Guide: Overclocking AMD And Intel CPUs On A Budget

[Gin Fushicho]

Why overclock?

Overclocking is a collection of methods for making components run faster than the manufacturer intended. Once little more than a hobby for die-hard geeks and value-seekers, overclocking has become a way—sometimes the only way—for performance fanatics to get the system performance they really want/need. With graphics and memory technologies forging ahead at a brisk pace, central processors are quickly becoming the second-most restrictive component in many high-end systems.

If you feel forced into overclocking just to get a high-performance benchmark from the best parts, mid-budget enthusiasts are certain to find their lower-cost parts mind-numbingly slow. Because most buyers can’t afford the best components, the majority of overclockers come from the mainstream market.

There are two groups who overclock out of perceived necessity: those who need more performance than the market provides, and those who need more performance than they can afford to purchase.

Tom’s Hardware puts much of its editorial efforts into testing and overclocking the latest high-end parts, but today we’re going to focus on a few processors that most mainstream readers can afford and enjoy: AMD’s Phenom II X2 and X4, and Intel’s Pentium Dual-Core and an entry-level Core 2 Quad.



Mitigating Risks

Though we’re obligated to tell everyone that overclocking is a great way to put important data at risk, many Tom’s Hardware editors even employ it on their all-important work PCs. Methods that ensure stability are just as important as those that assure longevity, and any data that can't be replaced should be backed up to at least two devices, regardless of whether or not the primary system is overclocked.

All machines wear out, and forcing a component to run beyond its specifications is a sure way to make it wear out faster. In electronics, the biggest source of wear is a phenomenon known as electromigration, whereby ions are slowly transferred from a structure to the adjacent structure under the force of electrical current. Major contributing factors include increased heat and voltage, but the limits of heat and voltage vary with different materials, different production technologies, and expected component lifespan.

Increased voltage allows a stronger signal to be carried between various components, reducing signal loss that can occur as the result of overclocking and thereby allowing higher component operational frequency. As we overclock today’s four processor samples, we’ll discuss the voltage and temperature limits we’ve chosen as well as the expected lifespan, testing each part for complete stability.  

[Gin Fushicho]

Understanding The Lingo



Many new overclockers leave our user forums never to return when they ask "How do I overclock?" and receive “Raise the FSB or HT clock” as an answer. But once you're armed with the lingo, the principles are fairly easy to understand. Let's cover a couple of the basics.

Frequency

A processor is made up of a complicated series of microscopic electronic switches (transistors) on a pulsating power circuit. The number of pulses (power cycles) per second is called the circuit's “frequency.” It takes at least one cycle for the transistor to change state between on (1) or off (0), and the ones and zeros become part of a data stream.

Modern central processors run at thousands of millions (billions) of cycles per second, or gigahertz. This is the same range of frequencies at which microwaves and mobile phones operate, so that a relatively short piece of wire can become a fairly good radio antenna. Preventing cross-communication between circuits, where one circuit acts as a transmitter an the other an unintended receiver, is extremely important.

The conductors on motherboards, called traces, are much longer than those of an integrated circuit, such as a central processor (CPU) or graphics processor (GPU). In order to reduce noise, signal loss and cross-talk, the pathways that connect various processors must run at slower frequencies.







The CPU Multiplier

As the need for increased data speed outstripped the ability of various busses to support it, companies developed a variety of methods to send more than one bit of data per conductor, per cycle. These methods include double data rate used in memory modules, quad data rate used by Intel’s front side bus (FSB), AMD’s HyperTransport (HT) interconnect, and Intel’s recent QuickPath Interconnect (QPI).

Because Intel’s most recent FSB uses quad data rate technology, its clock frequency is a quarter of its data frequency. That is to say, the clock rate of FSB-1333 is 333 MHz (megahertz, or millions of cycles per second). The CPU itself relies on an actual electrical frequency (the clock rate) to set its internal speed, so a CPU multiplier of 10x on an FSB clock rate of 333 MHz (FSB-1333) results in a CPU frequency of 3,333 MHz, or 3.33 GHz.

AMD’s internal HT link uses a 200 MHz clock speed with data rates of five to ten times clock speed, resulting in 1,000 to 2,000 transfers per second. But since HyperTransport supports full bandwidth in both directions at the same time, AMD doubles its name to HT 2,000 (1,000 MHz data rate, 200 MHz clock rate) and HT 4,000 (2,000 MHz data rate, 200 MHz clock rate). The most important thing to remember when overclocking is that both HT 4,000 and HT 2,000 use a clock rate of 200 MHz, so that a CPU multiplier of 10x would provide a CPU clock speed of 2,000 MHz, or 2.0 GHz.

Though we won’t use an Intel QPI-based system today, users should know that it operates in a similar fashion to AMD’s HT link, but at a slower 133 MHz base clock frequency.







Voltage

Frequent overclockers will discuss BIOS settings such as VCore (voltage of the CPU core), VDIMM (memory voltage), and various data pathway/memory controller voltage settings under a variety of different initializations. Some of these will be discussed in detail as we encounter them in BIOS screen shots.  

[Gin Fushicho]

Getting Started, The Hardware




Continually-falling DDR3 prices are allowing the memory technology to displace DDR2 in mainstream-performance builds. And with the future of DDR2 desktop memory drawing short, we selected two DDR3 motherboards from MSI to support our chosen AMD and Intel processors.



Picked for its best-in-class HT clock speed capability, MSI’s 790FX-GD70 should provide optimal results for our Socket AM3 overclocking tests. Choosing the standout motherboard from previous reviews allows us to set a high goal for owners of less-expensive motherboards to attempt using the same processor models.



Our budget limit for dual-core and quad-core processors was $125 and $250, respectively. AMD sent its Phenom II X4 955 Black Edition ($245 retail value) and Phenom II X2 550 Black Edition ($100 retail value) for today’s overclocking guide. Black Edition processors are special from other Athlons and Phenoms in their ability to manipulate the clock multiplier upward, allowing high overclocks to be achieved at or near the processor’s original 200 MHz HyperTransport reference clock.

We requested MSI’s top P45-chipset motherboard to maintain fairness between processor brands, and the firm responded with its P45 Diamond.






A higher average price gets buyers fewer graphics card slots. MSI makes up for the value loss with added features, such as a PCIe audio card and a chipset water block with copper line adapter kit.

Nobody said we had to spend our entire budget on processors. Focusing on the value segment brought us to the $70 dual-core Pentium E5200 for its high CPU to front side bus multiplier and good overclocking reputation, and the $160 Core 2 Quad Q8200 for its reasonable cost.







Intel doesn’t produce a 45 nm desktop quad-core with anything less than FSB-1333, and each model up gets us a slightly higher (0.5x) CPU multiplier at a noticeably higher price. Like AMD’s Black Edition, Intel also offers Extreme Edition processors with CPU multipliers that can be manipulated upwards, but Intel charges so much more for this feature that we couldn’t possibly consider any of these for use in a value-oriented overclocking guide.  

[Gin Fushicho]

Keeping It Cool




Cool processors clock higher and survive longer, but finding an inexpensive cooler in the preferred 120mm tower design able to support both AMD and Intel processors isn’t easy. Rosewill surprised us with a review sample that included an AMD-style clip, since its Fort 120 doesn’t advertise Socket AM2+/AM3 compatibility on the box. Readers should look forward to a review of this unit later this month.




This is the point where some die-hard overclockers might point out that, since we used top-end motherboards, we should also use a top-end liquid cooling system. But while budget overclockers might be able to find less expensive motherboards that replicates our results, the same cannot be said of liquid cooling. We wanted to provide a realistic, yet optimistic target for value-overclockers to use as a goal.

One other place we didn’t go cheap was in thermal compound selection. The Fort 120 cooler does not include enough thermal paste for multiple uses, so we instead relied on our established thermal grease choice.






Zalman’s ZM-STG1 was chosen for previous reviews based on its easy application, quick set in time, and upper-range thermal performance. Upon request, the firm supplied enough samples for each U.S. editor to have two bottles.



Thermal grease or paste fills small gaps between the processor and heat sink to provide a greater contact area. Many experienced builders swear that too heavy a layer will prevent proper sink contact, citing the lower conductivity of thermal compound compared to the aluminum or copper surface it fills, but most modern thermal materials are thin enough that heat sink pressure will squeeze out any excess. The real problem of applying too much paste is that it can make a mess of the motherboard, and its low-conductivity is still enough to potentially cause signal or voltage problems.  

[Gin Fushicho]

More Shared Hardware




Though processor families must be used with specific types of motherboards, other parts, such as the power supply, RAM, and hard drive work across multiple platforms.



Cooler Master’s RS850-EMBA power supply has far more capacity than needed for today’s guide, but was chosen because it was already on the bench. Its 80 PLUS rating should allow realistic comparisons of power draw between stock and overclocked speeds of each processor.



We didn’t need three modules to test this guide’s dual-channel systems, but two of the same parts can be used in dual-channel mode. Kingston’s DDR3-2000 wasn’t just handy; it’s also capable of low latencies at various speeds, available in single-module packages for dual-channel kits, energy efficient, and able to extract peak performance from each processor. Builders should look forward to a cost-conscious comparison of modern dual-channel kits later this month.






Western Digital’s VelociRaptor was again chosen for convenience, since its higher-than-average data rate allows quicker load times, but with little to no affect on most benchmark scores. It certainly won’t affect the outcome of today’s overclocks.  

[Gin Fushicho]

Overclocking AMD's Phenom II X2 550




We follow the overclocking scene fairly closely and have found several overclockers using AMD’s 45 nm processors at voltage levels between 1.50 and 1.56 volts since the Deneb quad-core was first introduced last winter. This level of voltage tolerance is far greater than that of competing Intel models, but to play it safe we chose a maximum setting of 1.50 volts (give or take a few millivolts) under full CPU load, mindful to keep peak unloaded voltage below 1.55 volts.

AMD publishes overclocking software under its “AMD OverDrive Utility” name that allows many of the most important settings to be changed inside Windows. While these can prove useful for finding the processor’s operational limits, many users will eventually want to make these adjustments semi-permanent through BIOS settings.

The traditional overclocking method is to increase clock speed and test for stability, in small steps, until it’s no longer stable. Then increase voltage slightly to make it stable, and repeat until either a thermal limit (too hot) or clock ceiling (where more voltage doesn’t help) is reached. But a little research on the Phenom II X2 550 showed that most samples continue to scale upwards at voltage levels beyond our desired limit. Because of this, we started with our target voltage and attempted to find the highest stable speed it would run at that voltage. The following BIOS images show the results of our efforts, so let’s discuss how we arrived at each setting.

The stock X2 550 clock speed of 3.10 GHz is attained by multiplying the HT clock of 200 MHz by 15.5. MSI's BIOS lists the HT clock as "CPU FSB frequency", though a technical inaccuracy, as AMD insists HT is not an FSB. Since this is a Black Edition processor, most of our overclocking efforts will focus on raising its 15.5x multiplier.












In MSI's BIOS, “CPU VDD Voltage” refers to the base voltage at which the processor is supposed to be detected, while the “CPU Voltage” setting acts as a type of fine-tuning for load voltage. We started with “CPU VDD Voltage” set to 1.50 volts and “DRAM Voltage” set to the memory manufacturer’s recommended 1.65 volts. The CPU multiplier, listed in the BIOS as “Adjust CPU Ratio,” was then increased to 16x.

We use Prime95 for stability testing, and find it handy that v25.8 build 4 (64-bit Window version) allows every core to be tested simultaneously from one application launch. A launch menu offers several types of tests. The “Small FFTs” option allows full CPU stress without much DRAM testing.

After around 20 minutes of load, we rebooted and increased the CPU multiplier to 16.5x, retested with Prime95, and continued this pattern until the system crashed at 18.5x. Detection program CPU-Z reported that the core voltage was dropping to 1.48 volts, so we went back into the BIOS and increased the “CPU Voltage” setting by 0.20 V (to 1.520 volts) as compensation.

Upon rebooting, the 18.5x setting was found to be Prime95-stable, so we continued making 0.5x increases until the system again crashed at the 21x BIOS setting.

Since we had already reached our target voltage, we tried dropping “Adjust CPU Ratio” in BIOS to 20.5x and let the stability test run longer. After around 45 minutes, the system again crashed. The same was true at a BIOS setting of 20x.

At a BIOS “Adjust CPU Ratio” setting of 19.5x, the system ran stable for several hours. Knowing that we could reach 19.5 x 200 but not 20 x 200, we began increasing the HyperTransport clock, which is the “200” part of 19.5 x 200. Using “Adjust CPU FSB Frequency (MHz)” in the MSI BIOS, we tried an HT clock of 202 MHz with great stability over a one hour test. We then tried 204 MHz and found the system crashed in around 45 minutes. At 203 MHz, the system crashed at around one hour of Prime95 test time, so we reverted back to 202 MHz and again found stability.













Though it also allows some overclocking adjustments, we used “AMD OverDrive Utility” primarily as a temperature monitor throughout testing. Note that its voltage monitor corresponds to the MSI BIOS' “CPU VDD Voltage” setting, not its “CPU Voltage” setting.

Note to non Black Edition CPU owners: Overclocking an AMD processor that doesn't allow multiplier increases requires raising the HT clock by a far greater amount. The higher data rate will eventually overwhelm the processor's internal HT link, but using the "Adjust CPU-NB Ratio" setting in BIOS to reduce the data multiplier can help. We generally try to keep the HT link data rate (listed below the adjustable setting in MSI BIOS) within 5% of its original speed when overclocking a "locked" AMD processor.

Though this isn’t a memory overclocking guide, we did want to optimize our modules for performance. Our Kingston RAM is rated at DDR3-2000, but the highest DRAM external clock rate available from AMD is four-times the CPU's HT clock. With a HT clock at 202 MHz, this corresponds to a DRAM external clock of 808 MHz and a DRAM data rate of 1616 MHz (see “FSB/DRAM Ratio” in the first BIOS screenshot above).

While the “DRAM Voltage” in the second screenshot above was set to the manufacturer’s recommended 1.65 volts, added stability can often be found by increasing memory controller voltage (“CPU DDR-PHY Voltage” in the same screenshot). With our DRAM data rate limited to 1,616 MHz, we used this added stability to enable lower latency, or wait time between operations, with our modules.



Starting at our memory’s DDR3-1866-rated CL-tRCD-tRP-tRAS timings of 8-8-8-20, we followed the same method as used for CPU overclocking to decrease each setting until the lowest stable timings were found. A bootable CD version of Memtest86+ v1.70 was used following each setting change for stability testing.  

[Gin Fushicho]

Overclocking AMD's Phenom II X4 955






Following the same method used for overclocking the Phenom II X2 550, AMD's quad-core Phenom II X4 955 Black Edition was first set to 1.50 volts for the CPU core, 1.65 volts memory, and 1.45 volts for the memory controller. As soon as we began increasing its CPU core multiplier (“Adjust CPU Ratio” below) however, we found that the CPU cooler simply couldn’t keep up with four cores at full load and our selected voltage.






Stability tests with four threads of 64-bit Prime95 revealed that our system would crash at a CPU core temperature of 59° Celsius, as monitored by AMD OverDrive Utility. We knew that 1.50 volts would be almost ideal for our tests, if only the CPU cooler could keep up. So, rather than start from stock voltage and work our way up, we began with 1.50 volts and worked our way down, until the core no longer reached the offending temperature.

At 1.48 volts the CPU would reach 59° Celsius at a multiplier of 18x, resulting in a Prime95 program error (worker stopped for one core, or program thread). Choosing 1.46 volts allowed a 19x CPU multiplier before the same error occurred at the same temperature. At 1.45 volts (and a 19x CPU multiplier) the program would crash before reaching 59°, indicating more voltage would be required to operate at this speed.

But those voltage levels were achieved using the “CPU VDD Voltage” setting in BIOS, where the “not enough voltage” crash occurred due to voltage fluctuation under full load. Increasing the “CPU Voltage” in MSI's BIOS to 1.480 volts allowed the system to actually run at 1.456 volts under full load with a maximum CPU temperature of around 55°.

Stable at 19 x 200, CPU frequency could still be increased slightly via HT clock, labeled “Adjust CPU FSB Frequency” in the first BIOS screen shot above. Adjusting in increments of 2 MHz, the system was found stable at 202 MHz HT clock, but crashed after around 40 minutes at 204 MHz. 203 MHz allowed it to run without error at full load for several hours, yielding a final overclock of 3.85 GHz.





Peak temperature increased to 56.5° Celsius, barely shy of the 59° limit where heat would cause it to crash. It’s important to note that temperature has far less effect on stability at lower clock speeds, so that systems unable to obtain this relatively mild temperature will be limited in overclocking capability.





At DDR3-1624, our memory supported the same minimum latencies as previously found in the X2 550’s DDR3-1616 DRAM data rate.  

[Gin Fushicho]

Overclocking Intel's Pentium E5200




As with AMD, Intel’s production technology already has well-known voltage limits affecting the majority of samples. For 45 nm processors based on the Core 2 architecture, CPU core voltage of 1.45 V is generally considered to be the maximum a processor can withstand over a period of many weeks or a few months. We’ve already seen an “office” system that was overclocked using 1.45 volts lose much of its overclocking capability over a period of around three months. This family of processors continues to scale well, even at much higher voltage levels, but cooling and longevity become an issue.

Because we want our overclocks to last at least several months (and we keep our fingers crossed for 1-3 years of reliable service), we chose 1.40 volts as a target setting under full CPU load and 1.43 volts peak under no load. Knowing the desired voltage level ahead of time negated the normal practice of increasing voltage in small steps until the system became stable after reaching a clock rate that would have otherwise been unstable.

The screenshots below show our final settings: P45 Diamond users must be forewarned that this memory setting required a jumper change which we’ll discuss further down this page.












Except for its Extreme Edition models, Intel doesn’t allow adjusting its CPU multipliers upward. Using a 12.5x multiplier and 200 MHz FSB clock (FSB-800 via QDR technology), the only way this 2.50 GHz CPU would go faster would be to set a higher FSB. Knowing that the Pentium E5200 should reach at least 3.60 GHz with air cooling, we first tried the next-higher Intel-standard FSB of 266 MHz (FSB-1066). The system booted normally and passed a 40-minute stress test using Prime95 v25.8 build 4. CPU-Z reported core voltage dropping to 1.38V under load however, so we increased the BIOS setting “CPU Voltage (V)” (second screenshot above) to 1.4132 volts. This resulted in 1.424 volts at idle and 1.408 volts at full load.

The MSI P45 Diamond supports most FSB clock settings, but we knew that the chipset would be most stable at or near one of Intel’s standard bus speeds. Our next attempt, FSB-1333 (333 MHz clock) would boot inconsistently, resulting in either a black screen or a system reset following the POST screen. After finding success at 320 MHz “most of the time,” we began to believe that our previous memory ratio problem was being caused by a dreaded bootstrap issue.









The P45 Diamond doesn’t have “bootstrap” settings in BIOS but does have two jumpers for altering detected bus speed. According to the manual, changing both jumpers from pins 1-2 to pins 2-3 would allow a 200 MHz FSB processor to be detected as a 333 MHz FSB version. Following those instructions solved both the memory ratio and boot inconsistency problems.

The system now booted at 333 MHz, but extended stability tests proved it wasn’t completely stable. In an effort to keep the system close to the 333 MHz “Intel standard” bus, we dropped the CPU multiplier to 12x.

The CPU clock of 12 x 333 MHz proved stable for a longer one-hour Prime 95 test. Jumping ahead slightly, 338 MHz was also stable. We continued increasing FSB and testing stability until it was found that the highest stable CPU speed was 4.1 GHz at 12 x 342 MHz FSB.







Using our memory manufacturer’s rated setting of 1.65 volts, we began chasing good memory performance to match that impressive 64% CPU overclock. Intel’s DRAM multiplier limit of 2 x FSB clock unfortunately meant that our RAM could be set no higher than a 684 MHz clock, corresponding to a data rate of DDR3-1368.

We began testing lower memory latency values to improve response time, using a bootable CD version of Memtest86+ v1.70 in the same manner that Prime95 was used for CPU stability testing. Since the memory controller is part of the northbridge, we experimented with increased “MCH Voltage” (second screenshot above) until it was found that anything higher than 1.352 volts provided no further improvement.  

[Gin Fushicho]

Overclocking Intel's Core 2 Quad Q8200





Intel’s value-priced Core 2 Quad Q8200 uses two of the same processor dice as the Pentium E5200, at a lower clock speed and a higher front side bus clock. The combination of moderate CPU frequency and higher FSB also requires a lower CPU multiplier, and Intel designs these so that the multiplier cannot be increased.

Intel typically uses a low FSB on mainstream processors to modulate performance and expand compatibility, so we’re not certain why the company chose FSB-1333 for its cheapest quad-core models. We do, however, know that many overclockers specifically select low-cost processors for the higher multiplier that typically accompanies a lower bus speed, so that its use of FSB-1333 for the Q8000-series has all but prevented its adoption amongst enthusiasts. After looking at the far higher prices of Q9000-series processors, we returned resolute to get big gains from the Q8200, viewing its lower multiplier as a challenge.

Unfortunately, the Q8200 would barely budge beyond its original 2.33 GHz frequency, regardless of how much voltage we applied to its core, reaching the same 2.5 GHz overclocked speed at core voltage settings from stock to 1.45 volts. The problem, it seems, is that FSB-1333 is almost the limit for these cores at stock FSB voltage.

Dual-die processors of this design use the front side bus for both CPU-to-chipset and die-to-die communication, and increasing the CPU FSB beyond 354 MHz (2.5 GHz CPU clock) would require an increase of the “VTT FSB Voltage” setting seen in the second screenshot below.









Our research showed that CPU FSB voltage had the same practical limit as core voltage: 1.45 volts peak and “something less” under load for continuous long-term use. As with the core voltage of the E5200, we chose 1.40 volts as a target voltage for Q8200’s FSB. We were then able to increase the FSB to 384 MHz, but the resulting 15% overclock is barely worth the risk and effort.










We really wanted to reach at least the next “Intel standard” FSB clock of 400 MHz, or FSB-1600, but getting there required far more “VTT FSB Voltage” than we can safely recommend. Further research into other far-more-successful Q8200 overclocks revealed that those units were actually cream-of-the-crop “Q8200S” models.

The CPU cores certainly wouldn’t need a full 1.40 volts at so low an overclock, so we began back-tracking. While “VTT FSB Voltage” remained at 1.40 volts for a stable 384 MHz FSB, we were able to drop the “CPU Voltage” setting to 1.30 volts. Anything less resulted in an eventual crash under Prime95 v25.8 build 4.

With the chipset’s maximum memory clock rate of twice the CPU FSB clock, the fastest selectable memory clock of 768 MHz provided a data rate of DDR3-1536. As with the E5200, we then began stability tests using Memtest86+ v1.70 at progressively lower DRAM latency settings until the best stable timings of 6-6-5-16 were determined.  

[Gin Fushicho]

Recommendations






Overclocking an unfamiliar processor to its limit usually requires increasing clock speed and voltage separately, in small increments, until additional voltage provides no increase in clock speed. But that unfamiliarity can lead to settings that significantly shorten the component’s lifespan. On the other hand, the old fashioned “safety” rule to increase voltage by no more than 10% would have left these products far short of their true potential.

Today we used four processors that have been on the market long enough to determine vital information such as the maximum tolerable voltage and life expectancy. Using slightly less than the “maximum long-term safe” voltage allowed an almost-stunning 64% clock speed gain on Intel’s Pentium E5200. Plus, if this architecture's history is a good indicator, there's a strong likelihood that the part will survive many months to several years of use. As stated in several System Builder Marathon articles, we specifically recommend this $70 processor for ultimate-value overclocking.

The biggest let-down was the Core 2 Quad Q8200, a part that actually contains two of the same dice as the E5200 under its lid. The problem of a “locked” multiplier becomes critical on an FSB-1333 CPU that can’t reach FSB-1600, and Intel doesn’t offer a cheaper, FSB-800 version to play with. If you wanted a more rewarding experience with this chip, you'd likely need to spend extra on the low-power 'S' version.

Anyone who really wants the best overclocking value from a quad-core could easily look to the Phenom II X4 955 Black Edition or Core i7 920 as safe bets. The Black Edition’s unlocked multiplier guarantees a lack of drama over HT clock, while the i7-920’s well-known tolerance to increased base clock speed makes unlocked multipliers an afterthought. But while our X4 955 Black Edition proved its capabilities at 3.86 GHz, the price of entry for the competing Core i7-920 was just a little beyond the budget of today’s guide. Disappointing headroom in the Q8000-series and a price-exclusion for Core i7 allow AMD's Phenom II X4 955 Black Edition to earn our value-overclocking recommendation for quad-core CPUs.
 

[Gin Fushicho]

CREDITS


All credits go to Toms Hardware