Core 2 Duo and Core 2 Extreme Evaluation

Product: Intel Core Architecture (Core 2 Duo/ Core 2 Extreme)
Author: Sean Kalinich
Category: All
Manufacturer: Intel
Spelling and Grammatical editor: Carlos Echenique

Introduction:
For what seems like an eternity AMD and Intel have squared off on opposite sides of the CPU arena. Back and forth these two companies have struggled. In the past couple of years AMD has seemed to take the offensive and to have gained a performance advantage that Intel has not been able to touch. Even the vaunted Extreme Edition CPUs have been taken behind the woodshed and pounded by AMD’s slower clocked CPUs.</p”>

Now it seems the Intel engineers have been working overtime and done a complete rework of Intel’s flagship CPU. Codenamed Conroe, this CPU will sell under the name Core 2 Duo and Core 2 Duo Extreme, these will be the desktop performance king of Intel’s line up.

In fact for the last few months, there have been Conroe reviews popping up like weeds on the Internet. These reviews that many have been flaunting have been with pre-release, non-(Intel)supported engineering samples that were “leaked” from Intel (the rumors say from Intel TW). You can read review after review that parade amazing numbers in front of your eyes and generate just as amazing page hit numbers for the sites that published them. The sad fact of most of these is that the CPUs they have been using are not what the consumer will be getting when they shell out their hard earned money for one. Although I have to marvel at the person that thought up this marketing strategy. It is a brilliant one, flood the Internet sites with samples so that the consumer will see amazing numbers from Conroe. This undermines the confidence in AMD as the AM2 is not showing that kind of performance. It also generates a demand for Conroe before it even ships. If this was not thought up by someone at Intel, I bet they are wishing they had.

Today we get to show you our experiences with two species of Conroe (These are Official Samples from Intel); the Core 2 Duo E6700 which runs at 2.66GHz and the Core 2 Duo Extreme which runs at a very nice 2.96GHz. We will see where it sits in the food chain.
Is it predator, or is it prey.

New and Improved just like the original.

Architecture:
Now my introduction above may seem to be a little harsh yet I cannot help but thinking that I have seen all of this before. I have visions of when the X1800XT came out from ATi. Many sites hailed this as a 7800GTX killer, able to soundly thrash nVidia’s top of the line card without so much as breaking a sweat. Yet when the real world testing was finished the two cards were so evenly matched it was a big letdown after all of the hype.

As mentioned above we have seen review after review touting the Core 2 Duo as the AM2 killer, etc, etc, etc, and blah, blah, blah. Are all of these reviews faked? I highly doubt that. Is this a case of either benchmarks optimized for CPU or CPU optimized towards benchmarks? Again I do not think that is entirely the case.

So what is it about Core 2 Duo that makes these amazing numbers possible? What have the mad scientists at Intel changed to give Intel the horsepower to pull off this kind of an upset?

One of the first things noticeable about the Core 2 Duo line is the L2 Cache. This has been bumped up to 4MB in all of the high end models (beginning with the E6600) . Normally this would indicate 2MB per core, with the Core 2 Duo this is not the case. The 4MB cache is shared by both cores with the ability to dynamically adjust cache usage based on work load and CPU core demand. Now this sounds like a wonderful idea on paper but how well would it work in practice? Would the CPU be able to quickly adjust to the demand and will there be enough bandwidth available to keep up with the demand from both cores. This added cache has been dubbed by some as “Memory Avoidance Technology” Intel calls it Advanced Smart Cache. Its purpose is to help alleviate the inherent memory access bottleneck between CPU, Northbridge, and memory. The less information that has to take that long trip, the faster the CPU can execute code. This is something that we were very interested in testing. As you will see later in our performance section were we quite surprised at our findings.

Some other architecture changes are:

Wide Dynamic Exectution: This allows for four full instructions to be executed simlutaneously. (As opposed to the three instructions the Mobile and Netburst architectures can handle)

Macrofusion: Macrofusion allows for similar instruction pairs to be combined into a single internal instruction (called a micro-op)

Improved Arithmetic Logic Unit: This assists in the macrofusion processing allowing for greater performance without an added cost in energy.

Memory Disambiguation: An advanced out-of -order instruction that allows the poccesor to intelligently load instructions from memory. While this sounds simple there is more to it. Normaly, a microprocessor cannot unload infomation stores to reschudle new loads due to possible memory dependancies. Memory Disambiguation allows the microprocessor to determine if there are any dependanices. If it finds that there are none, it can reschedule instructions to allow for faster and more efficient execution.

Advanced Digital Media Boost: Allows for execution of SSE instructions are a rate of one complete instruction per clock cycle, this is an imporvement over the previous one instruction per two clock cycles.

Power:
Intel and AMD have both made significant progress on making their CPUs more energy efficient. With the rising costs of Fuel and Power it is a welcome change. I have mentioned this before but my power bill is quite hefty each month from the various servers and workstations I have running at my house (not to mention the test rigs running). I am very pleased that both companies are now paying attention to this facet of computing. With Core 2 Duo Intel has definitely made improvements over previous generations of CPUs. The new Core 2 Duo CPUs are the energy efficient line of Conroe. While the Core 2 Extreme are for the high performance end.

The E6700 we tested is rated at 65 Watts,while the X6800 has a power rating of 75 Watts.

Heat:
Heat has been an issue with previous generations of Intel CPUs, with many likening them to small space heaters. The heat output from the new Core 2 CPUs has been cut back by a reduction of the power needed to run them, added in is also advanced speedstep to throttle back the CPU when the CPU is not under load. It has been brought to my attention on the eve of publication that some of the DX975XBX boards are miss-reporting CPU temperatures. Unfortunately this is another subject that will require a follow up evaluation.

Hardware Virtualization:
Here is something that is a touchy topic; for a while now Intel has offered a type of hardware virtualization and it is an improvement over the previous software only solutions. The rub (in this case) is that Intel CPUs have no integrated memory controller. This fact all by itself precludes the possibility of memory virtualization. So you have only a partial solution here. One of the most important parts of a virtual OS is the ability to access memory quickly and efficiently. Without the ability to do this, the Core 2 Duo is under a handicap when it comes to virtualization. Now you could argue that the current crop of Virtual OS software does not support true hardware virtualization either. We are exploring virtualization performance in depth and plan on releasing a full evaluation of both the X6800 and FX-62’s performance at a later date.  <p”>[Editor’s note: Hypervisor based virtualization is available today in the form of the Open Source Xen Project (which currently only supports Linux VM’s) and VMWare’s ESX Server (a closed source commercial product). Microsoft demonstrated Windows Hypervisor at WinHEC 2006, but it was unclear as to whether this is a true hypervisor or not.] 

Methodology:
For this evaluation I wanted to test the CPU with as many real world tests as I could. There will be very few synthetic benchmarks and no time demos in this evaluation. They do not serve to show off real performance.

In addition to the software used for testing, I also wanted to simulate a real world environment. For this I installed the following commonly used applications:
Microsoft Office 2003
SpeedFan
Diskeeper 10
Avast Home Edition
Acrobat Reader
FireFox

A HighSpeed PC TechStation was used for all testing.

Intel
Core 2 Extreme X6800 (2.93GHz)/ Core 2 Duo E6700 (2.66GHz)
Intel Desktop Mainboard DX975XBX
Kingston HyperX DDR2-800 CL4 (SLI Ready)
Leadtek PX7900GTX TDH Extreme 512MB
2x Western Digital 10,000RPM Raptor SATA HDDs (RAID0)
ThermalTake Pure Power 680Watt PSU
SilverStone  NT06 CPU cooler
Generic 16x DVD-ROM
Generic 1.44MB Floppy
Microsoft Windows XP Professional/ Microsoft Windows XP x64 Edition

AMD
AM2 FX-62 (2.8GHz)/ X2 5000+ (2.6GHz)
MSI K9A Platinum
Corsair XMS2 Twin2X2048 6400C4 DDR2-800 CL4 (SLI Ready)
Leadtek PX7900GTX TDH Extreme 512MB
2x Western Digital 10,000RPM Raptor SATA HDDs (RAID0)
SilverStone Zeus ST75ZF 750 Watt PSU
Stock AM2 Cooler
Generic 16x DVD-ROM
Generic 1.44MB Floppy
Microsoft Windows XP Professional/ Microsoft Windows XP x64 Edition

For my AM2 system I went with an ATi chipset to remove the nVidia optimizations (SLI Ready RAM Link Boost etc). Since I do not have an nVidia based mainboard to test the Intel CPUs on I wanted to get as even a playing field as I could.

Gaming:

The four game titles chosen have heavy physics and particle effects throughout the game. They are, in many cases, capable of bringing even the fastest system to a crawl.

All games were run at 1024×768 with no additional graphical options on. The only exception to this was HalfLife 2 Episode One where HDR was set to Full. The HDR used in HalfLife is a more CPU intensive integer based HDR. These settings, for the most part, removed the GPU from the equation and should show how the CPU effects the basics of gameplay. We will be testing the Core 2 Line in real-world game play in a follow-up evaluation. All testing was done during real gameplay (with reduced resolutions). Frame rates were recorded using FRAPS. 2.7.4

Need For Speed Most Wanted:
Need for Speed Most Wanted is a game that takes a lot of CPU horse power.There are many AI and Physics calculations going on in the game. Although the AI routines are simplistic, there are a large amount of them going on. Between the Police chasing you and the random vehicle movements, it can become a drag on system resources. The Physics routines apply to all objects in the game making it a very immersive driving game. On a different track, it will be interesting to see if future versions of Need for Speed will take advantage of Physics Processing Units like the Ageia PhysX or the GPU drive solutions from ATi and nVidia. For my test runs I used the challenge portion and ran race number nine. This race has a few nice curves to throw your speeding Corvette around at high speed. Even at 1024×768 the average frame rates never broke 100 fps.

In Need for Speed the Core 2 line enjoys a rough 10% lead across the board. With the E6700 outperforming even the FX-62.

F.E.A.R.
F.E.A.R. is another CPU intensive game, not only for physics but for the AI engine that powers the clone soldiers. The AI in F.E.A.R . is one of the best I have ever seen, allowing for very realistic combat. Add in the great “spooky” factor that Monolith is famous for and you have an immersive game that hammers away on your system.My test run was Interval 02 Insertion. I ran this level from the entry at the warehouse gate until the little girl (you know, the little girl that is on fire) throws you out of the window.

Here we see Intel’s new monster running away with an average 40% performance lead. Very Impressive indeed as the E6700 again out paces the FX-62 while clocked roughly 200MHz slower.

Call of Duty 2:
Unlike the other three games we have chosen to test the Core 2 Duo and Core 2 Extreme with Call of Duty 2 does not have a realistic AI system. What it does have is particle effects and physics everywhere. On the AI front the enemies use more of a mob tactic than a concerted effort to kill you off. For my test run here I ran the winter campaign from the end of the training (it is rather annoying that you cannot bypass that) until the destruction of the Panzerwerfer. This run involved many enemies smoke explosions and the other fun things about “mob style” war shooters.

Score another round for Intel as the Core 2 pair leads by an average of 30%.
HalfLife 2 Episode One:
The first new release from Valve and Steam in the HalfLife 2 series is Episode One (“Luke, I am your Father’s Brother’s Cousin’s Uncle!”) This “Mini” Game drags Gordon Freeman back to the ravaged earth, not by the G-Man but by the Vortigaunts, to fix what he has broken. Tagging along is Alyx, as you move through City 17 to escape the inevitable destruction of the reactor core in the Citadel. As with the other releases, HalfLife 2 Episode One is an incredibly immersive game. The textures, lighting, sounds, and colors all blend with the storyline to create a feeling throughout the game of oppression. The Physics and HDR in HalfLife 2 are also some of the best I have seen in a first person shooter. For my testing I chose a section of the LowLife level, where Gordon and Alyx move though the ruins of an underground parking garage full of Antlions. The numerous enemies and the need to move several destroyed cars to cover up the Antlions access holes allow for the physics engine to come into play. I continued with this level until I reached the elevator and was able to board it (after the fight in the dark with the various zombies.).

As if you could not figure out the results here. Intel again strolls effortlessly by with a commanding 33% performance lead.

Cinebench 9.5:
We Started using Cinebench 9.5 in our AM2 evaluation and have brought it back to see how Conroe fares. This test simulates both single and multi CPU rendering and works natively in both XP Professional and XP x64 Edition. This gives us a very good indication of the advantages we can expect from the Core microarchitecture in a 64 bit environment.
LightWave 3D:
Score another one for Intel as the FX-62 again has trouble keeping up with even the E6700

The second in our rendering applications Newtek’s Lightwave 3D has been the choice of many professionals for years now. Starting with version 8.5 Newtek included 64 bit support. In our article on the benefits of 64 bit over 32 bit found here we saw how a multiple core SMP system was able to take advantage of the 64 bit memory addressing and run rings around a similar 32 bit system. With the Core 2 CPUs we have multiple cores and 4MB of cache to buffer rendering information add into all of this, the EM64T instructions. My render sample is available from the content CD. It is the moonbase scene. I set the camera resolution to 3200×1600, for my AA I used a PLD – 5 Pass run with Gausian (sharp) for my reconstruction filter. Thread was set to 2 with segment memory at 128MB. Here we can see the new Core 2 run away from the AM2 with sheer horse power and an over abundance of L2 cache.

Time in Hours, Minutes, and Seconds (lower is better)

Here there is no contest. Intel out-renders the AM2 CPUs by a minimum of 30 minutes. Although it is interesting to note that the AM2 had a very slight performance increase from x86 to x64. While Intel had roughly the same performance loss between the two.

PhotoShop CS 2:
Adobe’s Photoshop CS is, as many would claim, the standard for image processing and manipulation. Photoshop uses a vector rendering system for images. Each new edit can be placed in its own individually editable layer. This rendering system allows for easier image manipulation. However this power and functionality comes at a price, Photoshop CS 2.0 needs a lot of horse power, not only in terms of CPU, but memory, HDD performance, and GPU. For our Photoshop testing we used Driverheaven’s Photoshop bench with Photoshop CS 2 (all updates as of this writing applied).

Although the AM2 CPUs are able to perform better in the “Dust and Scratches” filter they lag behind in every other filter. In some the Core 2 Duo is almost twice as fast.

SuperPi:
Hmmm. Nothing better than figuring out the number Pi to 32 million places on a long afternoon. Well not really, but the application that allows for this mind numbing exercise can be good to test your systems performance. SuperPi runs it’s calculations back and forth between the memory, CPU, and hard drives. A weak spot in this chain makes for a slow time. One thing I have been using SuperPi for lately is to check the ability of multi-core CPUs to run at 100% on each core. To do this I run 2 instances of SuperPi in parallel, I set affinity at one instance per core. Then I run them in parallel with no affinity. With the Core 2 CPUs I wanted to see how the 4MB of L2 cache would affect performance and if I would be able to overload this cache and see a significant slowdown on the times for a 32M run and a 1M run. Interesting results indeed.

Time in Minutes and Seconds (lower is better)

SuperPi 1M Single Run

Time in Seconds (lower is better)

32M Dual run with core affinity set

Time in Minutes and Seconds (lower is better)

32M Run No affinity

Time in Minutes and Seconds (lower is better)

Dual 1M with core affinity set

Time in Seconds (lower is better)

Dual 1M Run no affinity

Time in Seconds (lower is better)

The interesting thing about these results is in the dual runs. While the Intel CPUs still have enough raw horse power to bludgeon their way to a good time, on average they take a bigger performance hit than the AM2 does. While this is interesting and brings up questions about multi tasking performance it still does not take away from the fact that Intel soundly trounces the AM2 line. We will be looking at multi-tasking performance in-depth in a later review.
Easy DVD to DivX SVCD converter:
Video encoding seems to be a popular usage for peoples home computer systems these days. Many, myself included, set up Media Center or HTPCs to run their movies from instead of the hassle with DVDs or VCR tapes. The porting of movies from DVD or VHS to a computer has raised many legal and moral issues, thankfully I am not going to go into either of those here. Instead I am just going to tell you what type of encoding performance you can expect from the X6800 or the E6700 if you run a system similar to my test platform. For my encode I used Easy DVD to DivX VCD SVCD Converter Pro. (Version 3.0.36) with DivX codec 6.2.5 and CD quality audio. The Image size was set to infinite and saved to the local hard drive. As you can see below the Core 2 times are quite impressive.

Numbers Game:
This section is for those of you that just HAVE to know how well it does in PCMark05 and 3DMark06…. Oh and wait what does it do in Sandra? Well here are all of the numbers to make you happy.

The Sisoft Sandra 2006 Scores were only run on the X6800 and the FX-62 but are indicative of both lines.

So now you know that even in static benches Intel is able to snatch the crown from AMD’s head and run off with it.

Price:

Pricing for the Core 2 lineup will be as follows (in 1000 unit lots):

Considering the fact that an E6700 runs just about as fast as an FX-62 the prices are not just good they are very good.

Conclusion:
My original intention with this evaluation was to cover rendering, virtualization, and multi-tasking in depth. However the official release date was moved up by two weeks cutting my testing time significantly. I will be following up this hands-on evaluation with in-depth articles and evaluations on those topics, that (due to time constraints) had to be left out. The evaluation as it stands is still enough to show that Intel has finally pulled their collective heads out of the sand and gotten back into the performance game… and with a vengeance. They were not satisfied with meeting the AM2’s performance, they surpassed it significantly. Even with the latencies inherent in having the memory controller in the northbridge, Intel’s new architecture is able to pretty much pound AMD into submission. There will still be some areas that AMD will out perform Conroe. But they will be few and far between for now. Should you dump AMD and run to Intel? Well, again, that is something for you to decide. I am certain that AMD is not sitting back and shaking their heads. They are at work on something. Of course time will tell if that product will put them back in the game or not.

In all, as of this writing Intel’s new Core 2 CPUs are the predator in the CPU jungle. Running down and devouring the current competition.

So now you know the about the raw horse power of the new Core 2 line, how will it fare in real world usage and at real world resolutions? Check back for our follow on reviews where we will run through those topics and to tell you how Core 2 performs in the real world.

Acknowledgements:
I would like to thank Dan Snyder at Intel for providing the Core 2 Duo, Core 2 Extreme and DX975XBX mainboard used in this evaluation.