Sign in with
Sign up | Sign in
Overclocking Intel's Wolfdale E8000
By ,
1. The Tale Of Wolfdale: Power Requirements And Overclocking Analyzed

Only a few days ago we looked at the performance of Intel's new Core 2 dual core processors, the Core 2 Duo E8000 series, aka Wolfdale. While a 45 nm quad core Extreme edition processor (Yorkfield) has been available since early winter, the dual-core processors for the mainstream had not been released until recently. As the first Wolfdale review made clear, the new dual core processors provide a nice performance boost when compared to the 65 nm Core 2 Duo E6000 generation. This time we want to look at the overclocking potential and the power requirements of the new Wolfdale-based 45 nm Core 2 E8000 processors, as the enthusiast crowd has very high expectations since 65 nm Core 2 Duo Conroe already is an amazing overclocker Compare Prices on Core 2 Duo Processors.

Core 2 processors have served the overclocking community well since their debut. Unlike AMD, which must produce its 90 nm processors closer to their technical and thermal limits to stay competitive, Intel plays on its manufacturing prowess, which is at least 12 months ahead of that of AMD. The Santa Clara-based firm outputs most of its mainstream processors with a 65 nm process (internally called P1264) and has been producing 45 nm processors based on P1266 since the third quarter of last year. AMD continues to rely on the mature 90 nm process and is still optimizing its own 65 nm process as well as the Phenom design, which has been behind schedule.

If you look at the maximum clock speed AMD has been offering in its dual-core portfolio, which is 3.2 GHz in case of the Athlon 64 X2 6400+, and compare it to the clock speeds users reach by overclocking them (that would be only 100 - 200 MHz more), there is not much of a margin left. However, Intel Core 2 processors' in retail channels offer clock speeds up to 3.0 GHz, and many of these devices can easily be overclocked to more than 4 GHz. Even if you decided to opt for a low-cost Pentium Dual Core E2100 model (with a 1.6 GHz to 2.0 GHz default clock speed), you'd still be able to overclock the processor to at least 2.8 GHz. During the tests, the device even remained stable at 3.2 GHz, which equals a clock speed increase of over 50%. This certainly does not mean, of course, that AMD's processors, especially its lower-end mainstream 65 nm devices, are not geared for good overclocking results, either. However, AMD's overclocking margins haven't been as large compared to what Intel's modern processors offer.

We have published a number of articles on either the Core 2 topic or topics related to overclocking systems based on Core 2:

This article contains lots of basics on overclockings, components that can be overclocked and how to get started.

Intel's new dual-socket eight core Skulltrail platform can be overclocked to over 4 GHz.

This is an interesting article on overclocking very high-end devices using somewhat extreme cooling techniques.

Knowing that the 65 nm Core 2 Duo Conroe overclocks well at up to 4 GHz, and considering that Intel has released the new 45 nm E8000 processors at a maximum clock speed that is only slightly above the maximum speed of the E6000 series (3.16 GHz vs. 3.0 GHz), people are eager to see where the new core can take the power-efficient Core 2 architecture. We also wanted to know what the potential power savings are for the new generation. To look into this, we did not only measure the minimum and maximum power requirements of the test systems, but we tracked the system's power consumption over the duration of a certain workload using SYSmark 2007. These tests resemble what we did when we compared the power efficiency of current and older AMD and Intel processors:

Join our discussion on this article!

2. System Checkup

We used the Core 2 Duo E8500 engineering sample (3.16 GHz at FSB1333 system speed) for the overclocking tests. Since we wanted the results to fit into the CPU Charts to allow for easy performance comparisons across various platforms and between both AMD and Intel, we worked with the components used when we first created these charts in the summer of 2007 instead of the new Reference Test System. While the CPU Charts system is still up to date (an X38 Gigabyte X38-DQ6 and a Foxconn GeForce 8800GTX), it also offers a nice setting to facilitate overclocking. The Gigabyte BIOS includes an auto overclocking setting, which automatically adjusts voltages once you increase the Front Side Bus speed. Note that the CPU Charts test system does not support DDR3 memory, but only DDR2-800 (which is not much of an issue from a performance standpoint). You really need high DDR3 memory speeds that DDR3-1600 offers to at least get a small benefit compared to DDR2-800. You will find the overclocking test setup below.

We used the Tom's Reference System to analyze and compare the power efficiency of the new 45 nm Core 2 Duo E8000 processor against three older generations of Intel processors.

The Reference System, however, which we use as the base line for many reviews and roundups these days, is based on DDR3 RAM. We used this system based on the Asus P5E Deluxe (X38 chipset) for the power consumption and efficiency analysis for Pentium 4, Pentium D, Core 2 Duo and Core 2 Quad processors. The RAM speed we selected was DDR3-800 for the FSB800 processor generation, as the 1:1 FSB-to-memory ratio is the fastest setting we could select, and DDR3-1066 for FSB1333 CPUs.

Core 2 Duo E8000 Overclocking

The overclocking approach for mainstream Intel processors is always the same; since you cannot increase the processor multiplier, you have to increase the second factor, which is the Front Side Bus. Every high-end motherboard based on Intel's P35 or X38 chipset will allow overclocking from the default 333 MHz (FSB1333 quad pumped) to at least 450 MHz. Well designed motherboards will extend the threshold beyond 500 MHz.

3. First Step: 400 MHz FSB And 3.8 GHz

The Core 2 Duo E8500 processor reaches its 3.16 GHz core clock speed by multiplying the system speed of 333 MHz times 9.5. In the first step, we increased the bus speed to 400 MHz (FSB1600), which will be the default clock speed for the next Intel chipset generation (X48 and P45 aka Eaglelake and Bearlake). Multiplied by 9.5, this resulted in a core clock speed of 3.8 GHz, which ran absolutely stable at all times.

A clock speed of 3800 MHz was not much of a problem for the Wolfdale sample. The Gigabyte X38-DQ6 automatically increased the supply voltage from the default of 1.225 V to 1.345 V.

We set the memory to a 1:1 ratio in order to stay at the default 400 MHz clock speed of DDR2-800.
4. Second Step: 4 GHz (FSB1688)

A clock speed of 4 GHz could be reached using Gigabyte's auto overclocking mechanism, but it wouldn't provide more voltage to the motherboard components, which we found was required to achieve higher clock speeds.

We tried 4 GHz next. Since the automatic overclocking would not allow us to reach stable processor operation at 4 GHz and above due to missing voltage support, we decided to go ahead with manual settings. A 422 MHz system speed times 9.5 would give us a 4009 MHz clock speed. While 1.345 V core voltage was still enough to run 4009 MHz, we had to adjust the chipset voltage of the X38 by +0.25 V, the FSB voltage by +0.15 V and the memory voltage by +0.3 V, as the DDR2 memory clock speed increased from 400 to 422 MHz as well. We did not change the timings, hence we had to increase the voltage to maintain them.

I found it nice to see that Intel's Enhanced SpeedStep feature, which was designed to reduce the clock speed and the core voltage when the processor runs at low processing loads or idle, was fully implemented and worked properly in the test system. While the system speed and memory speed are not affected by SpeedStep, the core voltage typically is. In the overclocked scenario, however, we would not want the system to reduce the CPU voltage as well, as the 0.95 V core voltage required for a 2000 MHz core clock (333 MHz x6) at default settings would never be enough to maintain reliable operation at the overclocked 2532 MHz in SpeedStep mode (422 MHz x6). The reduction in clock speed alone provides for some power savings.

It was interesting to see that the Enhanced SpeedStep still worked when the system was highly overclocked. The multiplier drops from x9.5 to x6 when the system is idle.
5. Third Step: 4.2 GHz (FSB1772)

Next stop: 4200 MHz. Please hold on to your keyboard.

We decided to move in 200 MHz increments, which is why 4.2 GHz was the next step. As we saw stability issues and system crashes, we decided to further increase the processor voltage from 1.345 V to as much as 1.45 V. This helped to bring back stability for the overclocked system, now running at a FSB base clock speed of 443 MHz (FSB1772). Although we still did not relax the timings, the memory still worked fine at a +0.3V increase (1.5 V default).

The memory was also slightly overclocked at a 4200 MHz core clock speed, as the system speed, which runs at the same base clock speed as the memory, increased to 443 MHz.

Final Step: 4.3 GHz (FSB1812)

A 4.3 GHz clock speed was the limit. We couldn't get the sample to run faster, regardless of what we tried. Since the processors did not run hot at all during the overclocking, we assume that the system bus might be the bottleneck.

A 453 MHz Front Side Bus base clock speed (FSB1812) was the limit of the test sample. We tried increasing the voltage up to 1.5 V, and also increased the voltage of the other components up to +0.25 V for the Front Side Bus and +0.4 V for the memory. The "overvoltaging" also involved charging the chipset up to +0.35 V. We also tried different combinations of the above. Most likely the CPU sample just isn't capable of supporting much higher FSB speeds, which we have observed for Core 2 Duo E6x50 processors as well. Some of them could reach a 500 MHz base clock, while others do not.

Since this is only the beginning of the Core 2 Duo Wolfdale generation, we expect better clock speed margins to be available with upcoming processor steppings. The first 65 nm Core 2 Duo Conroe samples we had received in mid-2006 weren't able to reach 4 GHz. But today, a majority of the E6750 and E6850 processors can be overclocked to this speed with little effort.

Again, SpeedStep worked well even at 4.2+ GHz.
6. More Overclocking? Getting Rid Of The Heat Spreader

The mobile versions of Intel's Core 2 processors (this is a Core 2 Extreme X7800 for Socket 479) come without a heat spreader. Hence, the cooler is attached directly to the processor die. While this is efficient from a thermal perspective and also necessary considering the space constraints inside compact notebooks, both AMD and Intel have avoided this approach in the desktop segment, because the silicon die is rather fragile and it's easy to damage it. All desktop-class processors are covered by a metal plate, which is called the heat spreader.

As you hit the limits of an architecture, not yet hitting thermal limits (Wolfdale never got really hot during the testing), one might wonder whether or not it makes sense to remove the processor's heat spreader in order to directly attach a heat sink in an effort to improve heat dissipation. The heat spreader is the aluminum cover that sits on top of every AMD and Intel desktop processor to both protect the silicon die and to provide a flat surface to attach the CPU cooler to.

Removing the heat spreader doesn't provide much of an advantage in most cases, as the materials used to attach the heat spreader to the die conduct heat quickly enough to allow for quick heat conduction to the CPU cooler, where the heat gets dissipated by a fan or by a liquid cooling solution. However, extreme overclocking attempts require extreme measures to squeeze out a few more megahertz, which is the reason why hardcore enthusiasts worldwide have continued to remove the heat spreader once in awhile.

With the 45 nm Core 2 Duo Wolfdale, you need to be more careful than before, as the heat spreader is physically soldered to the processor die. Look at the photo below if you want to know what happens when you try to remove the heat spreader without desoldering it beforehand. If you intend to remove it, we recommend that you take a hot air gun and carefully move the heat spreader back and forth until it can be removed without damaging the processor die. We recommend not doing this, since finding a suitable cooling solution is very difficult. Remember, there is the socket 775 with its metal frame, into which the CPU cooler plate must still fit. Lastly, performance gains are marginal.

You will immediately kill your processor if you try to remove the heat spreader. Be aware of the risks if you want to try, and make sure you desolder the heat spreader using a heat gun.
7. Test Setup For Overclocking
System Hardware
AMD Platform AM2 Asus M2N32-SLI Deluxe, Rev.1.03G
Nvidia Nforce 5 NVIDIA nForce5, BIOS: 1001 (03/13/2007)
Intel Platform S775 Gigabyte X38-DQ6, Rev. 1.0
Intel X38 Intel X38, BIOS: F7 (01/02/2007)
Intel Platform S775 Gigabyte P35C-DS3R, Rev. 1.0
Intel P35 Intel P35, BIOS: F2o (05/11/2007)
Intel Platform S775 Asus P5B Deluxe/WiFi-AP, Rev. 1.03
Intel P965 Intel 965P, BIOS: 1101 (04/04/2007)
Memory 2x 1 GB A-Data DDR2-1066+ Vitesta Extreme Edition
DVD-ROM Samsung SH-D163A , SATA150
Graphics Card Foxconn Nvidia GeForce 8800 GTX
GPU: 575 MHz
Shader Clock: 1350 MHz
Memory: 786 MB DDR4 (900 MHz, 384 Bit)
Sound Card Creative Labs Sound Blaster X-Fi XtremeGamer
Power Supply Zalman, ATX 2.01, 510 Watt
System Software & Drivers
OS Windows Vista Enterprise Version 6.0 (Build 6000)
DirectX 10 DirectX 10 (Vista default)
DirectX 9 Version: April 2007
Sound Driver Vista Driver 2.13.0012 (15.03.2007)
Graphics Driver Nvidia ForceWare Version 158.18 (32 Bit) WHQL
Chipset Driver Version 8.1.1.1010 (21/11/2006)
Intel 631xESB/6321ESB/3100
Version: 8.3.0.1011
Intel 5400
Version: 8.5.0.1007
Intel 631xESB/6321ESB/3100 - SATA
Version: 8.2.0.1011
Intel 631xESB/6321ESB/3100 - USB
Version: 7.4.0.1005
Storage Driver Matrix-Storage Manager 7.0.0.1020
Nvidia Chipset nForce Driver: 15.00 (02.02.2007) WHQL
Java Java Runtime Environment 6.0 Update 1
8. Overclocking Benchmark Results

As some of our readers pointed out after we published the initial review of the Core 2 Duo E8000 series, the benchmarks are not yet optimized to take advantage of the SSE4 extensions. These will allow the processor to accelerate certain functions related to content creation or processing. However, SSE4 support depends on applications, and there hasn't been enough software available to justify spending more time on the subject at this point. We'll look into SSE4 once it will begin to make sense from a software-support standpoint.

9. Overclocking Benchmark, Continued

10. Overclocking Benchmark, Continued

11. Overclocking Benchmark, Continued

12. Overclocking Benchmark, Continued

13. Overclocking Benchmark, Continued

14. SYSmark 2007 Preview Results

Please note that we received the benchmark results for SYSmark 2007 Preview using a different test system, which has been the basis for the power-efficiency analyses of AMD Athlon 64 and Intel Pentium and Core 2 processors later on.

As you can see, the performance difference between the individual Core 2 processors is only significant if there is a lot of multi-threaded workload involved. This is the case in the video-creation section of SYSmark 2007 Preview. Apart from that, office and content creation performance doesn't vary a lot between a 65 nm dual core and a 45 nm Core 2 quad core processor.

15. Test Setup For Power Consumption Testing
Platform
CPU I Intel Core 2 Duo E8400 (45 nm; 3000 MHz, 6 MB L2 Cache)
CPU II Intel Core 2 Extreme QX9650 (45 nm; 3000 MHz, 12 MB L2 Cache)
CPU III Intel Core 2 Duo E6850 (65 nm; 3000 MHz, 4 MB L2 Cache)
CPU IV Intel Core 2 Extreme QX6850 (65 nm; 3000 MHz, 8 MB L2 Cache)
CPU V Intel Pentium D 830 (90 nm; 3000 MHz, 2 MB L2 Cache)
CPU VI Intel Pentium 4 630 (90 nm; 3000 MHz, 2 MB L2 Cache)
Motherboard I Asus P5E3 Deluxe, Rev: 1.03
Chipset: Intel X38, BIOS 0402 (2007-09-19)
RAM Crucial BL12864BA1608.8SFB
2x 1024 MB DDR3-1066 (CL 7-7-7-20 2T)
Hard Disk Drive Western Digital WD5000AAKS
500 GB, 7,200 RPM, 16 MB cache, SATA/300
DVD-ROM Samsung SH-S183
Graphics Card Gigabyte GV-RX385512H
GPU: Radeon HD 3850 (670 MHz)
RAM: 512 MB GDDR3 (830 MHz)
Sound Card Integrated
Power Supply Coolermaster RS850-EMBA
ATX 12V 2.2, 850 W
System Software & Drivers
OS Windows XP Professional 5.10.2600, Service Pack 2
DirectX Version 9.0c (4.09.0000.0904)
Platform Drivers Intel Version 8.3.1.1009
Graphics Drivers ATI Catalyst 7.11
Benchmarks and Settings
Sysmark 2007 Preview Version 1.02
Official Run

Power Consumption Test Results

All power-consumption tests involved different CPUs running at 3.0 GHz.

Idle And Maximum System Power Requirements

This difference is significant; using Core 2 Duo E8500 (3.0 GHz) instead of Core 2 Duo E6850 (3.0 GHz) results in a 5.2% decrease in the idle-system power draw (73 W instead of 77 W) and an amazing 18.2 decrease in the system-power requirement under maximum load. Keep in mind that all other system components including the power supply, the motherboard, the graphics card, the hard drive and the memory stay the same! As you could see in the overclocking benchmarks and in the first part of the Core 2 Duo Wolfdale review, the E8000 series provides even better performance while saving a lot of energy.

Average Power Requirement During SYSmark 2007

Clearly, the 45 nm Wolfdale makes a huge difference; if you use it to replace a Core 2 Duo E6850 (65 nm), it will decrease the average power requirement of the test system from 90.3 to 80.8 W for the duration of an entire SYSmark 2007 Preview run. This results in a total decrease of 10.5% for the entire system. Again, keep in mind that we only changed the processor from 65 nm to 45 nm.

16. Power (Wh) Consumed During An Entire SYSmark 2007 Run

The total power (in watt hours) required to perform an entire SYSmark 2007 run decreases by 5.2% from 106 to 100.5 Wh. While this doesn't look like much we should not forget that Penryn does not only require less total power to finish the workload, but it also delivers better performance.

SYSmark 2007 Score Per Watt hour

This is probably one of the most interesting results, as it relates to performance and power requirements. The chart shows the SYSmark 2007 Preview score divided by the total amount of power (in Wh) required to run the workload. We expected the 45 nm E8400 to be considerably better than the 65 nm E6750, but we did not quite expect it to outperform the Core 2 Extreme QC9650 quad-core processor, which is based on two Wolfdale dies inside the physical processor (Intel calls this Yorkfield). In other words, while the quad core does provide more performance for thread-optimized workloads, the power requirements increase more than the performance benefits.

17. Performance Per Watt Normalized To 3.0 GHz And Pentium 4 630

Once again we'd like to look back at one of the first Intel processors that reached 3 GHz. The Pentium 4 630 wasn't the first one ever to reach that clock speed (that was the Pentium 4 3.0 GHz on socket 478), but it was the last 90 nm Pentium 4 generation for socket 775. If we look at the SYSmark performance per Watt of the 45 nm Core 2 Duo E8400 in relation to the Pentium 4 630, the performance per Watt result has increased by over 460%!

Diagram: Performance Vs. Power Consumption

Click Image to enlarge

The diagram requires some explanation. The x axis shows the time required to complete the entire benchmark, while the y axis shows the power required at any given point of time during the benchmark run.

The brown line represents the Pentium 4 630, which required almost one hour and 40 minutes. The purple line stands for the 90 nm dual core Pentium D 830, which finished in one hour and 32 minutes. The green curve represents the 45 nm Core 2 Duo Wolfdale, which took one hour and 15 minutes. Finally, the blue and the yellow lines stand for the Core 2 Extreme QX9650 45 nm quad core and the QC6850 65 nm quad core while the red line stands for the Core 2 Duo E6750. All of them took roughly one hour and 10 minutes to finish the benchmark.

The results may be confusing, as Wolfdale (E8400) should finish this workload quicker than its predecessor Conroe (E6750). And in fact it does finish the individual workloads quicker, as you could see in the SYSmark 2007 benchmark results in the benchmark section above. But we found that SYSmark 2007 often waits for system idle tasks to complete before it initiates the next workload. You can easily track this in the green curve, as the processor is clearly running idle after four minutes, after 25 minutes, after 50 minutes and again after one hour and five minutes. We repeated the benchmark several times and haven't found out why this happened, but we can conclude based on two facts that:

  1. E8400 does perform better than E6750 in the individual SYSmark workloads;
  2. E8400 achieves much better performance per Watt despite being idle for several minutes during the duration of the SYSmark test.
18. Conclusion: Overclockers Wait! Everyone Else Go For It

Core 2 Duo Wolfdale remains an excellent new CPU that provides substantial performance while reaching new records in efficiency Compare Prices on Core 2 Duo Processors. We drove the processor all the way up to 4.3 GHz, which has been the limit for the test sample. Since this is only the beginning of the 45 nm generation, we expect future models and future steppings to deliver even better overclocking margins. We're also not sure whether it was the CPU core or the system bus that created the bottleneck that prevented higher clock speeds from being reached, as 4+ GHz can oftentimes be reached with the predecessor, Core 2 Duo E6000 Conroe. However, the 45 nm Wolfdale was optimized for performance per Watt instead of clock speed, and it clearly beats the 65 nm Conroe in all efficiency benchmarks and delivers better performance at decreased power consumption both in idle and under maximum load.

We found it amazing to see the dual-core Wolfdale beat the quad-core Yorkfield when it comes to SYSmark performance per watt. Clearly, combining two Wolfdale dies inside a physical processor, which is how Intel creates a quad-core processor, does provide much better performance (as you could also see in the overclocking charts), but the power consumption increases more than the level of performance. Even better thread optimization might help, but the results underscore the previous conclusion: Core 2 Duo E8000 is the most reasonable processor choice unless you really need much more performance. Please look at the CPU Charts for performance comparisons between the new Core 2 Duo Wolfdale processors and a long list of AMD and Intel CPUs.

Join our discussion on this article!