AMD, a brand known in the field of X86 processors, enters strongly in the market for processors in 1991, breaking up the monopoly of Intel chips with the AM386.With the Athlon 64 placed on the market in 2003, AMD becomes "the first company to create a desktop class 64-bit x86 processor". After that date, the overtake on Intel in performance becomes apparent with the Opteron series dedicated to the server and workstation, and in 2005 it produced the first dual core processor with integrated memory controller. Following it consolidates another successful, builds the first quad-core processor with the Athlon 64 X4. Another big jump is then with the acquisition of ATI brand in 2006.After various legal and economic problems with Intel, AMD introduces a new architecture completely redesigned from scratch, the Bulldozer architecture that characterizes the new FX series processors.
About a month and a half from the official launch of the new AMD FX, at which we presented a detailed architectural analysis, we are now going to sum up on this new architecture, after a series of extensive tests. We apologize to our readers for the delay in which this article was presented. The extensive suite of tests, the concomitant release of Intel's Sandy Bridge-E and the delay in sample delivery constitute the main reasons.
Although expectations were high for Bulldozer, for a month and a half after its release, it is now clear the atmosphere of disappointment, especially in the desktop segment. The competition in this segment with Intel puts AMD in a position of distinct disadvantage and Michael Silverman, a spokesman for AMD said that actually makes no sense to continue to pursue a direct challenge to Intel vs. AMD, preferring to pursue a differentiation, so to conquer market share that competition can not cover. In this perspective, the same Bulldozer architecture still makes sense, proposing effective solutions for the HPC market and cloud computing. We partly talked about in dedicated Opteron 6200 and 4200 solutions article, of which we recommend reading to those who have not already done so.
In this article we will focus, however, on the desktop version of Bulldozer and we will evaluate the FX-8150 as such, testing it with the software currently available. Probably these CPUs have still untapped potential, but until there is full implementation of the new instruction set proposed by AMD (mainly FMA4), with its code optimizations, the situation will reflect approximately the same tests performed on the following pages.
You can also read about it, reading our previous article on Bulldozer, where it is analyzed the architecture:
We summarize briefly the main features of the FX-8150 CPU:
- 8 physical cores
- 3.6 GHz (up to 4.2 GHz in turbo mode)
- 3 Levels of Cache
- 8MB of unified last stage cache
- 4 HyperTransport Controllers
- Integrated dual-channel DDR3 controller
Main Exclusive Features
The Bulldozer architecture is the most innovative project designed to date after the introduction of K10 architecture common to Phenom and Athlon, because it exploits the cluster concept as a basic element for all CPUs made with CMT architecture (Cluster Multi-Threading). Moreover Bulldozer is conceptually the union of 2 independent dual-core modules linked by a second-level cache, but also from a shared floating point unit, that can operate as a single unit with 256-bit or operate separately and simultaneously as two independent 128-bit units. Each Bulldozer module contains a second level L2 cache of 2 MB shared between the two cores. The diagram below illustrates the concept.
Recall that a more in-depth review of Bulldozer is available on our website with the article: AMD Bulldozer at debut.
Before continuing, a little update on the statistics of the Bulldozer die, for which, as stated also by AMD, were given incorrect data.
We conclude this section with a comparison of the AMD FX CPUs now available:
AMD Bulldozer press kit: FX-8150
The press kit sent by AMD is very rich and includes an Asus Crosshair Formula V, the FX-8150 CPU and a liquid-kit produced by Asetek branded AMD. Let's look at the CPU.
Inside you will find both the CPU and the air cooler that is absent in our case because was replaced with a liquid kit that will be presented later.
The CPU box is a black box with the logo printed on it. Opening it, there is, protected by a sponge on both sides, the CPU.
CPU-Z correctly reports the characteristics of this CPU: base clock of 200MHz, 18x multiplier for a default frequency of 3.6 GHz, 8 logical and physical cores.
In idle the frequency is reduced up to 1400MHz thanks to Cool'n quiet, lowering the operating voltage of our CPU from 1.248V to 0.984V.
The analysis of the tab that regard the cache we note 16KB per core for the L1 data cache and for the instructions we have a cache of 64KB per module. The associativity is 4-way for data cache and two-way for L 1 Instructions cache. Finally we have a 2MB L2 cache for each module and a shared L3 cache of 8MB (for a total of 16MB). Associativity for L2 and L3 is, respectively, 16 and 64-way.
AMD Bulldozer press kit: ASUS Crosshair V Formula
The board received in our labs from AMD to enable us to test Bulldozer as cutting-edge AMD processor for Scorpius platform, which we remind being the FX-8150, is the ASUS Crosshair V Formula. Over the years Asus has become synonymous with quality where the ROG (Republic of Gamers) production line is the highest expression in terms of quality and technology on both the hardware and software aimed at strongly support the growing set of people who practice the overclocking.
As shown from the series of images, Asus has paid particular attention to the packaging that appears to be both robust and elegant, yet full of information that show the features related to the product in question.
The layout is very neat and innovative and is separated from the traditions so far followed. While the color scheme is classic and always found in other ROG models, a clear advantage in terms of space around the socket is visible on the ASUS Crosshair V Formula . The availability of space on one hand makes more permissive in the choice of the heat sink for the CPU and the other hand let you use a simpler and more effective insulation when you run them under cold overclocking tests. Surely you've noticed that the area at the left of the socket is free and is devoid of classical cooling flange covering the 990FX because that was moved over on the same horizontal line of the power dissipation of the chip, connected by a central channel of Heat Pipe to maximize the dissipation of heat from warmer areas to cooler areas.
On the left we see the presence of 990FX under the sink.
The bundle is rich and full, as required for this class of product. Interesting, hidden in part on the adhesive, there is a USB cable for the monitoring of temperature, frequency and tension where you can act in real time to help us in the practice of overclocking without having to enter the bios or by acting on the test system.
In the area below the PCB of ASUS CROSSHAIR V Formula can not fail to notice the mounting of the socket retention bracket as well as the retention brackets of heat sinks adjacent to it.
The number of USB ports is remarkable, distinguished by color depending on the speed of connection and origin. One of the black ports, can be converted for use with ROG Connect, while we note the absence of a FireWire port and a second LAN port. Instead, from left, below the 2 USB ports group we have 1 e-SATA port. To end the blockade of ports, all the I/O sockets reserved for audio signals.
The viewed socket is the latest AM3+, designed to support the latest processor family. Including those of greatest importance, the Phenom and, the subject of our review, the processor of choice for the desktop market: the FX-8150.
Also on the adjacent part of the perimeter of the motherboard we have 2 extra power connectors for both the CPU socket area of the memory slots.
Proceeding to the left starting from the higher end of the motherboard you can see both the 4-pin molex connector for expansion slot power and the switch to activate/deactivate the ROG Connect connector.
Highlighted, covered by the adhesive, the ASUS Crosshair Formula V has a dedicated audio chip from Creative, the X-Fi 2 SupremeFX signal processing for high-definition audio.
In addition to the usual connectors depicted in the photos, Asus' ROG series include buttons to facilitate bench, overclock, power and reset operations.
Other highlighted connectors dedicated to replicate USB ports and connections for the fans. In total we can count 8 dedicated fan connectors on the PCB.
Framed in the figure are the 6+1 SATA3 6Gbps ports.
On the left of EATX Power 24-pin connector, in red there is the connector dedicated to the replication on the chassis of the USB 3.0 ports.
In the outer part there is a button GO BUTTON that when pressed at boot the system performs a test of RAM, while when pressed in Windows loads a profile specifically associated and stored in the BIOS.
Framed in the top the RAM slots distinguished by color, for installing DIMMs in pairs to support dual channel. Below, along the boundary line, the reading points of the various voltages of the main chips present on the Mainboard such CPU, RAM, NB, SB, etc.
The ASUS Crosshair V Formula has a total of 5 PCI-Express slots and 1 PCI. Supports both SLI and Crossfire configurations. Starting from the right slot and considering only the red, the first and third are at 16x while the second is at 8x and the fourth (i.e. the last) is at 4x.The PCI slot is black, supporting cards compatible with the old standard, while the only slot remained, gray colored, is always a PCI-Express but with 1x speed.
In the picture an elegant passive heat sink above the SB950 southbridge. Below, the support chip for iROG technology for system overclocking.
Following, the technical specifications of the Crosshair V Formula
AMD Liquid Cooling Kit
The pump body is one piece with the waterblock and was found (the pump) very quiet. To hear the rotation you need to get closer with the ear.
Same goes for the 2 fans that are Everflow Mod: R121225SU, of course of the size of the radiator, and should be mounted on 2 sides of the same to create the push-pull effect to improve the air flow in order to keep down the RPMs and then the noise created by the rotation. Recall that both the pump and the fans are connected directly to the system and automatically controlled by the software in the kit.
System configuration and testing methodologies
The FX-8150 CPU will be subjected to numerous tests over the course of this review.
- 3DMark Vantage
DATA COMPRESSION AND MULTIMEDIA
- Cinebench 11.5 and 10
- X264 HD Benchmark
- Hand Brake
- Fritz Chess
- Euler 3D
- SiSoft Sandra
Let's start now with all the tests that put stress on the processor, which we remind you to be an AMD FX-8150 with well 8 cores and Bulldozer architecture. This allows the OS to take advantage of 8 threads simultaneously.
The tests are performed with both the CPU at default and both setting a fixed frequency for all cores of 4 GHz, in this last case it is possible to evaluate more precisely the IPC of the various CPU and see who has at the same frequency an architecture capable of generating higher performance than the other.
3DMark Vantage CPU test
As can be inferred from the graph above, without a doubt the processing power of AMD is more than adequate to that of competitors. In fact, it performs just under the Intel i7-2600K, but at the same time you can see how it is not far from Intel CPUs without Hyper-Threading Technology.
Instead, setting a fixed frequency of 4 GHz, it is clear as the CPU loses further ground against other Intel CPUs, in this case the total score is under the i5-2500K Sandy Bridge CPU.
Above, we reported a screen carried out with the latest version of AIDA64 that shows the bandwidth of memory, L1, L2 and L3 caches at various phases of reading, writing and copying. The thing that we want to put special emphasis is a very unusual result, for example in the reading tests (Read) you can see that Bulldozer clearly shows an L2 cache speed slower than the L3 cache one. This result is obviously impractical and wrong, because such a situation can not be real. No doubt we can say that the results made with this benchmark are unfortunately to be taken with caution, and we just bring them back for the record.
Let's add the remainder of the charts made by the CPUs set to 4 GHz, of course, the same considerations apply as before.
In this benchmark the performance is such as to ensure a more than comfortable result also with this application. The performances are quite comparable to the Intel i7-2600K, but also you can see that the gap with the "old" generation of AMD Thuban, in particular, the 1100T, is not very high. We have a good margin in the compression phase, which also manages to be even higher than the Intel CPUs of the same segment.
This graph is very significant, if we saw in the previous case, the AMD CPU to deal with the i7-2600K, in this case we observe that the CPU is a definite step back. The end result sees the CPU to be stuck behind the CPU architecture with the old 1100T Thuban. A sign that the IPC on a single core is essentially unchanged, even in some cases is even lower than the older generation.
WinRAR 4.01 and 4.10
Other popular software compression and decompression of data archives. Inside it there is a benchmark utility that compresses a standard file suitable for this purpose; the software will return the compression speed in KB/s. We also performed the tests by turning off core parking in the operating system Windows 7 SP1. You can read our study of the topic at the following link:
This benchmark ran within the same application shows a double face. If we look at the result without affecting the parameter of Core Parking it can be seen that the new AMD Bulldozer architecture is able to have a really amazing performance compared to other competitor CPUs, even the Intel top of the range, the i7-3960x. But we note that all Intel CPUs with Hyper Threading in fact have the same performance of CPUs without this technology. In particular we see how the Intel i7-2600K has the same performance with and without HT. This made us go further in our analysis and we have examined the subject (please read the article listed above in this regard).Disabling the Windows core parking, via a registry key, performance have changed for many CPUs, and we rerun the tests especially for Intel CPUs with Hyper Threading, but also we re-ran the tests with the AMD FX-8150 itself.
Setting the operating frequency to 4 GHz there are the same considerations set forth above. Disabling the core parking the performance of the other Intel CPUs with HT increased significantly.
Software product from Maxon that allow, through the development of three-dimensional images and content, to test the CPU performance. This release allow to test the CPU using a single core, or all the cores inside the processor.
Going to another type of applications, rendering in particular, we see that AMD does not shine in some respect to others, especially those from Intel. Rendering on a single core (1 CPU) it loses a lot towards Intel rival but you can see as well towards the 1100T AMD CPU, with Thuban architecture and characterized by the presence of 6 physical cores, does not gain a lot on the single core test. Even the Speed-up test shows a score of 5x, which is very similar to the one scored by the Intel i7-3960x, that we remember to be 6 core.
Other very significant graphic, even in this case we note that the total score (Rendering xCPU) is at the AMD 1100T level. The score of single core instead sees again a step back, where it marked a score lower than a Thuban, but also against the Phenom II X4 CPU.
Software product from Maxon that allow, through the development of three-dimensional images and content, to test the CPU performance. This release allow to test the CPU using a single core, or all the cores inside the processor.
Same scenario as with the R10 version of this benchmark: on the single core we see a score of 1.03, or +0.03 compared to 1100T Thuban. While MP scaling ratio does mark a very good score of 5.83 x even if it is in fact a very similar score to the 1100T, a sign that this new architecture is not yet properly exploited by this benchmark. The distance on the single core from the Intel Sandy Bridge is very high, the Intel i7-2600K is +0.50 points of score on a single core. Objectively, we would have expected an higher score, but several considerations must be made on the production process not live up with expectations as fully explained in the given page.
The situation is quite the mirror image of what has already been seen in the case of the R10 version of the Cinebench. Bringing the CPU to 4 GHz, it does mark a step back once again on both the overall score and the score made on a single core. From here can be seen as the IPC of the single core CPU has dropped compared to the old architecture AMD: 0.99 is the lowest score ever on a single core.
X264 Benchmark HD 3.0
Software to measure CPU performance using x264 video encoding.
In this benchmark, we can see a really good score on 2nd Run, it is in fact higher than the Intel i7-2600K, but the difference in 1st Run rather reverses and does mark a step backward. We can however say that the overall performance is very similar, making an average, to the Intel reference.
At a frequency of 4 GHz, one can observe that while in the 2nd Pass maintains a degree of advantage with the 1100T and loses a little something towards 2600K, in the 1st pass once again we see how the CPU with Thuban architecture is forward, albeit little.
Handbrake 0.9.5: Multi-threaded video encoding software with which we will transform a file (a movie) to MP4; the process includes x264 video encoding, FAAC audio encoding and muxing into final MP4 container. The time it takes for the CPU to perform this task will be taken into consideration.
Real test that brings to the various CPU cores a movie to encode. As you can see, the AMD CPU does mark a time very similar to the Intel i7-2600K with a very little deviation of seconds. A sign that the architecture, if properly exploited by the software, can give the best.
Looking the test performed at 4 GHz, there is nothing more than point out what we have seen already in default test. The AMD FX-8150 achieves a good score, springing forward with a good margin than other CPUs.
Fritz Chess is the strongest chess engine and the evolution of what he faced the world champion Vladimir Kramnik in Bahrain in October 2002. Millions of chess fans watched the games live on the Internet. Fritz was able to equalize the match (4:4), and Kramnik showed all his respect for the performance of Fritz when he said: "DeepFritz is stronger than deepblue!" The new engine is based precisely on that Fritz8 version of "Bahrain" and was further improved in positional play.
Situation at 4 GHz does not change from before, the FX CPUs is located halfway between the i7-2600K CPU and the i5-2500K CPU.
Blender is an open source program modeling, rigging, animation, compositing and rendering of three-dimensional images. It also has features for UV mapping, simulations of fluids, coatings, particles, and other non-linear simulations, creating applications / games 3D. It is available for various operating systems: Microsoft Windows, Mac OS X, Linux, FreeBSD, along with unofficial port for BeOS, SkyOS, AmigaOS, MorphOS and Pocket PC. Blender has a robust feature set comparable in features and complexity, other popular programs for 3D modeling.
Again, using a single processing core, the loss in performance is particularly strong in relation to other CPUs.
POV-Ray is a ray tracing program available for a variety of platforms. It was originally based on DKBTrace, a program developed by David Kirk Buck and Aaron A. Collins. He was also heavily influenced by the ray tracing program Polyray with the consent of its author Alexander Enzmann. In newer versions the rendering engine has been updated to allow the calculation of the deep global illumination, caustics, and elements to create particle clouds, fire, steam. Now makes use of multithreading, then takes advantage of multicore processors on the computer, or multiple processors.
This benchmark highlights the excellent Bulldozer architecture that, when used properly, is able to express its best. You can see that in this case it can stand in front of an i7-2600K, although very marginally, but it is a good sign given the good multicore results of the architecture.
Again, all performance aligned to the Intel i7-2600K.
It is a software that can handle real numbers, complex numbers, intervals, matrices and arrays. Track 2D/3D diagrams and includes a modern programming language. All versions of Euler are freeware with GNU. EULER is very similar to Matlab, but the author points out that this is not a clone. Euler 3D is used for analysis of fluid dynamics, with a freely available integrated benchmarking mode that uses all available cores and threads of the processor.
Sisoft Sandra: Data Encryption
SiSoftware Sandra (System Analyser, Diagnostic and Reporting Assistant) is a diagnostic tool that can provide detailed information about your computer hardware and software configuration. A year ago, SiSoftware Sandra 2010 released offering full support to Windows 7, to 18 months after the launch of Windows 7, we have witnessed a real coronation of the PC to the media hub. A few months ago was released a benchmark dedicated to the Blu-Ray, now the authors have focused on delivering a new benchmark: the Media Transcoding based on Windows 7 Media Foundation. To all was given a new benchmark (GPGPU Cryptography) which allows a direct comparison between the performance recorded by the CPU (using the set of encryption available) and those of GPGPU.
Very interesting test, which highlights the new architectures that have hardware encoding for data encryption, using SHA and AES 256 instructions. You can see that in general the old architecture lacking of hardware encryption are far behind compared to other of new generation. But we see how AMD A8-3850 and Intel i3-2120 even if of the new architecture they are deprived of them. Instead observing Bulldozer we see clearly how hardware encoding is present, but the result is just below the Intel i7-2600K both with HT enabled and with HT disabled.
The situation with the CPU operating frequency at 4 GHz does not change from before. The CPU is located behind the Intel i5-2500K.
Consumption trend with changes in the Vcore and production process
Assuming that the CPU consumption is 125W under Lynx full load, considering the efficiency of the VRMs and the power supply and considering 1.2V as a reference Vcore and 250W as reference consumption for the entire system, you can calculate an approximation of the power dissipated by the CPU to each of these supply voltages. With Vcore of 1.1 volts, the system absorbs 215W, 35W less. Considering the loss in the VRMs and the power supply, this translates into just about 20W less absorbed by the CPU, i.e. 105W. The Vcore of 1.1V was chosen for the following reason: in ISSCC declarations of February 2011, a Bulldozer module was credited with a supply voltage between 0.8 and 1.3 V. From the AMD data sheet we know, however, that now a module Bulldozer requires 0.9 to 1,415 V depending on the frequencies. This leads us to suppose that the 32 nm Global Foundries production process requires approximately 0.1V more than expected. If Global Foundries would, over time, improve the process so you can get Bulldozer modules operating at 1.1 V, this would mean that the 4 modules CPU would consume a maximum of 105W at 3.6GHz. Not to mention that an improvement of the process may also lead to reductions in leakage. And if this process would improve to the point where they can afford higher frequencies than 3.6GHz with 1.1V, the upper limit would be given only by the 125W TDP. Assuming the power consumption goes linearly with the frequency at the same Vcore, we would have a default frequency of 4.285Ghz, i.e. close to 4.3GHz, up to 700MHz. This could have been the default frequency of Bulldozer, with a Vcore of 1.1V, if the production process had been more mature. We hope that Global Foundries is able to improve its 32 nm process in order to approach or exceed this level of consumption, whereas Intel with a 32nm HKMG process, but bulk, can get 3.4GHz and beyond on an architecture with a much higher FO4 (24-26 the FO4 of Sandy Bridge, against about 17 of Bulldozer), with a Vcore of 1.2V. On the Intel production process, a CPU with FO4 of 17, i.e. at least 40% lower than that of Sandy Bridge, excluding consumption problems, could go at a frequency of at least 30% higher than that of Sandy Bridge, at the same Vcore. That is a CPU with FO4 equivalent to that of Bulldozer, implemented on Intel's current production process, at 1.2V, excluding consumption problems, could go at over 4.4GHz.
On this page we decided to investigate a topic that often leaves a lot of confusion in the reader, that is how the performance change with varying type of memory installed. In this case, type refers to modules that differ in the operating frequency. For all three platforms (FX-8150, i7 2600K and i7 3960X) we decided to perform computational tests with memory set at the following frequencies and relative timings.
- 1333 MHz CL. 7-7-7-20 1T
- 1600 MHz CL. 8-8-8-22 1T
- 1866 MHz CL. 9-9-9-24 1T
- 2133 MHz CL. 10-10-10-26 1T
The reference software we primarily used was AIDA64, where we analyzed the bandwidth of read, write, copy and access times to memory. The second software is WinRar 4.01 (stable version), which typically can achieve good performance increase with faster memory. The last software we used is Cinebench R11.5, where we checked the score on the multi-core.
The second reference software is the known compression/decompression program WinRAR, where in effect for all architectures have shown increased performance rising the frequency of system memory. You can see below the graph we have achieved:
From the graph it is possible to deduce that for all the platforms we have the best results from using 2133 MHz memory, this is easily explained by the fact that WinRAR is a software which improves performance when you go to use the high frequency memories, which provide therefore a greater bandwidth. We note also that the AMD platform is less affected by this factor, in fact, going from a frequency of 1333 MHz at a frequency of 2133 MHz for the memory the increase is only 300 KB/s.
With Cinebench is clearly seen that the memories do not have any effect on the final result, in fact for all three platforms we have no substantive differences between using 1333 MHz memory and the use of memory at 2133 MHz.
When it comes to top range CPUs, one often wonders what advantages can make with a particularly performer graphics subsystem. On this page we will see, using two current top range graphics cards, what happens varying the operating frequency of the CPU starting from the default up to well 5 GHz. The system has two Nvidia GeForce GTX 580 in SLI mode. The reference CPUs, in addition to the AMD FX-8150, are always a Core i7-2600K and the current top-end Intel Core i7-3960X. The tests were performed first setting the frequency for each of them by default with turbo on, then we set a frequency of 4 GHz and finally a frequency of 5 GHz. In the last two cases the turbo for obvious reasons has been disabled for all CPUs.
Since most of the games currently on the market are unlikely to be limited by the CPU of this caliber, we decided to use some graphics benchmarks that use the CPU more intensively, heralding what could be the load application of future games. These tests will also be particularly interesting for all fans of overclocking, to see which platform is best suited to get the best scores. The software that we used are the well known 3D Mark in both Vantage and 2011 version. We tested the systems with the benchmark Unigine Heaven 2.5 setting Full-HD resolution and setting the tessellation on Extreme. We see below the various graphs generated.
Looking at the 3D Mark Vantage we can see how the scaling of Intel is less marked than the AMD, especially when switching from 4 GHz to 5 GHz, in this regard we can be seen the slope of the segment generated by the graph.
Turning to the 3D Mark 2011 instead we see an extraordinary inclination of the segment generated by AMD, where the frequency of 5 GHz is also going over the Intel i7-2600K. Considering the excellent overclocking margins of the FX-8150, in this test the Bulldozer platform could track particularly interesting records with respect to Intel platforms.
Noting however the fps generated by Unigine, which reflects a more real gaming, let's see how you have a good increase from AMD CPU in going from the default frequency to 5 GHz On the other hand it must be noted that the fps generated are lower than the other CPUs in question, although as you go up the frequency, the differences between the various CPUs decrease. In particular, if we refer to the Intel i7-2600K we start from a difference of about 22 fps at default, while at the frequency of 5 GHz the difference is reduced to about 14 fps. This result means that the scaling of the FX-8150 is very good, but in any case the fps generated are a bit distant from the other CPUs. Again the AMD proposal could become particularly interesting at frequencies above 7 GHz, that can not be reached by Sandy Bridge and Sandy Bridge-E systems.
CPU frequency scaling
An interesting test we conducted on this processor has up to verify the scalability of the operational frequency of the CPU from a base frequency. Also in this test we compared three CPUs; other than OCTA-core FX-8150 we have included in a comparative the Intel Core i7-2600K with quad-core architecture and an Intel i7-3960X with six-core architecture. Both CPUs are equipped with Hyper-Threading Technology. We started with a base frequency of 3 GHz for all CPUs to arrive at a frequency of well 5 GHz. We have used as software a very good compression/decompression program 7-zip, and the popular rendering program Cinebench R11.5 that uses the Cinema 4D rendering engine. The results were very interesting and not entirely foregone. Below you can see the graphs created for the two software used:
Looking at the first graph, 7zip, you can see that all CPUs have an excellent performance as the frequency increases. The CPU AMD has produced a near linear scalability, the sign of the excellent internal architecture; however, we note that the Intel i7 2600K is in fact able to increase, increasing the frequency, the gap differential compared to AMD. Finally, the scalability of the Intel i7 3960X is really impressive and manages to have a clear dominance over all CPUs.
Totally specular situation with Cinebench test that shows the same results seen with 7zip. The AMD CPU although showing a very linear scalability, can not keep up with rivals. And the gap at 5 GHz is much more marked against the other CPUs.
Continuing with the analysis of scaling, this time we decided to monitor the consumption of the entire system at various frequencies. The AMD FX-8150 was monitored at default both with turbo active and both with turbo off. Then we set an operating frequency of 4 GHz on all cores by turning off the turbo. At this frequency we did not need any kind of overvolt and then was left on default vcore. Finally we got up the operating frequency to well 5 GHz by assigning a voltage of 1.52v, which is necessary for the system to pass at least 10 complete cycles of the Lynx stress test. All consumption measurements were made by Lynx go running for at least 10 complete cycles. The comparison was made with two other Intel CPUs, both of Sandy Bridge architecture. The first CPU is an Intel i7-2600K, with 4 cores and 8 threads and compatible with LGA 1155 socket. The second CPU is an i7-3960X but with 6 cores and 12 threads and is compatible with LGA 2011 mainboards.
For the i7-2600K up to 4 GHz, it was not necessary any kind of vcore overvolt, same thing was true for the i7-3960X CPU. Increasing the frequency to 5 GHz our i7-2600K has needed well 1.60v (unfortunately this is an unfortunate sample), while the other Intel CPU, the Core i7-3960X, we assigned 1.48v.
The consumption of the systems, equipped with an SLI of GTX 580, were measured upstream the power supply with an high-precision current clamp. Finally, remember that Lynx is one of the heaviest stress test programs in circulation today, and it is a situation that can hardly be repeated in a desktop environment.
At idle we measured a smaller consumption for the Intel i7-3960X solution with only 138W, following with 20W spacing each other are the other two CPUs. Looking at the load situation at default CPU and active turbo we can see the power consumption very close to each other between the 3960X and FX-8150; the first has a consumption of 283W while the second 266W, the Intel i7-2600K CPU is rather more distanced with 209W.
Turning off the turbo for all CPUs, we note that for the FX-8150 consumption are increased (quite abnormal behavior), while for Intel CPU consumption down slightly or remain the same as in the case of 2600K. The reason for this increase on the FX-8150 may be related to the TDP management system, built-in with turbo that could automatically turn off some tricks to limit consumption of Bulldozer architecture.
Bringing the frequency to 4 GHz, which did not include an increase in operating voltage required for any CPU, we can observe that in this case the consumption recorded by AMD are very close to the Intel i7-3960x with a minimum difference. In this case, the i7-2600K CPU shows a consumption of less than the other two with a gap of about 50W.
Further increasing the frequency and pushing up to 5 GHz, overvolt is required for all CPUs. For the FX-8150 the vcore was equal to 1.52v; for the i7-2600K CPU voltage of 1.6V and finally for the i7-3960X CPU voltage of 1.48 v. Again we see how the two top range CPU of the manufacturers are marking their consumption very similar and this time the i7-2600K, despite the high vcore, does mark a radical departure from the other with a difference of more than 200W.
Old school overclock
AMD with the new FX CPUs did not want to betray the overclocking enthusiasts, especially extreme, giving much satisfaction to those who want to engage in this practice, despite a bit disappointing performance. Already before the official presentation, AMD experts had recorded the frequency record, using an FX-8150 CPU cooled with liquid helium. This record was recently beaten with the same top of the range AMD CPU, reaching the incredible frequency of 8585 MHz! The AMD FX CPUs, if properly cooled, tolerate very high voltages, above 2V.
The temperatures are generally the main hurdle to reach high frequencies. Extreme cooling systems are especially useful on these CPUs, unlike what is seen with Sandy Bridge and Sandy Bridge-E CPUs, that even lowering the temperature below freezing, can not earn much room for overclocking.
As for overclocking for daily use, overclocking margins are pretty good, allowing you to reach without too much difficulty 4.6/4.8 GHz and 5 GHz in some cases; in all cases, however, is necessary a good cooling system, since this CPU consumption is very high at full load, especially if you apply a consistent overclock and overvolt. With air cooler of modest size obviously the overclocking margin decreases.
The overclocking of the FX CPU is made in rather than classic mode. We can adjust several parameters: the bus frequency (HTT), the CPU multiplier (FX CPUs are unlocked), the multiplier of the Hyper Transport link and Northbridge multiplier. The voltages of greater interest for the overclock are obviously the Vcore and partially the CPU-NB.
In general, the easiest way to overclock is to increase the CPU multiplier, but the most profitable, both in terms of performance and frequency, include maximizing the bus (HTT).
Using the press kit that AMD has given us and in particular the Asus motherboard you can do overclocking effectively. Asus ROG series has introduced the RC TweakIt software that greatly facilitates the task of overclocking our components. Since the 'FX-8150 is a 32nm CPU, you can easily reach the frequency of 4.5 GHz. More difficult will be overcoming the frequency of 4.8 GHz with air cooling systems; the behavior will depend very much on the goodness of our CPU. The first thing you need to do is disable the various energy-saving technologies and set the parameter of the Load Line Calibration on high to stabilize the supply voltages under overclocking. Here are two screenshots of the BIOS in which these parameters are present.
After this, the next step is to enable communication by moving the micro switch selector to ON to allow the connection of the RC TweakIt application that we installed on a laptop/external PC and connected to the motherboard via the white USB cable supplied with. Here is a macro of the position of the switch on the PCB of the Crosshair Formula V and the user friendly software interface provided by the RC TweakIt software provided by the Crosshair Formula V.
We went up in frequency step by step by acting both on the voltage supplied to the processor (Vcore) and both on the bus (HTT). Setting as a starting point, from BIOS, the multiplier to x23, we climbed up to get the results shown below. Of course, an eye should be kept at temperatures, but the 5GHz for those on a liquid cooling are easily accessible if the processor is cooled by a cooling system designed for this purpose.
Following, with our overclocking team we decided to go further with the frequencies and operating voltages. This is of course possible using extreme cooling. To this end, we used a Cascade phase change by Dimas: this system allows you to get several degrees below zero °C, reaching even -120 °C without load. Below you can see the various screens that we performed with this system, pushing well beyond the 6 GHz for some tests.
To achieve the 7 GHz are required cooling systems with liquid nitrogen or liquid helium, that we have not been able to experience due to lack of time.
During this analysis several interesting aspects of this CPU emerged.
Starting with a pure consideration, as fully explained in the page "Consumption trend with changes in the Vcore and production process", AMD Bulldozer architecture was to be born with higher operating frequencies well within the limits of the TDP. This unfortunately has not happened since the Global Foundries Foundries have not been able to maintain this standard that were in AMD projects and the CPU required a much higher vcore. The 32nm have been very difficult in production and has had to make certain compromises especially in terms of frequencies and voltages to avoid to exceed the TDP limit. This has made sure that the CPU is very different from initial expectations. The CPU has to be evaluated in its current implementation, however, and compared with the competitor CPUs in these today conditions and in particular with the software currently on the market, that still does not exploit FMA4 and XOP instructions.
Focusing on pure performance, as seen in the previous pages, the FX-8150 CPU often is in most cases a little under the Intel rival reference CPU, which is represented by the i7-2600K, only in a couple of cases is able to be on par or slightly higher than the latter. Most obvious is a weak point derived from a relatively low per-core IPC, as can be seen in the tests: in some cases we also recorded a ipc under the older generation Thuban.
A clock to clock comparison sees the new AMD CPU to be far behind Intel CPUs and in some cases even behind the 1100T. A result that surprised us, and surely we could not expect from a new CPU.
Frequency scaling in the FX-8150 shows an excellent behavior; in 3D stress tests we remark really excellent result in 3D Mark 2011 where it scales really well. But in 3D Mark Vantage and in the Unigine, while retaining a good linear trend, sees the CPU score not up to the scores of other CPUs. In applications such as 7zip and Cinebench however scaling is a little less than the Intel CPU. Varying the type of memory installed, did not bring any major changes in the software used; the same was also valid for other CPUs.
Let's go instead to a negative point, consumption: they are very high when compared to the performance. If idle has a good behavior, we see a load consumption already aligned with the Intel i7-3960x, and this constant association with the Intel top range is also measured at 4 GHz and even at 5 GHz. The difference with an Intel i7-2600K set to a frequency of 5 GHz is well 200W, an high absolute value, even if we consider that Intel CPU was not particularly lucky and required a higher vcore than other CPUs of the same type. That efficiency is very low it is a fact known to AMD too, which has now accepted the defeat of Bulldozer in a fairly resigned way, announcing the battle, however, on other market segments.
The overclocking is definitely a strong point of this CPU, as long as to have a good cooling system. With a good liquid cooling system the 5 GHz are within easy reach, but it becomes difficult to go beyond this frequency while maintaining the same type of cooling.
The fun begins when you decide to put the CPU under pressure going to extreme cooling systems. In our case with a Cascade type phase change we have reached far beyond the 6 GHz frequencies: a result of good relief. The AMD FX-8150 CPU currently holds the record for the absolute frequency of microprocessors where they were well over 8 GHz with a cooling system more advanced as helium or liquid nitrogen.
For those about to buy a new CPU, the FX-8150, at a price of about € 230, is probably a good upgrade for those who already have an AM3+ system, particularly from the perspective of future longevity, while the 1100T at a price of 160 € is still a viable candidate. For those who are preparing to assemble a new system, proposals for Intel LGA 1155 are undoubtedly the most interesting. These considerations are valid especially for the majority of our readers generally interested in gaming. The situation may change for some professionals, who may be interested in having 8 real core at an affordable price, particularly attractive for those who need to virtualize many systems simultaneously. The FX-8150 could also prove very useful for those who make extensive use of professional applications optimized for the new instructions implemented in Bulldozer, applications, unfortunately, still very limited in number.
Translated by Marco Comerci