Sign in with
Sign up | Sign in
Clarkdale Efficiency: The Perfect Clock Rate For Intel's Core i5-661
By ,
1. What’s The Perfect Clock Speed For Intel’s New Dual-Cores?

Almost every processor can be easily overclocked on a decent motherboard. However, at some point, you'll have to boost the voltage to reach faster clock speeds, causing the processor to consume a lot more power.

There is a point at which every processor delivers its best performance per watt, though, and it happens way before you hit the chip's maximum overclock. Now it's time to determine this perfect clock speed for Intel's new Clarkdale dual-core architecture, which we know as the Intel Core i3 and Core i5.

We already executed this type of analysis on several other systems and discovered some interesting facts:

  • In the case of a Phenom II X4 quad-core (Deneb) on the Socket AM3 interface, the best performance at acceptable power consumption is reached at roughly 3.6 GHz.


Read the article Optimizing Your Phenom II Overclock For Efficiency

  • For a Core 2 Duo E8000-series (Wolfdale), based on the LGA 775 interface, the most efficient overclock speed is about 3.8 GHz.


Read the article Overclocking Core 2 Duo: Power Versus Performance

  • The Core i7-900-series quad-core (Bloomfield) running on an LGA 1366  interface delivers best performance per watt at 3.66 GHz.


Read the article Overclocking Core i7: Power Versus Performance

  • Finally, we also analyzed efficiency on the Core i5-700-series quad-core (Lynnfield) for LGA 1156 and found efficiency to be best at 3.2 GHz, with the Turbo Boost feature enabled and pushing performance up to 3.84 GHz.


Read the article Efficiency Explored: What’s The Perfect Clock Rate For Your Core i5?

While the 45nm Core 2 Duo reached top performance per watt at about 3.8 GHz, we expected higher clock speeds on the Core i5 dual-cores. Although the graphics unit is integrated with the processor, the new chips have a substantial advantage, thanks to Intel's new 32nm process.

Many online sites have reported getting Core i5 dual-cores to 4.2 or even 4.5 GHz on air cooling with comparatively little effort. You'll even find reports of hitting 6 to 7 GHz using liquid nitrogen, which is proof of the architecture having lots of overclocking headroom. Perhaps we'll even be able to reach 5 GHz on air cooling once Intel releases modified processor steppings in the next few months.

We grabbed our Core i5-661 dual-core sample (3.33 GHz nominal clock speed, 3.60 GHz with Turbo Boost operating on a single core, 900 MHz graphics clock speed) and overclocked it by raising the base clock (BCLK). At this point, it's important to watch the graphics core’s clock speed, as it also derives from the system’s base clock. Our overclocking focused on increasing the base clock in an effort to boost the processor’s nominal speed, leaving Turbo Boost and all power saving features, such as C-steps and SpeedStep, enabled.

2. Clarkdale And Turbo Boost

Intel currently has four Core i5 desktop dual-core processors at 3.2, 3.33, and 3.46 GHz. They all support the Turbo Boost feature, which means that the peak clock speed may be higher if there is sufficient thermal headroom and a high processor load. The following table provides the full overview on all processors currently available that are based on the Nehalem architecture:

Intel's Nehalem / Westmere Lineup For Q1/2010          

Model
Code Name
Clock
Max. Turbo
HT
CMOS
Cores /Threads
PowerPrice
Core i7-975 Extreme
Bloomfield3.33 GHz3.6 GHzYes45nm4/8130W$999
Core i7-950Bloomfield3.06 GHz3.33 GHzYes45nm4/8130W$562
Core i7-920Bloomfield2.66 GHz2.93 GHzYes45nm4/8130W$284
Core i7-870Lynnfield2.93 GHz3.6 GHzYes45nm4/895W$562
Core i7-860Lynnfield2.8 GHz3.46 GHzYes45nm4/895W$284
Core i5-750Lynnfield2.66 GHz3.2 GHzNo45nm4/495W$196
Core i5-670Clarkdale3.46 GHz3.73 GHzYes32/45nm2/473W$284
Core i5-661Clarkdale3.33 GHz3.6 GHzYes32/45nm2/487W$196
Core i5-660Clarkdale3.33 GHz3.6 GHzYes32/45nm2/473W$196
Core i5-650Clarkdale3.2 GHz3.46 GHzYes32/45nm2/473W$176
Core i3-540Clarkdale3.06 GHzN/AYes32/45nm2/473W$133
Core i3-530Clarkdale2.93 GHzN/AYes32/45nm2/473W$133
Pentium G6950
Clarkdale2.8 GHzN/ANo32/45nm2/273W$ 87


The 661’s graphics unit runs at 900 MHz while all other current i3/i5 models run at 733 MHz. We believe that the additional GPU speed is not a must-have feature, since few people would actually purchase a system with integrated graphics to run 3D-intensive applications or games. If you ask us, you should either choose one of the 73W models for multimedia systems or buy a discrete graphics card if you're planning on serious gaming. Although Intel has made quite a step forward here, integrated graphics still won’t impress anyone with enthusiast cred.

The main difference between Core i3 and i5 is the absence of Turbo Boost on Core i3. In addition, the Core i5 desktop processors all support Intel’s AES New Instructions (AES-NI) to accelerate encryption or decryption; Core i3 doesn’t, nor do its clock speeds reach as high. Still, due to Core i3's low price points and the fact that low-end chip overclocking margins that are typically only slightly behind the faster models, these should provide great bang for the buck.

Turbo Boost in Action

Turbo Boost can accelerate the Core i5-661's processor clock by one 133 MHz clock speed increment for both cores, or by two speed bins for a single core.

The idle clock speed of 1,200 MHz is set by the SpeedStep feature, which kicks in when there is little or no work for the processor to do.

The nominal clock speed of 3.33 GHz is either reached during normal operation with constant processor load or if thermal conditions prevent Turbo Boost from accelerating the processor any further. In the screenshot above, Turbo Boost provides an additional 133 MHz (3.46 rather than 3.33 GHz) for both cores.

If we throw a single-threaded workload at the CPU it will accelerate by two 133 MHz speed increments, to 3.6 GHz.

Since Turbo Boost is a very handy, practical feature, we believe it makes sense to overclock the processor in a way that maximizes Turbo Boost speed, but leaves the nominal speed considerably lower for the sake of power consumption. Traditional overclocking sets the CPU to a fixed clock speed, which may be fast, but is also inefficient.

3. Overclocking BCLK To 140 And 148 MHz

Wanting a real life situation to present, we decided to use the included Intel boxed cooler and marginal voltage increases. Users might want to boost system performance at little or even at zero cost, ideally without the increase in overall power consumption that significant voltage tweaks would entail.

140 MHz BCLK: 3.5 GHz Nominal Clock

Our first step was to 140 MHz, up from 133 MHz. This has little impact on the idle speed, but moves the nominal speed from 3.33 GHz to 3.5 GHz.

Turbo Boost goes up to 3.65 GHz when both cores are active...

...and to 3.8 GHz if only one core is being taxed.

148 MHz BCLK: 3.7 GHz Nominal Clock

Idle speed increases to 1,400 MHz with a 148 MHz base clock (9x multiplier). Nominal speed now is 3.70 GHz.

Turbo Boost accelerates from 3.7 GHz to 3.85 GHz when there is headroom to speed up both cores.

The peak frequency for a single core is 4.0 GHz now. Keep in mind that this all happens with only tiny voltage modifications (see the table on the following pages).

4. Overclocking BCLK To 156 And 160 MHz

156 MHz BCLK: 3.9 GHz Nominal Clock

A 156 MHz base clock means a 3.9 GHz nominal clock (25x multiplier) and 1,400 MHz idle clock speed.

If two cores are under heavy load, Turbo Boost offers a speed bump to 4.06 GHz.

With only one core in play, we’re now hitting 4.21 GHz.

160 MHz BCLK: 4.0 GHz Nominal Clock

The next stop is a 160 MHz base clock. Multiplied by the default 25x ratio, we now get a 4.0 GHz nominal speed.

With Turbo Boost applied to both cores, this results in 4.16 GHz.

And if a single core is sufficient to handle a comprehensive workload, the system will add two 160 MHz speed increments and reach 4.32 GHz. We had to increase the BIOS voltage by 0.135V to get here. Let's see how the increase affects overall efficiency.

5. Voltage Table And Test Setup

We tested using Intel’s reference LGA 1156 cooler, which comes bundled with all Core i5 dual-core boxed processors and can handle a Core i5 at up to about 4 GHz. It has a copper core and is quiet enough for an average desktop PC. But since it's half as high as the quad-core coolers designed to tackle 95W or 130W TDPs, you shouldn't expect too much from it. We decided to use it since not everyone would want to spend another $30 to $80 for a premium cooler. We recommend getting a tray CPU and purchasing a CPU cooler separately if you want less noise and/or better cooling.

System Hardware
HardwareDetails
Performance Benchmarks
Motherboard (Socket LGA1156)MSI H55M-ED55 (Rev. 1.0)
Chipset: Intel H55, BIOS: 1.1 Beta 1 (09/08/2009)
CPU IntelIntel Core i5-661 (32nm, 3.33 GHz, 2 x 256KB L2 and 4MB L3 Cache, TDP 87W, Rev. B1)
RAM DDR3 (dual)2 x 2GB DDR3-1600 (Corsair CM3X2G1600C9DHX)
GraphicsIntel HD Graphics
Hard DriveWestern Digital VelociRaptor, 300GB (WD3000HLFS), 10,000 RPM, SATA/300, 16MB Cache
Power SupplyPC Power & Cooling, Silencer 750EPS12V 750W
System Software & Drivers
Operating SystemWindows Vista Enterprise Version 6.0 x64
Service Pack 2 (Build 6000)
Drivers and Settings
Intel Chipset DriversChipset Installation Utility Ver. 9.1.1.1025
Intel Storage DriversMatrix Storage Drivers Ver. 8.8.0.1009


Benchmarks and Settings

Audio Benchmarks and Settings
BenchmarkDetails
iTunesVersion: 8.1.0.52
Audio CD ("Terminator II" SE), 53 min., Convert to AAC audio format
Lame MP3Version 3.98
Audio CD "Terminator II SE", 53 min., convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kbps)
Video Benchmarks and Settings
BenchmarkDetails
TMPEG 4.6Version: 4.6.3.268
Video: Terminator 2 SE DVD (720x576, 16:9) 5 Minutes
Audio: Dolby Digital, 48000 Hz, 6-channel, English
Advanced Acoustic Engine MP3 Encoder (160 Kbps, 44.1 KHz)
DivX 6.8.5Version: 6.8.5
== Main Menu ==
default
== Codec Menu ==
Encoding mode: Insane Quality
Enhanced multithreading
Enabled using SSE4
Quarter-pixel search
== Video Menu ==
Quantization: MPEG-2
XviD 1.2.1Version: 1.2.1
Other Options / Encoder Menu -
Display encoding status = off
Mainconcept Reference 1.6.1Version: 1.6.1
MPEG-2 to MPEG-2 (H.264)
MainConcept H.264/AVC Codec
28 sec. HDTV 1920x1080 (MPEG-2)
Audio:
MPEG-2 (44.1 kHz, 2-channel, 16-bit, 224 Kbps)
Codec: H.264
Mode: PAL (25 FPS)
Profile: Settings for eight threads
Application Benchmarks and Settings
BenchmarkDetails
Grisoft AVG Anti-Virus 8Version: 8.5.287
Virus base: 270.12.16/2094
Benchmark
Scan: some compressed ZIP and RAR archives
WinRAR 3.9Version 3.90 x64 BETA 1
Compression = Best
Benchmark: THG-Workload
WinZip 12Version 12.0 (8252)
WinZIP Commandline Version 3
Compression = Best
Dictionary = 4096KB
Benchmark: THG-Workload
Autodesk 3d Studio Max 2009Version: 9 x64
Rendering Dragon Image
Resolution: 1920 x 1280 (frame 1-5)
Adobe Photoshop CS 4 (64-Bit)Version: 11
Filtering a 16MB TIF (15000x7266)
Filters:
Radial Blur (Amount: 10; Method: zoom; Quality: good)
Shape Blur (Radius: 46 px; custom shape: Trademark sysmbol)
Median (Radius: 1px)
Polar Coordinates (Rectangular to Polar)
Adobe Acrobat 9 ProfessionalVersion: 9.0.0 (Extended)
== Printing Preferenced Menu ==
Default Settings: Standard
== Adobe PDF Security - Edit Menu ==
Encrypt all documents (128-bit RC4)
Open Password: 123
Permissions Password: 321
Microsoft Powerpoint 2007Version: 2007 SP2
PPT to PDF
Powerpoint Document (115 Pages)
Adobe PDF-Printer
Deep Fritz 11Version: 11
Fritz Chess Benchmark Version 4.2
6. Benchmark Results: Application Benchmarks

AVG’s virus check doesn’t benefit much from the faster clock speeds. This doesn’t necessarily apply to all anti-virus programs, though. We're looking into other AV suites, since the latest versions of this one have just fallen off with regard to scalable results.

3ds Max scales in a linear fashion, but this application can benefit much more if you go for a quad-core processor rather than trying to coax additional performance from a dual-core chip. You’ll almost halve the rendering time at similar clock speeds. Check out our current Desktop CPU Charts on 3ds Max for details.

PDF creation happens noticeably faster for our 115-page PowerPoint document. While the process takes 1:35 at stock speeds, the same PDF finishes in 1:17 at a 4 GHz nominal speed (4.16/4.32 GHz Turbo Boost).

There are noticeable benefits for image processing using Photoshop, as well.

WinZip 12 isn’t optimized to take advantage of additional cores, but it does benefit from any clock speed increases. This is another one we plan to replace, as there isn't much room left for compression/decompression titles that aren't written with parallelism in mind.

Fritz 11 is a chess program that takes advantage of multiple cores and clock speed increases.

7. Benchmark Results: Audio/Video

The Lame MP3 encoder still is popularly used to convert CD audio (.wav) into MP3, especially with sophisticated settings. The tool delivers much quicker processing times given quicker clock speeds.

DivX is a popular video codec for .avi container files. Transcoding an MPEG-2 SD video file into DivX 6 format is faster with the overclocked settings, but the difference is not as significant as it'd be with a quad-core processor.

Results are similar on XviD. It benefits from higher clock speeds, but don’t expect serious performance boosts.

And a last set of similar results, this time on MainConcept’s video encoder. It takes HD MPEG-2 video and creates an H.264 file.

8. Benchmark Results: Power Consumption

Idle power depends on the processor’s idle clock speed and applied voltage. Once we started to increase the processor voltage, we saw an increase in system idle power as well. The idle clock speed went from 1,200 MHz for the 3.33 GHz/133 MHz BCLK default setting to 1,440 MHz at 4.0 GHz/160 MHz BCLK. However, 40W instead of 35W sounds acceptable in the light of the performance gains.

Peak power also increases from 86W at maximum clock speed (3.46 GHz) to 114W when running 4.16 GHz. That’s quite a bit, but keep in mind this only applies under heavy load, and that the CPU goes back to idle state once open tasks are completed.

9. Benchmark Results: Power Efficiency

Let’s start with the runtime of our efficiency benchmark, which consists of the following applications: 3ds Max, DivX, Xvid, Lame, MainConcept, PDF Creation using Adobe Acrobat 9 and Microsoft PowerPoint 2007, Photoshop CS4, AVG Anti-Virus, WinRAR, and WinZip 12. We saw a runtime decrease from almost 30 minutes on stock clock speeds to less than 26 minutes at the highest overclock.

These are the average power requirements for the efficiency test run and each of the test clock speeds.

We also tracked the total power used during this benchmark. It's interesting to see that the first three settings require exactly the same total power, while performance keeps going up from one test to the next. Now we can look at the efficiency score that divides the performance result by the power used.

You could almost tell by the total power used what performance you’d get this is the ranking. Overclocking to 3.7 GHz with 3.85/4.0 GHz Turbo Boost clock speeds is the most reasonable overclock, delivering the best performance per watt.

While other processor architectures show increasing power efficiency with a bit of overclocking, the Core i5 dual-core does not. However, going after too much clock speed causes a decrease in efficiency (while still delivering more performance, of course).

10. Conclusion

We don’t want to spend too much time talking about the Clarkdale-based Core i5/i3 dual-core processors themselves. We’ve done that in great detail, and it’s obvious that these CPUs deliver great performance and impressive efficiency, albeit at a significant price premium compared to AMD’s offerings. Instead, this article looks at the performance gains and the impact on power efficiency when overclocking a Core i5-661.

Our testing revealed that the processor shows fairly steady efficiency up to about 3.9 GHz. This means that every voltage and clock speed increase causes an increase in power consumption that is equivalent to the performance impact. However, if you exceed 4 GHz nominal clock speed, you will have to turn up the processor voltage (Vcore), and this clearly decreases performance per watt. Our recommendation is to stay in the range of 3.7 to 3.9 GHz nominal speed. Intel’s Turbo Boost technology will still be there to add another one or two clock speed increments.

Two conclusions can be drawn: first of all, the difference in power consumption between the stock settings and our highest overclock—4.0 GHz nominal and 4.32 GHz with Turbo Boost max'ed out—is relatively small. Our measurements say that the delta is around 5W at idle and roughly 30W at peak load. Even though efficiency decreases above 4 GHz, it might still make sense to go higher. The chips have decent clock speed margins, and the additional power draw isn’t a huge deal. All you’ll need is a better cooling solution. Last but not least, there's room for improvement on the power supply side, as the 750W PC Power and Cooling device we’ve been using for various reviews is by no means an ideal partner for this test system. Go for a more conservative PSU to match power requirements so you can operate within power supply’s highest efficiency range, and thus save some power.

Our second conclusion refers to Turbo Boost: we want more of it! Obviously, there's headroom. Intel could take better advantage of the existing architecture, so why not work on more sophisticated power schemes? Turbo Boost and SpeedStep are conceptually similar, but they originate from different requirements. The PCU, Intel’s on-die power control unit, is already there to take care of dynamic clock speed adjustments. We would love to see processors differentiated mainly by their thermal specifications, meaning that a 65W desktop CPU would feature large power saving potential, but limited (reasonable) “overclocks” to stay within the power envelope. At the same time, a 130W high-end CPU could offer smaller power saving options but much higher clock speed boosts at peak loads, given the available thermal headroom, of course. Finally, the Extreme Edition flagship could allow users to modify thermal thresholds at their own risk. This is only a vision, but it would make more sense than keeping the processor world revolving around clock speeds.