by Ilya Gavrichenkov
11/01/2006 | 10:31 PM
At the IDF that took place in San Francisco about a month ago we have already got our first experience with the new quad-core Intel processor also known as Kentsfield (for details see our Intel Kentsfield Preview). However, that was hardly full-fledged experience. Everything we did was strictly monitored by Intel guys, who wouldn’t let us veer even a little bit away from the plan. Luckily, we have finally got some Kentsfield samples for our lab, therefore, we are proud to share with you the results of real indepth test session. Of course, this review was only possible because these processors are officially announced today. So, if you get really interested in the new Intel Kentsfield processor, you should be able to easily buy one for your system or to get a complete Kentsfield based platform from one of Intel’s partners.
<%BANNER[article]%>Therefore, we can state that today we witness the beginning of new processor expansion into the desktop market, only these CPUs have more than two cores onboard. Note that it took Intel only one and a half years to move from dual-core processors towards quad-core processor designs, which proves that Intel is being very serious about pursuing higher parallelism. However, I would like to stress that it was not the contemporary software that stimulated the introduction of additional cores into the new processors. Even though there are quite a few applications that can split the work into four parallel threads quite efficiently, they are all of the same type and deal mostly with media content processing or 3D rendering. So, the idea to increase the number of processor cores comes solely from the processor developers. They need to continue developing their solutions, but they can no longer raise the processor working frequency, because of the growing CPU power consumption. In other words, Intel’s new quad-core processor launch is intended to demonstrate the software developers that the time has come for dramatic revision of common working algorithms towards higher parallelism.
At this time Intel will position Kentsfield processors for the top price market. So, they will be primarily targeted for such computer enthusiasts as hardcore gamers. However, so far we cannot think of any 3D game that could be able to utilize all four cores. Luckily, this situation should change next year when at least five games should come out that will fully support Kentsfield. These will be: THQ Supreme Commander, Remedy Alan Wake, Valve Half-Life 2: Episode 2, Epic Unreal Engine 3 and Ubisoft Splinter Cell: Double Agent. So, I believe we have every right to say that even the developers of traditional single-threaded applications start adapting their software for the realities of the today’s processor market.
I have to say that Intel managed to get a way ahead of AMD with the quad-core processor announcement. AMD is planning to launch similar designed CPUs no sooner than in the middle of next year. However, in the meanwhile, they will roll out the so-called 4x4 platform that will include two dual-core processors that should help them retain their positions against the background of the competitor. We are definitely going to pay special attention to this new AMD platform in the nearest future, but today let’s talk about the new Intel’s quad-core CPU aka Kentsfield.
Intel introduced the first dual-core processors on Core microarchitecture in mid summer. In fact, not so much time has passed since then, especially from the IT industry standpoint. However, Intel is ready for the next step forward: the launch of new CPUs with quad-core design based on the same Core microarchitecture!
It wasn’t that hard for Intel to introduce Core microarchitecture into the quad-core processor design. They used the same approach they worked with in the times of Presler CPU. In fact, Kentsfield is actually none other but a combination of two Conroe (Core 2 Duo) dies within the same processor package. In other words, the new quad-core Intel processor is a combination of two dual-core CPUs with Core microarchitecture.
Note that this is a very advantageous approach that makes a lot of sense. For example, it helped Intel save quite a lot on engineering and technology, which led to very early arrival of quad-core microarchitecture to the market. At the same time, Intel has also taken care of possible production issues that are typical for the early manufacturing stage of new semiconductor dies. Of course, the yields during manufacturing of two dual-core Conroe dies will be higher than in case Intel had to manufacture single dies twice as big in size and featuring about twice as many transistors.
Here I have to add two more things. Firstly, by using two dies instead of one Intel managed to save about 12% of the die size. And secondly, they got great opportunity to select the best dies for their quad-core solutions. Kentsfield is expected to utilize dies with lowest heat dissipation, which will create very attractive thermal envelope for these CPUs.
So, Kentsfield should be regarded as another CPU on Core microarchitecture that we have already discussed in detail in our article called Getting Ready to Meet Intel Core 2 Duo: Core Microarchitecture Unleashed. It boasts all the same advantages as Conroe processors with one single exception. Since the quad-core newcomer is built of two dual-core physical semiconductor dies, its L2 cache memory will consist of two 4MB parts. Each 4MB part is shared (thanks to Intel Advanced Smart Cache technology) between the corresponding pair of cores. In other words, the cores located in two physically different dies will exchange and share data in an old-fashioned manner, i.e. via the system bus and RAM.
At first, that is starting today and until the end of this year, Intel will be shipping only one Kentsfield modification aka Core 2 Extreme QX6700. This CPU will cost $999, i.e. it will be a quad-core competitor to the dual-core Core 2 Extreme X6800 processor.
The specifications of the newcomer look as follows:
If we compare Core 2 Extreme QX6700 against the top of the line dual-core processor on Core microarchitecture, we will see that Core 2 Extreme QX6700 works at 266MHz lower frequency and features 75% higher typical heat dissipation. However, both CPUs cost absolutely the same, so that the user is facing a very difficult choice between these two completely different solutions.
The diagnostic CPU-Z utility reports the following about Core 2 Extreme QX6700:
As we see, Intel released new B3 core stepping specifically for quad-core processors: the latest Core 2 Duo core stepping is currently B2. Other than that the screenshot shows what we have expected.
Together with the Core 2 Extreme QX6700 quad-core processor we also received a slightly slower, non-extreme Kentsfield model aka Core 2 Quad Q6600. This processor hasn’t been officially announced yet, it should come out in the beginning of next year.
Core 2 Quad Q6600 differs from Core 2 Extreme QX6700 by lower clock speed and lower price point that is expected to be around $850, according to some preliminary data.
The specifications of Core 2 Quad Q6600 are given in the table below:
Intel Core 2 Quad Q6600 | |
Clock frequency | 2400 MHz |
Cores | 4 |
Packaging | LGA775 |
Vcore (max) | 1.35 V |
Bus frequency | 1066 MHz |
Typical heat dissipation | 105 W (?) |
L2 cache | 2 x 4MB, shared |
Production technology | 65 nm |
Intel Hyper-Threading | None |
Intel Virtualization Technology | Yes |
EM64T (Enhanced Memory 64 Technology) | Yes |
EIST (Enhanced Intel SpeedStep Technology) | Yes |
And here is what the CPU-Z utility reports:
Note that the “non-extreme” Kentsfield processor will also boast slightly lower typical heat dissipation than the top-of-the-line Core 2 Extreme QX6700. Therefore, if you do not want to face the same issues as with NetBurst based CPUs, Core 2 Quad Q6600 may be a better choice for you.
Well, it doesn’t make a lot of sense to dig deeper into the theory about the new quad-core processors. We have already discussed Core microarchitecture in our numerous articles devoted to Kentsfield’s “halves”. As for the peculiarities of quad-core CPUs, we have also dwelled on them in our IDF Fall 2006 Coverage. So, let’s move to the most exciting part of our review: the practical test session.
When we worked with Core 2 Extreme QX6700 and Core 2 Quad Q6600 we decided to test their performance not only in multi-threaded applications. Since multi-core processors allow running a few resource-hungry applications simultaneously, we felt it would be interesting to check out situations like that, too. But let’s start from the very beginning here. Before we start, please take a look at the configuration of the systems that were assembled for our tests:
The mainboard BIOS was set for maximum performance.
The first test we performed let us make very precise conclusions about the quad-core CPUs. These processors are definitely faster than the dual-core CPUs. And this performance advantage is first of all noticeable in digital content creation and processing tasks. In fact this is quite natural. Most applications of the kind are optimized from multi-threading prospective, therefore Core 2 Extreme QX6700 appears almost 8% faster than Core 2 Extreme X6800, even though the dual-core CPU works at 10% higher clock speed.
As for the performance in typical office applications, the results demonstrated by Kentsfield are not as rosy anymore. The workload in office applications is very rarely multi-threaded, so the performance can hardly be increased by adding a few more cores. By the way, I would like to remind you that when we transition from single-core processors to dual-core ones, the Office Productivity index in SYSMark 2004 SE still rises. In other words, these data suggest that it doesn’t make any sense to use more than two processor cores in office applications.
The popular PCMark05 benchmark does support multi-threading, however, it implies not only the ability to process two computational threads simultaneously. There are two subtests included in this benchmarking suite that create four simultaneous threads. As a result, Kentsfield shows better performance index in this benchmark than any dual-core processor.
The victory in 3DMark06 results from the processor subtest that affects the total score significantly. The results of the CPU test are given on the second diagram. This test uses multi-core architecture to calculate physics and AI for multiple objects interacting with one another. Of course, a task like that can be easily split into several parallel threads, and the obtained results prove this point very well.
Of course, relatively old games like Far Cry and Half Life 2, that do not have even a hint of multi-threading support, cannot benefit from four processor cores in any way. Quake 4 is known to support dual-core CPUs, however it didn’t impress us on a Kentsfield based system. Looks like the so-much-appraised multi-threading support in Quake 4 was designed specifically for dual-core solutions, so the game can work only with two computational threads at a time. However, the implementation of multi-threading support in F.E.A.R. allows Core 2 Extreme QX6700 to get slightly ahead of Core 2 Extreme X6800. Although these results are still far from being a tremendous success. Only the next-generation gaming engines are expected to fully implement the potential of quad-core solutions for environmental physics and AI calculations, and the first games featuring these engines should come out next year.
Multi-threading support found its way into video and audio codecs quite a while ago already. However, the results below demonstrate that far not all the codecs can really load all the four cores with sufficient amount of work. Most of them generate only two threads and hence Kentsfield cannot boast any advantages here. However, there are a few absolutely opposite examples, too.
Some codecs recognize all the four processor cores in the system and prove highly efficient in Kentsfield based platforms. The relative advantage of Core 2 Extreme QX6700 over the predecessor, Core 2 Extreme X6800 makes 20-35% in Xvid and TMPCEnc.
However, it turns out that not all the codecs can still engage all the four processor cores simultaneously. Many of them can demonstrate a significant performance boost only when we switch from single- to dual-core processors which makes Kentsfield absolutely useless in this case.
In other words, even the applications that we used before to demonstrate the advantages of dual-core architectures may not serve the same purpose in case of quad-core CPUs performance analysis.
As we have already said above, the digital content creation and processing tasks are the first ones to benefit from the quad-core Kentsfield processors.
For example, Core 2 Extreme QX6700 is about 18% faster than Core 2 Extreme X6800 working at 266MHz higher frequency in Adobe Photoshop CS2.
And the applications for non-linear editing demonstrate even more significant performance boost due to quad-core CPUs. New Kentsfield processors appeared 60% faster than Conroe CPUs.
To check the performance of our testing participants in professional applications we used the latest 3ds max 9 version.
The performance during final rendering scales perfectly depending on the number of the processor cores. Kentsfield based systems may become an excellent choice for 3D designers. Rendering runs 30% faster when we switch from Core 2 Extreme X6800 to Core 2 Extreme QX6700. Unfortunately, we cannot say the same about the work in viewports. It is not the number of processor cores, but the clock speed that affects the performance here the most.
Kentsfield proved even more efficient in Cinebench that demonstrates final rendering speed in another popular suite aka Cinema 4D. Core 2 Extreme QX6700 performs 50% faster than Core 2 Extreme X6800.
The latest versions of popular WinRAR archiving tool support multi-threading just fine. And Kentsfield is exactly what we need here: it ensures very decent performance boost.
The results in MATLAB suite do not get any better from the suite of four processor cores: the best result here belongs to Conroe based CPUs. However, I have to point out that Kentsfield doesn’t fall too far behind Conroe in applications that cannot take advantage of parallel data processing. The reason is the relatively small difference in working frequencies of the dual-core and quad-core processors: the top Kentsfield model is only one clock multiplier step away from Conroe – 266MHz.
We also decided to include a chess game benchmark that uses a popular Fritz algorithm to measure the CPU performance. The calculation of all possible moves for a given chess position turned out to be very easy to parallelize and hence could demonstrate the advantages of having four cores in the system. Really, if the working frequency is the same, then Kentsfield proves 94% faster than Conroe, which is close to the theoretical maximum.
In this section we performed a few tests aimed at revealing the systems performance where there are a few applications running simultaneously. We launched several resource-hungry tasks in different applications and measured the time it took the test platforms to complete them.
Here we were processing an image in Adobe Photoshop and at the same time compressing a folder with files using WinRAR utility. I have to admit that we were not at all surprised with the results. Multi-core processors coped with multitasking easily, leaving dual-core Conroe more than 30% behind.
Then we were editing video in Adobe Premiere Pro and encoding an MP3 file using Apple iTunes. Looks like the computational workload was higher than in the previous case. At least Kentsfield proved 93% faster than Conroe working at the same clock frequency.
The third test is probably the most sophisticated. Our test platforms were busy working on three tasks at a time: image editing in Adobe Photoshop, final rendering in 3ds max and video encoding into MPEG4 format. And once again quad-core processors proved to be the best choice having outperformed dual-core Conroe pretty noticeably. So, we have every right to say that multi-core processors will come in very handy in systems running multiple tasks at the same time.
Besides the performance of the dual-core and quad-core platforms running several parallel tasks, we also decided to investigate the influence of background processes on the performance in certain resource-hungry applications. We measured the fps rate in Quake 4 game with several copies of WinRAR utility running in the background.
At first glance the results appeared pretty discouraging. We would expect a CPU with more cores to be faster, no matter how many applications are running in the background. But the real state of things turned out different. Although Kentsfield based platform runs faster with a few applications in the background, we cannot observe the same tendency for situations with higher number of background apps. In case of 6+ applications in the background, the performance of the “primary” game drops faster in a quad-core system.
Let’s try to get to the roots of things here. It is important to understand that the performance of the game drops not because the background archiving tools eat up all the processor resources. Windows OS task manager distributes all processes very nicely, so that all background tasks get sent to the idling cores. However, all background tasks do not require only CPU resources. They also need access to other platform subsystems, such as front side bus and memory bus. This seems to be the problem for Kentsfield. Since this processor consists of two different semiconductor dies, they exchange data via front side bus and system memory. And WinRAR is a multi-threaded application that involves a few cores at the same time. So, if you have this utility running in the background it eats up some of the memory bus and front side bus bandwidth in order to exchange data between different application threads. Therefore, if you have a few copies of this program running on a Kentsfield based platform, the bus bandwidth gets used up much sooner than in a Conroe based platform where the data is exchanged via the shared L2 cache.
However, this is no tragedy at all. We used this relatively artificial test to show the potential bottleneck of Kentsfield architecture. It will be really hard to stumble upon the same situation in real life, because most background processes are not that resource-consuming. And if any “heavy-duty” tasks are running in the background, there are not too many of them, as a rule.
Core micro-architecture has already proven extremely economical and efficient from the performance-per-watt prospective. Conroe processors that we have already tested before turned out not only the today’s fastest desktop CPUs but also consumed the least power of all. For example, Intel declared the typical heat dissipation for their Core 2 Duo E6700 at only 65W. The new quad-core processors are built with two Conroe cores inside, so their typical heat dissipation doubled. So, the quad-core Core 2 Extreme QX6700 working at the clock speed of Core 2 Duo E6700 dissipates 130W of heat. Therefore, theoretically it boasts the same heat dissipation as Pentium D processors on Presler core. I have to admit that it might be a very alarming sign for those of you who very well remember what kind of cooling systems were required for the last NetBurst processors. However, before we make any final conclusions, let’s take a look at the practical benchmark results, maybe things are not as threatening as they might seem at first glance. So, we will measure the processors power consumption that equals their heat dissipation according to the energy conservation law.
As always, we used a special S&M utility to measure the maximum power consumption (you can download this utility here). We measured the current that goes through the CPU power circuitry. So, the numbers given below do not take into account the efficiency of the CPU voltage regulator laid out on the mainboard.
First of all, we measured the processors power consumption in idle mode. Cool’n’Quiet, Intel Enhanced SpeedStep and Enhanced Halt State power saving technologies were disabled in this test.
Well, we do not observe anything alarming in idle mode. Core 2 Quad Q6600 consumes almost as much as the dual-core Core 2 Extreme X6800 on Conroe core. As for the power consumption of the quad-core Core 2 Extreme QX6700, it is 8W higher, but still lower than the power consumption of Athlon 64 FX-64 processor in the same testing conditions.
Now let’s see what the results will be in case of 100% CPU utilization.
As we can see, the practical power consumption of Kentsfield processor is 75% higher than that of Conroe CPU working at the same clock frequency, which is pretty close to the theoretical numbers. However, despite this fact the new quad-core Intel processors are still more economical than the dual-core Athlon 64 FX-62 and the latest revisions of the dual-core Presler based CPUs.
In other words, you shouldn’t really worry about the high heat dissipation of the new Kentsfield processors. Our test session suggests that from the heat dissipation prospective Core 2 Extreme QX6700 is comparable with AMD Athlon 64 X2 5000+, and Core 2 Quad Q6600 – with the Energy Efficient modification of Athlon 64 X2 4200+.
In conclusion let’s take a closer look at the frequency potential of the new Kentsfield processors.
These CPUs can be overclocked in pretty traditional manner. You do exactly the same thing as you did with Conroe. Since Core 2 Extreme QX6700 is targeted for hardware enthusiasts, its features an unlocked clock frequency multiplier and hence can be overclocked somewhat easier than any other members of the same processor family. At least you can raise the CPU frequency above the nominal on any mainboard. This is however not true for the not yet announced Core 2 Quad Q6600 processor that features a locked clock frequency multiplier of 9x. So you will need a mainboard that can work stably at high FSB frequencies, if you want to succeed in overclocking this processor.
During our overclocking experiments we didn’t use any special cooling solutions. All tests were run with a popular Zalman CNPS9500 LED air-cooler. We used ASUS P5B Deluxe mainboard on Intel P965 Express chipset as our ultimate overclocking platform. Processor Vcore was set at 1.5V for both CPUs. As you remember, the nominal Vcore for our Core 2 Extreme QX6700 processor equaled 1.3V, and for Core 2 Quad Q6600 processor – 1.2V. During our overclocking experiments we manipulated only the FSB frequency, without changing the clock multipliers, even for Core 2 Extreme QX6700.
The results are the following: Core 2 Extreme QX6700 could run stably at 3.5GHz.
So, we can say that the top Kentsfield processor demonstrated pretty impressive frequency potential of more than 30% beyond its nominal rate.
As for the second CPU, the Core 2 Quad Q6600, it hit a slightly lower frequency of 3.42GHz.
Nevertheless, this is also a very good result, equal to 43% frequency increase above the nominal.
So, Kentsfield processors have every chance to be an interesting overclocker’s solution. Even though Conroe based CPUs can hit higher speeds, Kentsfield doesn’t fall too far behind them. Our testing participants could work stably at 3.5GHz without any additional cooling.
Summing up I have to state that it will be very hard to draw a final line today. The thing is that Kentsfield processors have evidently got far ahead of their time. There are not that many applications yet that could use the potential of all four cores and load them to the full extent. In fact, these are only 3D rendering tools, video editing tools and a few codecs. These are the few applications where multi-core architecture can show its real best and prove adequate to its theoretical potential. Since there are not that many optimized applications, Kentsfield processors cannot yet become the ultimate leaders from the performance-per-watt prospective. Dual-core Conroe based CPUs still retain the leadership here.
However, despite this fact, Core 2 Extreme QX6700 launch is definitely a success. Firstly, Intel was brave enough to push the multi-core concept into the market. Intel was the one to give software developers to understand that the time has come to revise their algorithms dramatically. The upcoming year 2007 should become a turning point: we expect a lot of new applications that would benefit from systems on multi-core processors.
Moreover, smart pricing policy makes Core 2 Extreme QX6700 a very attractive purchase today already. Its frequency is only 10% lower than that of the top Conroe CPU, Core 2 Extreme X6800. So, Core 2 Extreme QX6700 will be just a little bit slower than the predecessor in applications that do not support multi-threading. However, the are priced equivalently, so that the users looking at the price-to-performance ratio in the first place will be able to consider Kentsfield as a possible good choice. Especially since it will be extremely efficient in case of several tasks running in parallel even if they are not optimized for multi-threading. And in optimized apps, Core 2 Extreme QX6700 is unattainably fast.
So, even if you do not yet see any tangible benefits from the quad-core Kentsfield processors for your particular case, the situation will undoubtedly change very soon. This new Intel solution has big future ahead.