (64 bit)
• Games
  • Assassin's Creed III* V. 1.0.2 (DirectX 11 - Savegame)
  • Crysis 3* (DirectX 11 - savegame)
  • Serious Sam 3* V. 1.5.1466 (DirectX 9 - Savegame)
  • The Elder Scrolls V: Skyrim* V. 1.8.151.0.7 (DirectX 9 - Savegame)
  • tomb raider* V. 743.0 (DirectX 11 - Savegame)
• Encryption
  • 7-Zip
  • TrueCrypt 7.1a
  • WinZip

SiSoft Sandra 2013.05.19.44

Overall, we tried to focus strongly on real applications with the benchmarks and to distance ourselves from synthetic tests. Whenever possible, we also used the 64-bit version.

A special remark should be made about LAME: By default, LAME is created with an Intel C ++ compiler. In the past, this has repeatedly caused irritation, which is why we also use a version in our new benchmark course that we created ourselves with the help of Visual Studio 2010 from Microsoft. In terms of performance, however, it does no difference.

Other tools

• CPU-Z
• CPU-Z Latency Tool
• CoreTemp
• Core2MaxPerf
• Fraps

Test methodology

Apart from the remarks already made on this and on the previous page regarding our test philosophy, we want to briefly summarize the essential points again. Unless otherwise stated in the direct test description, the following points always apply:

• All available energy saving mechanisms are activated.
• If the CPU has a turbo mode, this is activated.
• If the CPU supports hyperthreading / core multithreading (CMT), this is activated.
• If not mentioned otherwise, always comes MSI Radeon HD 7970 Lightning for use.

Technology

The new FX-8320E

In the technical sector there is of course nothing new to report today. The technical implementation is basically the same as it was when the Bulldozer in late 2011 by AMD, of course including the advantages of the current Piledriver cores.

AMD's current innovations - shown with the FX-8370E, as now also with the FX-8320E - are more of a “cosmetic” nature. The manufacturer simply reacts to the allegations that the AMD CPUs in the middle class segment are too power hungry and thus tries to present a more attractive TDP class. The two new E-models are therefore in the 95-watt TDP range and no longer in the 125-watt class, like the 8370 and 8320.

The implementation takes place in simple steps, namely in the form of clock and voltage lowering - presumably also the DIE selection. The FX-8320E only has a base clock of 3,2 instead of 3,5 GHz. In turbo mode, however, it still clocks up 4 GHz, just like the previous FX-8320.

In addition, the FX-8000 series remains with models with four modules, whereby each of the modules can handle two threads and AMD therefore speaks of eight CPU cores.

Tabular comparison

In the middle class, the FX-8150 and FX-8350 are practically yesterday's news, but both are still available in stores. The actual offers are represented today by the models FX-8320 and FX-8370 and their E versions.

The nomenclatures are again not always conclusive, because an FX-8320 (E) offers less clock rate than, for example, an FX-8150 and should theoretically not have a higher number. With the FX-8370 (E) you can at least outperform an FX-8350 in terms of turbo clock.

TDP and price as weapons

Over the past few years, since the introduction of the FX-8000 models, it seems to have become clear that you can no longer win a flower pot with high GHz clock speeds alone. Especially if the high clock rates in your own architecture are clearly inferior to the lower clock rates of the competition. And so the new motto is quite obvious: less is more!

And in fact, AMD is currently losing less with the E variants than it is winning. The lower clock rates are not clearly noticeable in the performance, because the turbo clock often bridges here. At the same time, however, you can now name the lower TDP class of 95 watts.

If you look at the market in the area of ​​the FX-8000 series and the AM3 + socket, the FX-8320 (E) in German-speaking countries is probably the cheapest option to get started here. The current staggering also shows the low margins with which the manufacturer has to act here. Often individual models are simply separated by only 10 euros - outliers in our table can be explained by discontinued models in sales or fluctuating exchange rates.

We do not know how much AMD still has to select a good three years after the introduction of the FX-8000 processors in order to receive E and non-E models. But the production should really be mature enough by now that enough high-quality models can be captured, which can be operated with lower voltages in order to be sold as e-models. In this area - with four modules - the 65-watt class remains simply utopian.

Practice

overclocking

Despite the goal of being able to achieve a lower TDP, AMD also offers this FX processor without a multiplier blockade, and of course the end customer is free - at their own risk - to tease out the maximum performance from the CPU. This is also relatively easy using the free multiplier. However, you have to struggle with a few restrictions.

If you play on the CPU's multiplier, the processor's idle rate is no longer lowered as much as it would without manual intervention. This is of course counterproductive when it comes to energy efficiency. When it comes to bare clock rates and performance at the lowest possible price, deactivating the power-saving mechanisms and increasing the voltage in overclocking can even have an effect on an FX-8000 model. After that you don't have any savings in terms of power consumption, but you can get the maximum performance from the CPU squeeze out.

In our test, however, we do not want to deal with the intricacies of overclocking FX processors, so that we do not resort to means such as water cooling or additional voltage increases. Instead, we use our usual air cooling and just turn the clock up above the multiplier.

We were able to increase the clock rate by a further 600 MHz at the base clock rate before the chip lost its stability. This shows that there is definitely potential in the FX-8320E, but again it is not really a surprise compared to the regular FX-8320.

With the increase in clock speed, the power consumption also increases and reaches a value of over 196 watts. Compared to the previously measured 150 watts, this is a significant increase. The increase in clock speed also ensures that values ​​were around 20 watts higher in idle mode.

Performance index [OC]

Games [dGPU]

The increased performance is quite impressive, because the average across all benchmarks is around seven percent. Here everyone has to decide for themselves whether they want to accept the declared disadvantages.

But the means of the benchmarks is of course not the panacea for everyone. As our breakdown shows, the differences lie in the details and so the overclocking measure can hardly bear fruit in one case, but it may increase by up to 18 percent in another. This means of our tool makes it easier to judge personally where one applies one's own standards - including all possible negative consequences.

Direct comparison

Practice: power consumption

In the following, we determine the average consumption of the entire system without a monitor. A standard energy cost meter is used here, in our case an Energy Check 300. Over a period of 20 minutes, we record the course and, of course, state the average value in watts. We use Core2MaxPerf for all processors as a full load scenario.

It remains clear that this measurement of the entire system cannot, of course, be as accurate as previous measurements on our part, in which we only reduced the CPU load and power consumption with special mainboard modifications. Unfortunately, everything flows into such measurements. Sudden spikes caused by hard disk access, programs starting in the background that request higher CPU loads, or similar scenarios. One can only try at this point to exclude all evildoers. But it can never succeed in the end, and so the following diagrams are only indicative - always based on our selected test system!

Based on our explanations, two circumstances clearly stand out in idle mode. On the one hand, the FX-8000 models are basically on the same level, but unfortunately none of the candidates can really score compared to the ranking. The next smaller level are the AMD APUs, which also work in roughly the same segment. Another fundamental limitation of such a consideration of the overall system is the selected components, such as the mainboard or the power supply.

However, the clear lead of Intel offshoots, which operate on different motherboards to AMD, but are also equipped with identical components such as power supply unit, hard drive and so on, cannot be disputed.

And so the Intel processors are clearly ahead in this comparison.

A look at the load scenario clearly shows that AMD is right to separate the 125-watt TDP class from the 95-watt class. The measurements of the total system power consumption show a difference in the range of 30 watts, in some cases even more. It's nice to see, but unfortunately it's still a bit too high. After all, over 30 watts separate this mid-range AMD CPU from Intel's Socket 1150 top model i7-4790K.

In principle, this only gives a rough indication, because at this point the dependency of the load tool is also decisive. This tendency is evident in the tool we have chosen. Intel's top model i7-4790K operates with a TDP of 88 watts, AMD's mid-range model operates with a TDP of 95 watts - roughly the same region. How much leeway the two different CPU models have for maximum TDP under typical load scenarios is unfortunately unclear. It seems that the AMD models are approaching their limits faster here.

And so in the end there is only one look at the benchmarks, which have to clarify where exactly which CPU is.

Benchmarks: Synthetic

PCMark 05

Information on the benchmark

Information on the benchmarkAs the name suggests, PCMark 05 dates from 2005 and is therefore already several years under its belt. Nevertheless, the benchmark suite from the Finnish company Futuremark is still very well suited to classifying the computing power of processors and their memory performance. We only use the CPU and memory suite, so that conclusions can only be drawn for these components. The CPU suite is based on eight different tests from the fields of packing / unpacking, encryption, and audio and video processing, and thus allows an overview of the everyday performance of the processors. The CPU suite benefits equally from high clock frequencies and from multiple cores. In the memory suite that is also used, however, factors such as cache rate, cache size and memory bandwidth play a role, since the tests essentially consist of reading, copying and writing data of different sizes to the memory.

PCMark 7

Information on the benchmark

Information on the benchmarkPCMark 7 was only launched this year and represents the latest system benchmark from Futuremark. We only use the Computation Suite to draw conclusions about the computing power of the processors tested. The suite comprises three different tests from the fields of video transcoding and image processing. It allows you to see the everyday performance of the processors. In addition to a high clock rate, the tests benefit from multiple cores.

Euler3d benchmark

Essentially, it is a CFD (Computational Fluid Dynamics) application that simulates the flow around and in a certain object. For such applications it is quite common that large caches and many CPU cores can result in a significant increase in performance. More information on the Euler3d benchmark Is there ... here.

Benchmarks: Audio Editing

Now we come to the “correct” everyday applications. We want to start with music editing software. All tests are based on a wave file of around 710 MB, which we convert to MP3 files with the help of iTunes, LAME and the Nero AAC encoder. A conversion to the Ogg Vorbis format is also used. All programs are strictly single-threaded, so they only make use of one core.

iTunes

Information on the benchmark

iTunes is a multimedia program from Apple that allows you to play, convert, organize and buy all kinds of music. The first version of the very successful software came on the market in 2001. There is now the ninth revision. We are currently using this with version number 9.1.2.5. However, this version does not yet make use of multi-core processors either. The SSE units are used very often for this.

Nero AAC

Information on the benchmark

The Nero AAC encoder is a freely available encoder that is called from the command line and is used in Nero 10, for example. We are using the latest version 1.5.1, which, like iTunes, is not yet multithreaded. High-performance SSE units are therefore the most important criterion for high performance.

LAME

Information on the benchmark

LAME is an open source encoder for converting audio files into MP3 format. The big difference to the Fraunhofer-Gesellschaft's MP3 encoder is that LAME is free. This is why LAME is also used in a large number of software products. We are using the latest version 3.98.4 from March 2010, which does not yet support multi-threading. However, we do not use the fully compiled package but only the source code and create our own program package with the help of VisualStudio 2008 and the integrated C ++ compiler from Microsoft. Normally, however, LAME uses an Intel compiler. In order to avoid any differences, we have created our own version.

OggEnc

Information on the benchmark

OggEnc is again a free audio encoder. It converts audio files into the Ogg container format. The structure and structure of Ogg are similar to the MPEG-4 file format MP4. We use version 2.87 of OggEnc and, like the other three programs mentioned, does not support multi-threading.

Benchmarks: image editing

When it comes to image processing, we rely on the free programs GIMP and IrfanView. Both programs make very weak use of multiple cores, but are more likely to be classified as single-threaded.

GIMP

 

 

Intel socket LGA-1155

• Core i7 3770K:
  IB architecture, E1 stepping, 3,5 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Intel Core i5-3570K:
  IB architecture, E1 stepping, 3,4 GHz, 4 cores, turbo mode active, 4 x DDR3-1600
• Intel Core i7-2700K:
  Self-service architecture, D2 stepping, 3,5 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Core i7 2600K:
  Self-service architecture, D2 stepping, 3,4 GHz, 4 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Core i5 2500K:
  SB architecture, D2 stepping, 3,3 GHz, 4 cores, turbo mode active, 4 x DDR3-1333

For Intel's “Sandy Bridge” and “Ivy Bridge” processors for the LGA-1155 socket, 4 x 4 GB G.Skill Ripjaws Z DDR3-1600 are used, operated on DDR3-1333. The memories are operated with the CL9-9-9-24 2T latencies. That comes as a mainboard MSI Z77A-GD65* used with BIOS version 7751vP0. All energy saving mechanisms are activated in the BIOS.

 

 

MSI Z87-G43

Intel socket LGA-2011 and LGA-2011-3

• Core i7-5960X
  R2 stepping, 3,0 GHz, 8 cores, turbo mode active, HTT active, 4 x DDR4-2133
• Core i7-4960X
  S1 stepping, 3,6 GHz, 6 cores, turbo mode active, HTT active, 4 x DDR3-1600
• Core i7-3960X
  C2 stepping, 3,3 GHz, 6 cores, turbo mode active, HTT active, 4 x DDR3-1600

For the processors of the LGA-2011 socket, 4 x 4 GB G.Skill Ripjaws Z DDR3-1600 and an ASUS Rampage IV Gene with BIOS 4901 are used. The socket LGA-2011-3 is measured with 4 x 4 GByte Corsair Vengeance LPX DDR4-2666, operated with DDR4-2133 and timings of 15-15-15-36. An MSI X99S Gaming 7 with BIOS V17.4 is used as the mainboard.

 

 

 

Hardware: AMD systems

AMD Socket FM2 +

• A10-7850K
  Steamroller architecture, A1 stepping, 3,7 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A8-7600
  Steamroller architecture, A1 stepping, 3,3 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A10-6800K
  Piledriver architecture, A1 stepping, 4,1 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A10-6700
  Piledriver architecture, A1 stepping, 3,7 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• A10-6500T
  Piledriver architecture, A1 stepping, 2,1 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600

The processors for FM2 and FM2 + were measured on the MSI A85XA-G65. The Kaveri models and the A10-6800K also on the MSI A88XM-E45.

AMD Socket AM3 +

• FX-8370E:
  Piledriver architecture, C0 stepping, 3,3 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• FX-8350:
  Piledriver architecture, C0 stepping, 4,0 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600
• FX-8150:
  Bulldozer architecture, B2 stepping, 3,6 GHz, 4 modules, turbo mode active, CMT active, 4 x DDR3-1600

The same G.Skill modules are used for AMD's Bulldozer processors for the AM3 + socket as on the Intel systems. That comes as a mainboard ASUS Crosshair V Formula* (990FX chipset) used with BIOS 1703. All energy saving mechanisms are activated in the BIOS.

 

 

ASUS Sabertooth 990FX

Due to the lack of support for the new processor from our test board ASUS Crosshair V Formula, we had to change the board. The measurement results for the AM3 + socket were therefore completely created on the new ASUS Sabertooth 990FX.

AMD Socket FM2

• A10-5800K

Trinity architecture, A1 stepping, 3,8 GHz, 4 cores, 4 x DDR3-1600

The G.Skill DDR3 memory mentioned above works here as well. The Gigabyte GA-F2A85X-UP4 with BIOS F4 is used as the mainboard. All energy saving mechanisms are activated in the BIOS.

More hardware

Graphic card:

• MSI Radeon HD 7970 Lightning
• AMD Radeon HD 3450 (DDR3):
  only for measurements of power consumption

Memory:

• 16 GB (4 x 4 GB) G.Skill Ripjaws Z DDR3-1600
  SPD operation: DDR3-1600, 9-9-9-24 at 1,5 volts

 

 

 

Power adapter:

• be quiet! DARK POWER 550W R10

Hard disk:

• Seagate ST2000VX000

Cooler:

• Scythe Katana III
• Noctua NH-U12S

 

 

 

Measurement:

• Tenma 72-6185 clamp ammeter
• Voltcraft Energy Check 3000
• Tenma infrared thermometer 72-820
• Metex multimeter M-3640D
• Profitec KD 302 energy cost meter

Software and benchmarks

Operating system and driver

• Windows Ultimate 7 (64-bit) including all updates up to May 2013
• AMD / ATI Catalyst 13.4 WHQL
• Windows 7 supplied drivers for the chipset and network card

CPU benchmarks

• Synthetic benchmarks
  • PCMark05 v120_1901
  • PCMark 7 v1.4.0
  • Euler 3D
• Audio editing
  • iTunes 11.0.2
  • LAM 3.98.4 (compiled with Visual Studio 2008)
  • Nero AAC Encoder 1.5.1
  • OggEnc 2.87
• image editing
  • Gimp 2.8.4
  • Irfan View 4.35
• video editing
  • HandBrake 0.9.9 RC1 64bit
  • Avisynth 2.58 and x264 encoder 0.132.2310
• Packer
  • 7Zip 9.20 (64 bit)
  • WinRAR 4.20 (64 bit)
  • WinZip 17.5 Build 10480g (64bit)
• Rendering
  • Blender 2.67 (64 bit)
  • Cinebench 11.529 (64 bit)
  • Pov-Ray 3.7 RC7 (64 bit)
• Games
  • Assassin's Creed III* V. 1.0.2 (DirectX 11 - Savegame)
  • Crysis 3* (DirectX 11 - savegame)
  • Serious Sam 3* V. 1.5.1466 (DirectX 9 - Savegame)
  • The Elder Scrolls V: Skyrim* V. 1.8.151.0.7 (DirectX 9 - Savegame)
  • tomb raider* V. 743.0 (DirectX 11 - Savegame)
• Encryption
  • 7-Zip
  • TrueCrypt 7.1a
  • WinZip

SiSoft Sandra 2013.05.19.44

Overall, we tried to focus strongly on real applications with the benchmarks and to distance ourselves from synthetic tests. Whenever possible, we also used the 64-bit version.

A special remark should be made about LAME: By default, LAME is created with an Intel C ++ compiler. In the past, this has repeatedly caused irritation, which is why we also use a version in our new benchmark course that we created ourselves with the help of Visual Studio 2010 from Microsoft. In terms of performance, however, it does no difference.

Other tools

• CPU-Z
• CPU-Z Latency Tool
• CoreTemp
• Core2MaxPerf
• Fraps

Test methodology

Apart from the remarks already made on this and on the previous page regarding our test philosophy, we want to briefly summarize the essential points again. Unless otherwise stated in the direct test description, the following points always apply:

• All available energy saving mechanisms are activated.
• If the CPU has a turbo mode, this is activated.
• If the CPU supports hyperthreading / core multithreading (CMT), this is activated.
• If not mentioned otherwise, always comes MSI Radeon HD 7970 Lightning for use.

Technology

The new FX-8320E

In the technical sector there is of course nothing new to report today. The technical implementation is basically the same as it was when the Bulldozer in late 2011 by AMD, of course including the advantages of the current Piledriver cores.

AMD's current innovations - shown with the FX-8370E, as now also with the FX-8320E - are more of a “cosmetic” nature. The manufacturer simply reacts to the allegations that the AMD CPUs in the middle class segment are too power hungry and thus tries to present a more attractive TDP class. The two new E-models are therefore in the 95-watt TDP range and no longer in the 125-watt class, like the 8370 and 8320.

 

 

 

The implementation takes place in simple steps, namely in the form of clock and voltage lowering - presumably also the DIE selection. The FX-8320E only has a base clock of 3,2 instead of 3,5 GHz. In turbo mode, however, it still clocks up 4 GHz, just like the previous FX-8320.

In addition, the FX-8000 series remains with models with four modules, whereby each of the modules can handle two threads and AMD therefore speaks of eight CPU cores.

Tabular comparison

In the middle class, the FX-8150 and FX-8350 are practically yesterday's news, but both are still available in stores. The actual offers are represented today by the models FX-8320 and FX-8370 and their E versions.

The nomenclatures are again not always conclusive, because an FX-8320 (E) offers less clock rate than, for example, an FX-8150 and should theoretically not have a higher number. With the FX-8370 (E) you can at least outperform an FX-8350 in terms of turbo clock.

TDP and price as weapons

Over the past few years, since the introduction of the FX-8000 models, it seems to have become clear that you can no longer win a flower pot with high GHz clock speeds alone. Especially if the high clock rates in your own architecture are clearly inferior to the lower clock rates of the competition. And so the new motto is quite obvious: less is more!

And in fact, AMD is currently losing less with the E variants than it is winning. The lower clock rates are not clearly noticeable in the performance, because the turbo clock often bridges here. At the same time, however, you can now name the lower TDP class of 95 watts.

 

 

 

If you look at the market in the area of ​​the FX-8000 series and the AM3 + socket, the FX-8320 (E) in German-speaking countries is probably the cheapest option to get started here. The current staggering also shows the low margins with which the manufacturer has to act here. Often individual models are simply separated by only 10 euros - outliers in our table can be explained by discontinued models in sales or fluctuating exchange rates.

We do not know how much AMD still has to select a good three years after the introduction of the FX-8000 processors in order to receive E and non-E models. But the production should really be mature enough by now that enough high-quality models can be captured, which can be operated with lower voltages in order to be sold as e-models. In this area - with four modules - the 65-watt class remains simply utopian.

Practice

overclocking

Despite the goal of being able to achieve a lower TDP, AMD also offers this FX processor without a multiplier blockade, and of course the end customer is free - at their own risk - to tease out the maximum performance from the CPU. This is also relatively easy using the free multiplier. However, you have to struggle with a few restrictions.

If you play on the CPU's multiplier, the processor's idle rate is no longer lowered as much as it would without manual intervention. This is of course counterproductive when it comes to energy efficiency. When it comes to bare clock rates and performance at the lowest possible price, deactivating the power-saving mechanisms and increasing the voltage in overclocking can even have an effect on an FX-8000 model. After that you don't have any savings in terms of power consumption, but you can get the maximum performance from the CPU squeeze out.

In our test, however, we do not want to deal with the intricacies of overclocking FX processors, so that we do not resort to means such as water cooling or additional voltage increases. Instead, we use our usual air cooling and just turn the clock up above the multiplier.

We were able to increase the clock rate by a further 600 MHz at the base clock rate before the chip lost its stability. This shows that there is definitely potential in the FX-8320E, but again it is not really a surprise compared to the regular FX-8320.

With the increase in clock speed, the power consumption also increases and reaches a value of over 196 watts. Compared to the previously measured 150 watts, this is a significant increase. The increase in clock speed also ensures that values ​​were around 20 watts higher in idle mode.

Performance index [OC]

Games [dGPU]

The increased performance is quite impressive, because the average across all benchmarks is around seven percent. Here everyone has to decide for themselves whether they want to accept the declared disadvantages.

But the means of the benchmarks is of course not the panacea for everyone. As our breakdown shows, the differences lie in the details and so the overclocking measure can hardly bear fruit in one case, but it may increase by up to 18 percent in another. This means of our tool makes it easier to judge personally where one applies one's own standards - including all possible negative consequences.

Direct comparison

Practice: power consumption

In the following, we determine the average consumption of the entire system without a monitor. A standard energy cost meter is used here, in our case an Energy Check 300. Over a period of 20 minutes, we record the course and, of course, state the average value in watts. We use Core2MaxPerf for all processors as a full load scenario.

It remains clear that this measurement of the entire system cannot, of course, be as accurate as previous measurements on our part, in which we only reduced the CPU load and power consumption with special mainboard modifications. Unfortunately, everything flows into such measurements. Sudden spikes caused by hard disk access, programs starting in the background that request higher CPU loads, or similar scenarios. One can only try at this point to exclude all evildoers. But it can never succeed in the end, and so the following diagrams are only indicative - always based on our selected test system!

Based on our explanations, two circumstances clearly stand out in idle mode. On the one hand, the FX-8000 models are basically on the same level, but unfortunately none of the candidates can really score compared to the ranking. The next smaller level are the AMD APUs, which also work in roughly the same segment. Another fundamental limitation of such a consideration of the overall system is the selected components, such as the mainboard or the power supply.

However, the clear lead of Intel offshoots, which operate on different motherboards to AMD, but are also equipped with identical components such as power supply unit, hard drive and so on, cannot be disputed.

And so the Intel processors are clearly ahead in this comparison.

A look at the load scenario clearly shows that AMD is right to separate the 125-watt TDP class from the 95-watt class. The measurements of the total system power consumption show a difference in the range of 30 watts, in some cases even more. It's nice to see, but unfortunately it's still a bit too high. After all, over 30 watts separate this mid-range AMD CPU from Intel's Socket 1150 top model i7-4790K.

In principle, this only gives a rough indication, because at this point the dependency of the load tool is also decisive. This tendency is evident in the tool we have chosen. Intel's top model i7-4790K operates with a TDP of 88 watts, AMD's mid-range model operates with a TDP of 95 watts - roughly the same region. How much leeway the two different CPU models have for maximum TDP under typical load scenarios is unfortunately unclear. It seems that the AMD models are approaching their limits faster here.

And so in the end there is only one look at the benchmarks, which have to clarify where exactly which CPU is.

Benchmarks: Synthetic

PCMark 05

Information on the benchmark

Information on the benchmarkAs the name suggests, PCMark 05 dates from 2005 and is therefore already several years under its belt. Nevertheless, the benchmark suite from the Finnish company Futuremark is still very well suited to classifying the computing power of processors and their memory performance. We only use the CPU and memory suite, so that conclusions can only be drawn for these components. The CPU suite is based on eight different tests from the fields of packing / unpacking, encryption, and audio and video processing, and thus allows an overview of the everyday performance of the processors. The CPU suite benefits equally from high clock frequencies and from multiple cores. In the memory suite that is also used, however, factors such as cache rate, cache size and memory bandwidth play a role, since the tests essentially consist of reading, copying and writing data of different sizes to the memory.

PCMark 7

Information on the benchmark

Information on the benchmarkPCMark 7 was only launched this year and represents the latest system benchmark from Futuremark. We only use the Computation Suite to draw conclusions about the computing power of the processors tested. The suite comprises three different tests from the fields of video transcoding and image processing. It allows you to see the everyday performance of the processors. In addition to a high clock rate, the tests benefit from multiple cores.

Euler3d benchmark

Essentially, it is a CFD (Computational Fluid Dynamics) application that simulates the flow around and in a certain object. For such applications it is quite common that large caches and many CPU cores can result in a significant increase in performance. More information on the Euler3d benchmark Is there ... here.

Benchmarks: Audio Editing

Now we come to the “correct” everyday applications. We want to start with music editing software. All tests are based on a wave file of around 710 MB, which we convert to MP3 files with the help of iTunes, LAME and the Nero AAC encoder. A conversion to the Ogg Vorbis format is also used. All programs are strictly single-threaded, so they only make use of one core.

iTunes

Information on the benchmark

iTunes is a multimedia program from Apple that allows you to play, convert, organize and buy all kinds of music. The first version of the very successful software came on the market in 2001. There is now the ninth revision. We are currently using this with version number 9.1.2.5. However, this version does not yet make use of multi-core processors either. The SSE units are used very often for this.

Nero AAC

Information on the benchmark

The Nero AAC encoder is a freely available encoder that is called from the command line and is used in Nero 10, for example. We are using the latest version 1.5.1, which, like iTunes, is not yet multithreaded. High-performance SSE units are therefore the most important criterion for high performance.

LAME

Information on the benchmark

LAME is an open source encoder for converting audio files into MP3 format. The big difference to the Fraunhofer-Gesellschaft's MP3 encoder is that LAME is free. This is why LAME is also used in a large number of software products. We are using the latest version 3.98.4 from March 2010, which does not yet support multi-threading. However, we do not use the fully compiled package but only the source code and create our own program package with the help of VisualStudio 2008 and the integrated C ++ compiler from Microsoft. Normally, however, LAME uses an Intel compiler. In order to avoid any differences, we have created our own version.

OggEnc

Information on the benchmark

OggEnc is again a free audio encoder. It converts audio files into the Ogg container format. The structure and structure of Ogg are similar to the MPEG-4 file format MP4. We use version 2.87 of OggEnc and, like the other three programs mentioned, does not support multi-threading.

Benchmarks: image editing

When it comes to image processing, we rely on the free programs GIMP and IrfanView. Both programs make very weak use of multiple cores, but are more likely to be classified as single-threaded.

GIMP

Information on the benchmark

GIMP is a free and free image manipulation program under the GNU General Public License (GPL). Similar to the paid Adobe Photoshop, GIMP offers numerous filters and allows intensive image processing. The biggest advantage of the GIMP, besides the fact that it is free, is the platform independence. So GIMP can be used under Windows, Linux and OS X.

A 9000 × 9000 pixel JPEG image is used as a test scenario, on which we first apply a blur filter, then a Gaussian soft focus and finally a red-eye filter. All three filters make little use of multiple cores in version 2.6 we are using.

IrfanView

Information on the benchmark

In contrast to GIMP, IrfanView is more of an image viewer than an image editing program. Nevertheless, this free software also offers some welcome additional functions, such as something for scaling multiple images.

We use precisely this function to scale down an image pool with a total of 256 Mbytes of JPEG images to a size of 256 × 156 pixels each. We trust the Irfanview version 4.27, which does not yet handle extensive multithreading.

Benchmarks: Video Editing

When it comes to video editing, Handbrake and MainConcept are used, which use different codecs with different quality settings. The source file is always a 380 MB HD video. While software that has not made use of multiple cores has primarily been used up to now, MainConcept and Handbrake are two prime examples of parallelization. Even six cores are optimally used.

Handbrake x264

Information on the benchmark

HandBrake is a software for converting video files. Originally developed for BeOS, the free program is now available for OS X, Windows and Linux. We are using Handbrake version 0.9.5, which makes massive use of several cores as well as SSE commands up to SSE 4.2. However, AVX is not yet supported. H.264 is used as the codec.

We pass two benchmarks with Handbrake. In the first, we convert the above video into an iPod-compatible format of 320 × 176 pixels. In the second, however, we rely on a Full HD resolution of 1920 × 1080 pixels.

Avisynth & x264 encoder

Information on the benchmark

GIMP is a free and free image manipulation program under the GNU General Public License (GPL). Similar to the paid Adobe Photoshop, GIMP offers numerous filters and allows intensive image processing. The biggest advantage of the GIMP, besides the fact that it is free, is the platform independence. So GIMP can be used under Windows, Linux and OS X.

A 9000 × 9000 pixel JPEG image is used as a test scenario, on which we first apply a blur filter, then a Gaussian soft focus and finally a red-eye filter. All three filters make little use of multiple cores in version 2.6 we are using.

IrfanView

Information on the benchmark

In contrast to GIMP, IrfanView is more of an image viewer than an image editing program. Nevertheless, this free software also offers some welcome additional functions, such as something for scaling multiple images.

We use precisely this function to scale down an image pool with a total of 256 Mbytes of JPEG images to a size of 256 × 156 pixels each. We trust the Irfanview version 4.27, which does not yet handle extensive multithreading.

Benchmarks: Video Editing

When it comes to video editing, Handbrake and MainConcept are used, which use different codecs with different quality settings. The source file is always a 380 MB HD video. While software that has not made use of multiple cores has primarily been used up to now, MainConcept and Handbrake are two prime examples of parallelization. Even six cores are optimally used.

Handbrake x264

Information on the benchmark

HandBrake is a software for converting video files. Originally developed for BeOS, the free program is now available for OS X, Windows and Linux. We are using Handbrake version 0.9.5, which makes massive use of several cores as well as SSE commands up to SSE 4.2. However, AVX is not yet supported. H.264 is used as the codec.

We pass two benchmarks with Handbrake. In the first, we convert the above video into an iPod-compatible format of 320 × 176 pixels. In the second, however, we rely on a Full HD resolution of 1920 × 1080 pixels.

Avisynth & x264 encoder

Information on the benchmark

The x264 encoder is a platfo

Assassin's Creed III

Universal encoder for the video format H.264 (MPEG-4 AVC) and is published under the GNU General Public License. The special thing about it is that the source code can be viewed, so you can always understand what the encoder is doing. By default, the x264 encoder does not have a GUI for operation, because this is done exclusively via the console with the help of AviSynth scripts. We are currently using the x264 encoder in revision r2053, which already includes initial support for the AVX vector extension. In addition, the encoder makes massive use of several cores and SSE commands up to SSE 4.2.

With the encoder we convert a 720p video into the H.264 format. We use a 2-pass conversion which we complete four times. Then we determine both the mean value for the conversion time and the mean frame rates for each of the two runs.

Benchmarks: Packers

We use 7-Zip, WinRAR and WinZip as packing programs, whereby WinZip and 7-Zip are used both once with AES encryption and once without this feature. In both cases, the highest compression level (Ultra) is selected.

7-Zip

Information on the benchmark

7-Zip is a free packing program that was launched by the Russian programmer Igor Pavlov in 1999 and is still intensively maintained today. It represents the reference implementation of the Lempel-Ziv-Markow algorithm (LZMA) that he developed.

We use version 9.20 in the 64-bit variant. LZMA2 is used as the compression method. This supports unrestricted multithreading. 7-Zip is of course used as the archive type.

WinRAR

Information on the benchmark

WinRAR is different from 7-Zip shareware. Nevertheless, you can use it almost without restriction for free. Similar to 7-Zip, it supports various archive formats, including the RAR format that gives it its name and is used by us. We use version 4.01 in the 64-bit variant. This is fully multi-threaded and makes optimal use of all existing cores. However, like us, you have to create a real archive and not use the integrated benchmark, because it only supports a few cores.

WinZip

Information on the benchmark

WinZip isn't a free program either, but the shareware version offers enough functionality and can run without restrictions. In the version 14.5 we use, WinZip supports AES encryption, which is supported, for example, by "Sandy Bridge" but also by some processors from the first Core i generation. In terms of multithreading, the same applies as with 7-Zip. A maximum of four cores are used optimally only here. The eponymous zip format is used as the archive format.

Benchmarks: rendering

In the rendering segment we trust Blender, Cinebench version 11.5 and POV-Ray. All three programs make massive use of multi-core processors, so each additional core saves real time.

Blender

Information on the benchmark

Blender is a 3D graphics software designed completely for multi-threading, which contains functions to model, texture, animate and render three-dimensional bodies. Originally, Blender was an in-house program for the Dutch animation studio NeoGeo. The chief developer Ton Roosendaal founded the company Not a Number Technologies (NaN) in 1998 to further develop and sell blenders. After the company went bankrupt, however, the creditors agreed to put Blender under the free software license GNU General Public License (GPL) for an amount of 100.000 euros. Thus, Blender is now freely available and expandable. We use Blender as a 64-bit offshoot in version 2.59.

POV Ray

Information on the benchmark

POV-Ray is a free graphics program which, like Blender, makes heavy use of multithreading. With POV-Ray, not only 2D but also 3D scenes can be created. The software has been developed for over 20 years and new functions are constantly being added. An integrated test is used as a benchmark. As with Blender, we rely on a 64-bit offshoot of version 3.7 Beta 38 for the program version. This version supports both SSE2 and SSE4 extensions to increase performance. The new AVX commands are not yet integrated.

Cinebench 11.5

Information on the benchmark

We tested Maxon's Cinema4D rendering software from Germany using Cinebench 11.5, which is available for both Mac and Windows. Maxon uses the underlying Cinema4D engine not only in Cinebench, but also in commercial products that were involved in the development of the movie “Spider Man”, for example. The engine uses massive multithreading to reduce runtime. As usual with Cinebench, we present both the result of just one core and that of using all cores or threads.

Benchmarks: encryption

Truecrypt

Information on the benchmark

TrueCrypt is encryption software with which entire data carriers or individual files can be encrypted. The program is freely available and can be used under Windows as well as Mac OS X and Linux. AES, Twofish and Serpent and combinations of these three can be used as encryption algorithms. If you rely on AES encryption and use a processor with the AES instruction set extension, the encryption or decryption can be greatly accelerated with the aid of the extension.

We use version 7.0a. This supports both the AES extension and multithreading. We use the integrated benchmark with a buffer size of 1000 Mbytes.

7-Zip: AES

Information on the benchmark

7-Zip is a free packing program that was launched by the Russian programmer Igor Pavlov in 1999 and is still intensively maintained today. It represents the reference implementation of the Lempel-Ziv-Markow algorithm (LZMA) that he developed.

We use version 9.20 in the 64-bit variant. LZMA2 is used as the compression method. This supports unrestricted multithreading. 7-Zip is of course used as the archive type.

WinZip: AES

Information on the benchmark

WinZip isn't a free program either, but the shareware version offers enough functionality and can run without restrictions. In the version 14.5 we use, WinZip supports AES encryption, which is supported, for example, by "Sandy Bridge" but also by some processors from the first Core i generation. In terms of multithreading, the same applies as with 7-Zip. A maximum of four cores are used optimally only here. The eponymous zip format is used as the archive format.

Benchmarks: Games [dGPU]

We use two resolutions below. We show benchmarks with 1366 x 768 pixels on the one hand, and benchmarks with a resolution of 1680 x 1050 on the other. We show medium detail levels for the latter resolution; for the first resolution we manually lowered the detail levels from medium a little bit come close to “notebook quality” with an integrated GPU.

The scenes used are not identical to our usual graphics card benchmarks. At this point, of course, we tried to choose game sequences that had a CPU rather than a GPU limit.

Assassin's Creed III

In-game test scene

Crysis 3

In-game test scene

Serious Sam 3

In-game test scene

TES V: Skyrim

In-game test scene