Intel has now launched its first processor with four cores and the new flagship Core 2 Extreme QX6700 has been in our test lab to endure our procedures. We take a closer look at Intel’s quad-core architecture and how the new processor performs.
The frequency hysteria that ruled the CPU market just some years ago is almost gone now and CPU manufacturers had to come up with new ideas when it came to increasing performance. Some of these ideas concerned optimizing the system around the CPU, which in most cases actually was slow and still are. Another idea is the use of several cores in a single CPU. Systems based on CPUs with two cores already cover a great amount of the CPU market with just a year after their release. Tendencies throughout the microprocessor industry shows similar results, parallel performance is the way to go in the future.
Today, Intel presents us with the logical successor to its dual core CPUs based on the x86 architecture and on this great milestone the name from several rumors found on the net during the previous months – Kentsfield. From the development name to the finished product the name has changed to fit in with the Core 2-series of CPUs. In stores, the Kentsfield can be found under two names, the Core 2 Quad and the Core 2 Extreme, both which will be examined here at NordicHardware today.
Let’s take a peak at what Kentsfield really is.
On the server market we’ve seen a lot of motherboards featuring two or more sockets to get a system with multiple cores. Dual sockets came to ordinary personal computers during the Pentium 3 era, not to a great extent, but it did exist nonetheless. A more attractive solution was to create CPUs with more than one core, but still using one socket. These dual core CPUs were released little more than a year ago from both AMD and Intel. Both companies chose a different approach to making dual core CPUs. As Intel chose to use the so called “dual die” technique, AMD chose to use one die with two cores baked together. From the release of the dual core CPU named Conroe, Intel also started to use single die (the two cores baked in a single die) CPUs. Now, when we look at Kentsfield, Intel decided to return to their former dual die technique, as the Kentsfield really is two Conroe dies crammed together thus having a total of four cores, also known as Quad Core. To get a clear overview please look below.
|Processor||Code-name||Circuits||Physical cores||Logical cores|
|Pentium4 800FSB||Northwood/Prescott/Cedarmill||1st||1st||2st (HT)|
|Pentium D Extreme Edition||Smithfield/Presler||2st||2st||4st (HT)|
|Core 2 Duo||Conroe||1st||2st||2st|
|Core 2 Quad||Kentsfield||2st||4st||4st|
The careful reader notices that there are a few exceptions, but these are very few which we will not expand our table with.
As we can see, Intel has used some different paths when it comes to creating multi core CPUs. The benefit from just cramming two dies next to each other is that you can use already existing dies, and you don’t need to create a whole new chip or die, which is very expensive. The flaw of this otherwise so great design is that the dies communicate using the older FSB interface. The FSB (Front Side Bus) also communicates between the CPU and memory, with the northbridge in between. The problem here however is that the FSB isn’t the fastest mean of transportation around and certainly wasn’t designed to be able to feed a Quad Core with the amounts of data needed to keep all cores working at 100%.
Conroe was however designed from bottom up to be a dual core CPU in one silicon die, where the two cores shares the same level 2 cache (also known as L2 Cache). Here the two cores can communicate using the extremely faster L2 cache, which is one of the reasons why Conroe performs as superbly as it does.
But why wait for a die with four cores? Why not use the older technique? The answer was, you guessed it, Kentsfield.
In other words, Kentsfield is the development name of Intel’s first CPU featuring four physical cores for personal computers and workstations. We’re using the term “four physical cores” because of Intel’s former attempts with Smithfield/Presler, where each physical core had a logical core too. These weren’t real quad cores but featured four logical cores using a technology known as HyperThreading. Despite the fact that the Kentsfield has four physical cores doesn’t make it look different from other s775 CPU’s, as shown above.
On the backside of the CPU we do see some minimal differences compared to the Core 2 Extreme next to it. The placement of capacitors and resistors has been altered, which isn’t unusual at all when it comes to new CPUs. We’ve even seen design specific differences between the different steppings from Intel as well. On the third picture we can see that Intel had to add a couple of extra layers, no doubt for the communication between the two dies and to provide them with power. The biggest difference between its Core 2 cousins is what’s under the IHS.
Pictures from www.thetechrepository.com
Here we can clearly see the difference between a CPU using a single die (left) and one using dual dies (right). To clarify, the Smithfield, Presler and Kentsfield CPUs uses the dual die principle (right picture), and what separates Kentsfield from the two other CPUs is that one die in the Kentsfield CPU features two physical cores. Whereas the Presler and Smithfield CPUs has, like we wrote before, one physical core per die but acts it acts as two.
Let’s continue our study of Kentsfield.
Since Kentsfield is simply two Conroe cores in one and the same package there isn’t much separating the architectures except from the fact that it has two such circuits. We’ve done a thorough dissection of this earlier on and simply direct you to our previous article for any deeper technical explanations.
Above we see a simple schematic of how Kentsfield has been constructed in comparison with Conroe. The lines with long dashes shows where the limit between processor and the rest of the system is and the dotted lines are the borders between the physical circuits. What we’re trying to illustrate here is how the two pairs are separate and communicate via the system bus in Kentsfield and that is also the interface through which the communication with the rest of the system is handled. One realizes quite fast that the two pairs are communicating quite fast internally, but not quite as fast with each other. Whether this affects the performance is something we will investigate later in this article.
Kentsfield will be launched on November 2 and is expected to hit the stores on November 14, in the form of the Extreme Edition XQ6700 and at the start of next year as Q6600. Both of these use a 266MHz/1066FSB and with a 4MB L2 cache per die and we have a total of incredible 8MB. The heat dissipation is for obvious reasons higher, which we will also take a look at later in the article.
Time to start the testing.
|Motherboard||Asus P5W64 WS Professional|
|Processor||Intel Core 2 Extreme QX6700|
Intel Core 2 Extreme X6800
Intel Core 2 Duo E6300
|Memory||OCZ PC6400 ATI CrossFire Edition|
|Graphics card||nVidia GeForce 7900GT|
|Power Supply||OCZ PowerStream 520W|
|Operating system||Windows XP (SP2)|
|Drivers||Intel Chipset Driver 220.127.116.111|
|Graphics drivers||nVidia ForceWare 91.47|
|Benchmark programs||SiSoft Sandra 2005 SR3|
Because of the short time we’ve had the processor the number of references are a bit limited, for the same reason we’ve been unable to create a better test suite with more multithreading focus. All of the processors are tested on the same motherboard with the same hardware and settings. The only thing we’ve done is to change processor and multipliers in the BIOS. We’ve simulated the Q6600 with the QX6700 by lowering the multiplier one step and the same goes for E6700 and E6600 which has been simulated with the help of the X6800 processor. We’ve also included an E6300, which is the cheapest model in the Core 2 series and is limited to only 2MB L2 cache. The graphics settings have been left to the standard default for the drivers and has not been changed during the testing.
We continue with some synthetic benchmarks.
Sisoft Sandra doesn’t necessarily say that much about the everyday performance of processors, but it indicates what kind of performance which is available if all applications were multithreaded. The latest version of Sisoft Sandra had some tendencies of not recognizing the quad core processors correctly and sometimes presented rather irrational results. When everything worked as it should we could see a clear difference between 2 and 4 cores.
Next is Everest.
Everest has a number of tests for measuring the performance of the processor. As we can see from the results the benchmarks are multithreaded and thus an easy win for the Core 2 Quad processors. We are close to doubling the results at the same frequency when comparing QX6700 with E6700 and Q6600 with E6600, which points to that we’re dealing with small datasets, and very few dependencies between the threads. The more threads a program has, the more there is to gain from a quad-core.
We move on to some more practical benchmarks.
With WinRAR we can see a clear difference when comparing with the dual-core processors. The advantage with multithreaded applications of today is that they are not moving from being single-threaded to just double-threaded to generally multithreading right away. This results in that quad-core processors, and even more cores in the future, will continue to result in better performance. With Cinebench we can see a nice increase in performance and from 1 thread to multiple the dual-core processors increase the rendering speed with about 87%, while the quad-cores go up by 210%. From the ideal case of 100% and 300% respectively we can see that it’s not possibly to take advantage of all of the theoretical performance, as the cores share resources such as cache, FSB and memory. This is a general problem for multi-processor systems and can be located with more or less all systems with multiple cores/processors.
We move on and take a look at MP3 compression.
Here we can see a clear example of an application which is just double-threaded. When dealing with media compression it’s exceptionally hard to create multithreaded programs because of the compression algorithms. Audio and video are often compressed in relation to how the audio sounds or what picture looks like relative to the rest of the file, which makes it impossible to juts split the file and let each thread compress each part. Therefore, we can’t see any gain when moving from two to four cores.
Next is 3DMark.
None of these benchmarks are multithreaded to any major extent, except one – 3DMark06’s CPU test, which you can clearly see. Theoretically we should get the same performance at the same frequency with the other programs, but as you can see the quad core processors slips behind the dual core versions in 3DMark2001 and 3DMark03, although with very small margins. We ran these tests a few more times to check, with the same results. Exactly why this is, we can’t say for sure, but we will continue investigating this matter.
We take a closer look at the processor performance with 3DMark06 and PCMark05.
Here we can clearly see how the CPU test in 3DMark06 is multithreaded and the quad core processors are performing far better than the dual-cores. PCMark05’s CPU test is divided into segments where some are single-threaded, some are double-threaded and some multithreaded. Therefore the results are not as impressive when moving from two cores to four as it was with the CPU test in 3DMark06.
Next are a few games.
Farcry is heavily limited by the graphics card of this system and it’s just the E6300 processor which doesn’t make the 7900GT the bottle neck of the system. The same goes for Doom 3 where it’s just the E6300 processor which deviates from the top scores. With Quake 4 we have a broader range, but we can see that the frequency is the deciding factor and not the amount of cores.
Finally, we will take a look at the power consumption of the processors.
We measured the system’s total power consumption with an ampere meter connected before the power supply. Thus it’s the power consumption of the entire system which is presented. To load the processor we used multiple instances of Prime.
The power consumption is as you can see significantly higher with the quad core models during full load. But we should point out that during full load we also have twice the performance. With normal systems you will most likely not even get close to these figures. We can see that the power consumption is higher that the difference between the TDPs of the processors which also is because of the rest of the system, such as FSB and memory, is loaded a lot more with the quad core processors. Another aspect is that we get a higher loss in the motherboard’s voltage regulators during higher load. You should thus consider these figures with a pinch of salt.
Let’s move on and summarize our Core 2 Quad experiences.
As we’ve seen in the benchmarks Core 2 Quad has its sweetspot with heavily multithreaded environments, hardly surprising. Because of the lower frequency, when compared to X6800, QX6700 isn’t quite able to keep up when there is just one or two threads. The FSB which is used by both the cores and the northbridge, and in turn the RAM makes it impossible for the processor to reach its full potential. When the choice is between a QX6700 and a X6800 the user should really think about what the processor will be used for. When it comes to heavy multitasking QX6700 is the obvious choice, while applications and games where the frequency is the decisive factor X6800 is a better choice.
Improved performance comes at a cost when it comes to the power consumption of modern processors. As we can see from our measurements of the entire system’s power consumption the quad core processors consume considerably more than the dual core siblings, with an increased heat dissipation following it. We haven’t been able to test Intel’s new cooler for the Quad series, which is a heatpipe construction for keeping the temperatures at bay. We ran all tests with a Zalman CNPS-9500, which handled all of the processors just fine, which should apply to all third-party coolers in the performance segment.
The Core 2 Quad processors doesn’t require any new motherboards, but all boards that supports Core 2 Duo will also be able to handle Core 2 Quad. The motherboards which are included are those based on the i975X and P965 chipsets, but some may require a BIOS update for full support. Up until today we’ve tested Intel’s BadAxe2, Asus P5W64 WS Professional, Abit AW9D-MAX, which all work with the latest BIOS updates. Intel says that the first generation BadAxe will also support the processor from revision 304, with an updated BIOS.
We didn’t have time for any thorough overclocking tests here in this article, but as yours truly, together with Marcus ’Kinc’ Hultin, visited Intel’s own lab in Kista north of Stockholm the past weekend and, the least to say, massaged an QX6700 and you can expect a follow up. Keep your eyes open for a longer overclocking article very soon. We can already reveal that the processor has a very broad margin with several ’heavier’ cooling setups.
Intel has stated that the Core 2 Extreme QX6700 processor will be available, in large numbers, in stores on November 14, but as it is an enthusiast processor the sales will still be limited when compared to the cheaper models. We haven’t received any exact pricing, but the Extreme series usually costs between $900 and $1,000, and this one should be no exception. The Q6600 model will not appear on the market until the first quarter of 2007 with same specifications as we’ve presented today. It’s too early to speculate about the price of this model.
There’s no doubt that Intel has once again released a performance ace for PCs, which will offer truly good performance in multithreaded environments. We see that games are overall rather poor to utilize the capacity of this processor, but then again we’re certain this will change in the future, which is just another argument for why one would consider a Core 2 Quad processor.
Intel Core 2 Quad Q6600
We would like to thank Intel, Asus and Overclockers.se for the hardware.