by Ilya Gavrichenkov
04/12/2010 | 10:30 AM
All CPU tests we conducted in our lab always include a few mandatory aspects. Among them are performance, overclocking potential as well as power consumption and heat dissipation measurements. This set of practical data turns out more than enough to shape up very insightful and detailed opinion about the solution using its price as a starting point. However, in this case a number of important issues that could be of interest to both: enthusiasts as well as just curious users, remain left out. So, today we decided to veer away from our traditional plan and carry out a not very common CPU test session. Namely, we are going to put aside pricing and absolute performance and focus mainly on overclocking and power consumption.
There is hardly anything surprising about the increase in power consumption when the processor clock frequency increases. We have long known that these two parameters are in proportional dependence to one another. However, overclocking requires core voltage increase, then the thermal and electrical parameters will increase according to square-law. However, these two elementary rules also contains a few unknown proportion coefficients depending on processor microarchitecture, number of cores, manufacturing process used for the CPU die, etc. that is why there is no universal formula that will allow to estimate the heat dissipation and power consumption of any overclocked processor. And since we can’t seem to find a definite theoretical answer regarding the effects of overclocking on the CPU power consumption and heat dissipation, we have no other choice but to turn to practical experiments.
This test session is going to be extremely interesting and acute because overclocking has become a very popular phenomenon these days. The times, when only selected enthusiasts with extensive knowledge of computer hardware and advanced soldering skills could handle overclocking are long gone. Today most existing platforms allow CPU overclocking that can be configured with a few simplest adjustments in the mainboard BIOS Setup. As for contemporary processors, they possess significant hidden frequency potential. Even without any specific technical tricks their frequency can almost always be increased by 20-30% above the nominal, and if you are lucky, you can push it up to 50% higher.
All this results from new way the CPU manufacturers started to declare their product clock speeds. It is no news any more that they use the same manufacturing process for the fastest as well as slowest models in the lineup, so that the junior models can eventually work at the top model frequencies. Overclockers have been taking advantage of this peculiarity of the production process for over two decades already. However, these days there is a new aspect to it. The top processor frequencies used to be set according to the frequency potential of the semiconductor dies. Now that the CPUs have become much more complex, which resulted into higher thermal and electrical readings, they also take into account such parameters as power consumption and heat dissipation when they determine the nominal CPU clocks. In other words, it is often not the potential of the semiconductor dies used for the processor in question, but their heat dissipation that act as a limiting factor for their nominal clocks.
For example, contemporary desktop processors have several typical limitations for average heat dissipation under heavy load: 130 W or 95 W for performance models, 73 W or 65 W for mainstream and budget models. As a result, although many mass production semiconductor processor dies can easily work at 4 GHz frequency, the actual processor built on them cannot boast having a nominal frequency like that, because in this case their heat dissipation will be beyond the acceptable maximum. The TDP restrictions are absolutely justified: first of all, they are determined by the efficiency of existing cooling systems priced acceptably for each given price range; and second, by the constructive peculiarities of processor voltage regulator layout on the boards.
This is where two possible conclusions come to mind. First, overclocking as a way to boost the performance is one of the effective and pretty affordable options for anyone. Contemporary processors have substantial hidden frequency potential, which can be put to good use without much effort. Most contemporary mainboards provide the users with all necessary tools for that matter. Second, if you decided to resort to CPU overclocking, you have to be prepared to deal with its increased heat dissipation and power consumption even when these values go beyond the calculated TDP thresholds. Processor cooling system should be efficient enough, and its onboard voltage regulator circuitry should be able to handle currents exceeding the nominal values.
In our today’s article we are going to discuss how processor power consumption (and heat dissipation that is directly connected to it) changes during overclocking. Namely, we do not need any tests to show us that platforms power consumption increases in this case. But it is definitely very interesting to see how greatly the electrical and thermal parameters of the processors in question change when their clock speeds increase. How big of a reserve should overclocker mainboards and cooling solutions have? How seriously will the system’s energy-efficiency suffer during overclocking? What power supplies would be sufficient for enthusiasts’ needs? These are a few questions our today’s article intends to answer for you.
In order to make the results of our today’s test session interesting and useful for as many users as possible, we decided to study the effects of overclocking on power consumption using several popular processor models featuring different internal design and microarchitecture. Therefore, we put together four popular platforms: an LGA775, LGA1156, LGA1366 and Socket AM3. That is why the list of computer components involved in our today’s test session turned out pretty substantial:
We used our in-house testing equipment that was described in detail in the article called “PC Power Consumption: How Many Watts Do We Need?”. Without going too much into details about the design of this unit we would like to say that the advantage of this device over electrical meters, amperemeters, current clamp meters, shunts and voltmeters is that it not only provides more precise readings, but also allows to interactively monitor the current changes along different power lines. As a result, we will offer you not only the average results, but also the maximum power consumption readings. We will also check not only the current sent to the mainboard through a special 12 V processor power connector, but also the currents the mainboard receives via standard 24-pin ATX power connector along 12, 5 and 3 V power lines. We must take all this data into account because many contemporary processors, mostly the ones made by Intel, use combination power circuitry that involves not only the individual 12 V processor power line.
I would also like to mention that when we talk about total system power consumption today, we will actually imply the reading taken not in front of but past the power supply unit. In other words, we do not take into account the PSU efficiency, but deal with “pure” power consumption readings taken off the individual system components separately and together as a whole.
Before we get to analyze the power consumption readings taken under different operational loads, let’s take a closer look at the CPUs that will participate in our today’s test session. In this part of our article we will try to overclock each processor and will see how its maximum power consumption changes.
AMD Athlon II X2 255 is the top processor model in Regor family that is based on proprietary dual-core semiconductor dies manufactured with 45 nm process. These processors are designed for Socket AM3 platform and are among the most affordable solutions for this platform. The nominal frequency of AMD Athlon II X2 255 is 3.1 GHz, and there is a 1 MB L2 cache for each of the two processor cores. You can see all CPU specifications on the following CPU-Z screenshot:
The nominal Vcore for our processor is 1.4 V, while the voltage of the North Bridge built into the processor is 1.175 V. The calculated TDP under load for this particular processor model is 65 W, according to the official specification. Our testbed equipped with this processor working in its nominal mode (without any overclocking) consumed the total of 111 W. Note that the power consumption measured along the 12 V processor power line was around 63 W, which is fairly close to the claimed calculated TDP.
As for overclocked mode without any Vcore increase, the CPU remained perfectly stable up until its clock speed hit 3.6 GHz. Further frequency increase was only possible if we increased the processor core voltage. As soon as we raised it to 1.5 V, we got our AMD Athlon II X2 255 working stably at 3.8 GHz.
I have to say that during our experiments we overclocked Athlon II X2 255 processor by raising the clock generator frequency, which means that not only the processor frequency increased, but so did the frequency of the North Bridge built into the CPU. However, it didn’t cause any problems for us: Athlon II processors do not have any L3 cache that is why they are less sensitive to overclocking than their elder brothers from the Phenom II series.
To reveal the type of dependence between the power consumption and clock frequency of our Athlon II X2 255, we took the corresponding readings in several key knots as stated in the table below:
All other voltages not mentioned in this table remained at their defaults.
The total power consumption of the test platform taken in the above mentioned system knots is shown on the graph. All readings were taken under maximum operational load created by LinX 0.6.4 utility.
You can easily notice that the power consumption increases mostly when the CPU Vcore is increased. Until this moment the power consumption graph remained pretty flat: 16-percent clock frequency increase from 3.1 to 3.6 GHz caused an only 8-percent increase in the total system power consumption. However, when the clock frequency increases from 3.6 to 3.8 GHz after raising the processor Vcore by 0.1 V, it leads to an additional 17-percent boost in power consumption.
The graph showing currents along the major mainboard power lines is an even better illustration for these numbers:
As we see, the biggest load falls upon the 12 V processor power line that delivers energy to our CPU. So, while overclocking barely affects the currents going through the 24-pin mainboard power connector, the power consumption along the CPU’s individual 12 V power line changes from 62 W to 91 W. Moreover, when we go from 3.6 GHz to 3.8 GHz frequency (namely, when further overclocking requires processor Vcore increase), CPU power consumption gets more than 20 W higher.
The second AMD processor participating in our today’s study, Athlon II X4 635, belongs to the same Athlon II family, but in reality is dramatically different from the X2 models. This CPU is based on a different semiconductor die aka Propus that is a monolithic quad-core die manufactured with 45 nm process. From the user prospective Athlon II X4 635 processor is especially interesting because it is one of the least expensive quad-core CPUs available in the today’s market. As for the specs, Athlon II X4 635 is designed for Socket AM3 platform and works at 2.9 GHz clock frequency. I have to point out that unlike Phenom II processors, Athlon II X4 has no L3 cache, and provides 512 KB of L2 cache per core.
The nominal Vcore of our Athlon II X4 635 was 1.4 V, and the voltage of the North Bridge integrated into the processor was set at 1.175 V. In other words, Athlon II X4 635 works with the same voltages as its dual-core brother. Nevertheless, twice the number of computational cores did affect the calculated TDP of this processor. For Athlon II X4 635 it is 95 W. As far as the practical values are concerned, our system equipped with this processor working at 2.9 GHz clock frequency consumed 146 W under heavy load, which is 35 W more than the same platform equipped with a dual-core Athlon II X2 255 would consume. The practical power consumption along the processor power line was 96 W.
I have to say that Propus processor family should be considered the least overclocking-friendly contemporary CPUs. While most widely available processors can hit frequencies up to 4 GHz, the tested Athlon II X4 635 could only go as high as 3.5 GHz. Moreover, to ensure that it remained stable at this frequency we had to increase its core voltage by 0.1 V. The maximum frequency we could get this CPU to work at without adjusting the core voltage was 3.4 GHz. Just like in the previous case, we overclocked by changing the clock generator frequency, because Athlon II X4 635 has a locked clock frequency multiplier.
Like in the previous case, we took a number of readings to check out the type of dependence between the power consumption and frequency. The table below described the key knots and major settings for our testbed:
All other voltages, not mentioned in the table above remained at their defaults.
The graph below shows total power consumption under maximum workload depending on the processor frequency.
This is not a new picture for us. While the CPU Vcore remains unchanged, power consumption increases strictly linearly and with a pretty low coefficient. However, once we slightly increase the CPU core voltage, we immediately see a dramatic increase in power consumption on the graph. For example, in our case the increase in CPU Vcore from 1.4 V to 1.5 V produces 25 W higher power consumption reading, although all other system voltages do not change and the CPU clock speed gets only 100 MHz higher.
The second graph here shows the changes in currents feeding the CPU and the mainboard during overclocking:
I would like to draw your attention primarily to the curve showing increase in the CPU current. At least, the change in processor frequency barely affects the currents fed to the mainboard through a 24-pin power connector, just like in the previous case. As for the processor power consumption, when we overclocked our Athlon II X4 635 from 2.9 to 3.5 GHz, it changed from 96 W to 137 W, and the lion’s share of this increase occurred in the interval between 3.4 and 3.5 GHz, when we had to increase the core voltage.
Besides Socket AM3 processors sold under Athlon II brand, we also took higher-end AMD CPUs for our today’s test session – Phenom II. Since the manufacturer offers solutions with different number of computational cores within this same family, we took one quad-core and one dual-core processor for our today’s investigative testing. For the dual-core model we went with Phenom II X2 555 - top Socket AM3 CPU with two computational cores and L3 cache memory. This processor is built on the same 45 nm semiconductor Deneb die as the quad-core Phenom II X4 CPUs. However, in this case two cores out of four are disabled. According to the specifications, Phenom II X2 555 works at 3.2 GHz frequency. Each of the two processor cores has its own L2 cache memory – 512 KB per core. Besides that, the CPU also has a shared 6 MB L3 cache.
Since AMD used the same 45 nm process for all their CPUs, it is not surprising that their electrical characteristics are similar. For example, our Phenom II X2 555 is designed for 1.4 V nominal core voltage, but differs from the above discussed Athlon II processors by slightly higher voltage of the integrated North Bridge that also contains L3 cache, which is set at 1.2 V.
The presence of L3 cache and higher clock frequency make the calculated TDP of the Phenom II X2 processors higher than that of Athlon II X2. The TDP for Phenom II X2 555, just like the TDP for other representatives of the same processor family, is set at 80 W. In reality, our Phenom II X2 555 processor consumed 74 W under maximum load and the total power consumption of the system equipped with it reached 123 W.
Deneb processor die is one of the best overclockable dies AMD has these days. Processors based on dies like that can often go as high as 4 GHz without any extreme cooling systems involved. Our Phenom II X2 555 was also no exception: we managed to push it as high as the notorious 4 GHz. However, we had to increase its Vcore by 0.15 V. but even if we hadn’t resorted to voltage increase, we would have still overclocked it quite nicely: it remained perfectly stable at 3.8 GHz with the nominal 1.4 V Vcore.
I have to say that Phenom II X2 555 belongs to the Black Edition series, which means that it has an unlocked clock frequency multiplier. Therefore, it is extremely easy to overclock without increasing the clock generator frequency or changing the frequency of the HyperTransport bus, memory or North Bridge integrated into the CPU. We definitely took advantage of this opportunity during our preliminary “estimate” test.
The table below shows the frequencies at which we took power consumption readings:
All other voltages that were not mentioned in the table above remained at their defaults. Moreover, I have to point out that this time we overclocked using unlocked clock frequency multiplier, which is a very convenient but not the most energy-efficient approach. The thing is that when you change the clock multiplier or set it to a certain fixed value, it disables Cool’n’Quiet technology that would drop the processor frequency in idle mode. Therefore, if you want to improve your system performance by overclocking and do not want to sacrifice the system energy-efficiency in idle mode, we wouldn’t recommend using the unlocked clock multiplier of AMD Black Edition processors.
The results of our total power consumption tests for the Phenom II X2 555 system taken under maximum load are given on the following graph:
Well, this is a pretty familiar picture that doesn’t surprise us at all. While we overclock our processor without increasing its core voltage, namely in the interval from 3.2 to 3.8 GHz, the total system power consumption increases linearly, by about 2-3 W per every 200 MHz. After that we see a sharp increase in power consumption caused by increased processor Vcore. As a result, the next 200 MHz on the CPU cost us quite a lot: 37 W.
The platform power consumption along the power lines is laid out below:
When the processor frequency as well as its core voltage rise, we can see the current increase along only one of the lines – the processor power line using a separate 8-pin cable. All other currents remain unchanged during overclocking. By the way, take a look how greatly processor power consumption may increase during maximum overclocking. Even the dual-core Phenom II X2 555 working at 4.0 GHz frequency needs 120 W of power, which is in fact 1.5 times higher than its calculated TDP. However, if you overclocked this processor without changing its core voltage, then the processor current won’t increase by more than 10%. Therefore, this overclocking should be considered absolutely safe: any mainboards, even the budget ones without enhanced processor voltage regulator circuitry will be able to cope with it just fine.
The last AMD processor that we decided to include into our today’s test session is Phenom II X4 965. It is the fastest and the most expensive Socket AM3 solution for desktops these days. Just like Phenom II X2 555, this processor uses a 45 nm Deneb semiconductor die, but unlike the former has four fully-functional computational cores. Each core has its own 512 KB L2 cache and they all share a 6 MB L3 cache. Phenom II X4 965 works at 3.4 GHz nominal frequency, which is the maximum clock frequency today’s AMD processors have hit so far.
As you can see from the CPU-Z diagnostic tool screenshot above, the default Vcore of our Phenom II X4 965 processor was set at 1.4 V. It must be the most popular Vcore setting for AMD processors manufactured with 45 nm process. The integrated North Bridge containing HyperTransport bus controller, memory controller and L3 cache worked at 1.1 V voltage.
AMD Company also ships Phenom II X4 965 modifications with different TDP of 140 or 125 W. We tested a newer modification of this processor based on C3 die revision. Its TDP was 125 W. This relatively high calculated TDP is actually not just a number: when we installed this processor into our testbed we registered significantly higher power consumption than in all previous cases. When the CPU utilization was at its maximum, the total system power consumption read 186 W. The highest power consumption of our Phenom II X4 965 processor working in nominal mode registered 137 W along the processor 12 V power line.
By the way, here is one interesting fact: the actual power consumption of the quad-core Phenom II X4 965 is almost twice as high as the actual power consumption of the dual-core Phenom II X2 555. It means that most of the energy inside the CPU is spent specifically by computational cores, while the shared parts, such as L3 cache or memory controller, contribute but slightly to the total power consumption score.
As I have already said, Deneb based processors overclock pretty well. Phenom II X4 965 once again proved that this reputation was well deserved. Our test processor working at 1.4 V Vcore remained perfectly stable up until its clock frequency hit 3.8 GHz mark. We managed to win another 100 MHz by increasing processor Vcore to 1.5 V. However, we still failed to hit 4 GHz barrier: the system would boot and would even pass some benchmarks successfully, but would fail the fully-fledged LinX stability test.
To check out the dependence of power consumption on the processor frequency, just like in the previous cases, we ran the tests in two modes with 200 MHz increment. Since Phenom II X4 965 belongs to AMD’s Black Edition series and has an unlocked clock frequency multiplier. So, we certainly used this feature during overclocking,
All other voltages not mentioned in the table above were at their defaults.
The curve below shows the dependence of power consumption on frequency in the above described modes:
The results are totally typical. While we do not touch the processor core voltage, there is a linear dependence between the power consumption and frequency (with certain acceptable measuring error margin). However, as soon as it comes to adjusting the processor Vcore, power consumption jumps up abruptly. In case of Phenom II X4 965, a 0.1 V Vcore increase results in about 40 W of additional power load on the PSU.
Note that all 40 W are consumed by the 12 V processor power line. The current along this line reaches an impressive value of 16 A during maximum overclocking of our Phenom II X4 965 processor.
It turns out that when we overclock our Phenom II X4 to 3.9 GHz, it starts to consume 190 W of power. This number is a perfect illustration of how greatly the processor voltage regulator circuitry on the mainboard gets overloaded in this case. Therefore, if you are going to overclock your system and increase the processor core voltage, then you should ensure that your mainboard has a quality voltage regulator that is capable of handling currents exceeding the typical ones.
While all current AMD solutions are unified for the same Socket Am3 form-factor, Intel processors are pretty diverse in this respect. At this time CPUs from this manufacturer support three independent platforms: LGA775, LGA1156 and LGA1366. We are going to start with the oldest socket type today. And the first CPU we picked for our LGA775 tests will be Core 2 Duo E7600 based on a 45 nm Wolfdale core that saw the light of day back in 2008. Processors designed in this form-factor has already moved into the low-end price segments, but they are still desired by many computer enthusiasts due to their excellent performance and overclocking potential. Unlike pilot CPU models on Wolfdale core, Core 2 Duo E7600 has slightly simpler specifications. Its clock frequency is 3.06 GHz, it supports only 266 MHz system bus and has a shared 3 MB L2 cache.
Although Core 2 Duo E7600 is manufactured with the same 45 nm process as all contemporary AMD processors, its nominal voltage is considerably lower than that of the AMD solutions. It was set at 1.275 V for our particular CPU. Moreover, it shouldn’t exceed 11.3625 V for the mass production processors of this type anyway. As we have already seen during our first tests, core voltage does affect power consumption and heat dissipation a lot, that is why it is not surprising that the TDP of our Core 2 Duo E7600 processor is only 65 W. In reality a complete system equipped with this processor consumed no more than 96 W (when the CPU was loaded to the fullest extent, but not the graphics card), and it means that Core 2 Duo E7600 is considerably more energy-efficient than Athlon II X2 255. Another great example, is the power consumption along the processor power line – during our tests in nominal mode it never exceeded 45 W.
Keeping in mind that Core 2 Duo E7600 supports only 266 MHz bus, it is fairly easy to overclock, even though Intel locks the clock frequency multipliers for all their processors except the most expensive ones. Without touching the core voltage we managed to get our CPU to work stably at 3.6 GHz, and he best result obtained in our testbed was 4.0 GHz. The table below lists all intermediate steps undertaken in order to reveal the connection between the frequency and power consumption:
All other voltages not mentioned in the table above were at their defaults.
I have to say that in this case the dependence of power consumption on the clock frequency promises to be somewhat more interesting than what we saw by AMD CPUs. Here we had to increase Vcore not only in order to conquer that last overclocking threshold, but even a little sooner than that. As a result, the flat curve on the graph and its abrupt growth should start not at the last overclocking increment, but at the one before.
And it is really so. As we see, power consumption starts increasing dramatically only when the processor core voltage increases. When we overclock with the default Vcore settings, every additional 200 MHz frequency increase produces only 2-3 W of extra power consumption. In other words, in terms of dependence of the total power consumption on the processor frequency and core voltage, LGA775 platform behaves just like Socket AM3.
However, the power consumption layout along the mainboard power lines looks completely different:
In fact, we could probably say that if we take a mainboard with a different implementation of the processor voltage regulator circuitry (for example, one from a different manufacturer), the picture will be different. Nevertheless, we notice significant currents going along 3 V power line that increases a little bit during overclocking. It seems logical to assume that this line powers the chipset North Bridge that contains the memory controller in LGA775 systems. As for the power consumption along the 12 V processor power line, it doubles during Core 2 Duo E7600 overclocking. It turns out that although this processor in its nominal mode consumes about 45 W of power under heavy load, 30% overclocking pushes its power consumption to 94 W. moreover, power consumption increases mostly at the last two increments, when in addition to raising the FSB frequency we also had to increase the CPU Vcore to ensure system stability.
The second LGA775 processor we selected to participate in our power consumption tests is a quad-core Core 2 Quad Q9505. This CPU doesn’t have a unique semiconductor die and is based on a combination of two Wolfdale dies manufactured with 45 nm process. Therefore, it is not surprising that this double die aka Yorkfield has unusual structure of the L2 cache memory that consists of two 3 MB parts, each shared between two cores. As for the frequencies, Core 2 Quad Q9505 works at 2.83 GHz nominal clock and supports 333 MHz FSB that serves not only to connect this processor with the chipset, but also to ensure proper communication between the core pairs that do not share any cache-memory.
It would be quite logical to expect the typical TDP of a quad-core processor like that to be twice as high as that of dual-core Core 2 Duo based on Wolfdale core. However, this is not quite the case: Core 2 Quad Q9505 TDP is set not at 130 W but at only 95 W. Of course, lower CPU clock frequency than that of the dual-core solutions is definitely a factor here as well as the peculiarities of the employed production process. The thing is that Intel selects semiconductor dies with better heat dissipation readings for their quad-core processors. As for less energy-efficient dies, they are being cut in half for dual-core processors. Therefore, it is not surprising that during our tests of the Intel Core 2 Quad Q9505 processor in nominal mode, its power consumption under heavy load was only 70 W. the total system power consumption in this case was about 125 W, which can be considered yet another piece of evidence that LGA775 platform is more energy-efficient than Socket AM3.
Overclocking quad-core LGA775 processors is not the easiest task to complete. The thing is that at certain bus frequency these processors start “acting out”, and very often the problems occur at 450-475 MHz FSB speed. Luckily, Core 2 Quad Q9505 has a relatively high multiplier of 8.5x, which allowed us to overclock it to 3.9 GHz without any serious problems. I have to say that just like with Core 2 Duo E7600, our quad-core test CPU worked stably at 3.6 GHz frequency and its default Vcore of 1.275 V.
To investigate the way power consumption changes during overclocking we tested Core 2 Quad Q9505 with 200 MHz increments, just like in all other cases. The major system parameters are listed in the table below. All other system voltages remained at their default values during this test session:
So, the system power consumption under full processor load depends on the CPU frequency as follows:
Since Core 2 Quad Q9505 overclocked pretty well in relative terms, we managed to take the power consumption readings off seven different knots. As a result, we can clearly see that if the processor core voltage remains constant, the dependence between its clock frequency and power consumption remains linear. After that, as soon as we pass 3.6 GHz mark, where not only the clock frequency but also the CPU Vcore start to change, every 200 MHz increase costs about the same in terms of power consumption as every 600-800 MHz before the 3.6 GHz mark. Overall, 27% frequency increase from 2.8 to 3.6 GHz produced a 19% increase in power consumption. However, overclocking to 3.9 GHz caused a 50% power consumption growth compared with the TDP in nominal mode.
As for the power consumption along different power lines, we can say that just like in the previous case overclocking causes a natural increase in currents going along 12 V power line assigned to the processor voltage regulator. Also the currents increased along 3 V mainboard power line, which we believe feeds the chipset North Bridge.
Processor working at its nominal frequency under 100% load in LinX test consumes 71 W of power. At 3.6 GHz frequency, when its Vcore remains at the default 1.275 V, its power consumption hits 89 W. during maximum overclocking to 3.9 GHz, when all major system voltages get set about 10% above their nominal values, processor power consumption rises to 136 W. this is, certainly, a lot, but it can’t compare with the power readings taken off the overclocked Phenom II X4. Therefore, the conclusion we drew above about LGA775 processors still being more energy-efficient than their Socket AM3 alternatives not only in nominal mode but also during overclocking remains intact and true.
Besides LGA775 we also included Intel processors designed in other form-factors. Namely, we couldn’t leave out relatively new solutions from Clarkdale family designed for LGA1156 platforms. The primary argument in favor of including them into this test session was the fact that one of the two semiconductor dies used in these processors is manufactured with the today’s most advanced 32 nm technological process. This die is in fact a combination of two computational cores. The integrated memory controller together with the built-in graphics core is inside the second die of the Clarkdale processor, which is manufactured with 45 nm process. We have already said that this configuration composed of two semiconductor dies within the same processor packaging doesn’t perform its best. Now let’s see what is going on with the power consumption of a CPU like that, when we overclock it to different frequencies.
For our tests we took Core i3-540. This is a mainstream Clarkdale model. On the one hand, it supports Hyper-Threading technology, on the other – doesn’t support Turbo Mode, which is otherwise not very handy for overclocking. The nominal clock frequency of this processor is 3.07 GHz. It has 256 KB of L2 cache per each of the two cores, and a 4 MB shared L3 cache.
32 nm production technology allowed using pretty low core voltage for Core i3-540 CPU. Our processor, for instance, required only 1.125 V of power. However, processor North Bridge that is located inside the second 45 nm die uses its own voltage regulator circuitry and its own voltage of 1.1 V for our processor. In this case the total calculated TDP for Core i3-540 processor is set at 73 W, which means that despite the finest manufacturing process, Intel didn’t really make any breakthroughs in terms of improving their processors energy-efficiency. In fact, 73 W is even more than the TDP of dual-core LGA775 processors based on 45 nm dies and priced in the same range. However, our test session revealed a completely different picture. A Core i3-540 based system with the CPU loaded to its fullest extent consumed only 86 W of power in nominal mode, which is lower than the power consumption of an Intel Core 2 Duo E7600 based system. It must be a significantly simpler chipset used in LGA1156 systems that contributed to this result, as now it only consists of the South Bridge, since all the functions of the North Bridge have been transferred over to the processor.
The clock frequency of LGA1156 processors is derived from the base clock generator frequency (133 MHz in nominal mode) times the multiplier preset and locked at a certain value for each particular processor model. Therefore, we overclocked our Core i3-540 processor by raising the base clock. I have to say that 32 nm Clarkdale behaved quite unexpectedly during our overclocking experiments. We managed to hit only 3.2 GHz maximum without touching the processor core voltage. After that we could only continue to overclock our CPU if we gradually increased its Vcore. And in order to get our Core i3-540 to work stably at 4.2 GHz we also had to raise the voltage of the integrated North Bridge. The table below shows all system settings, at which we measured the power consumption:
All other voltages not mentioned in the table above were left at their default values.
So, let’s take a look at the total power consumption of our Core i3-540 based system during overclocking:
I have to say that the graph looks quite unusual. There is no plateau phase and no dramatic increase in the end. Since we had to resort to voltage adjustment already in the second overclocking increment, all significant power consumption peaks have been distributed all over the graph. Nevertheless, 37% clock frequency increase during Core i3-540 overclocking to 4.2 GHz produces a tangible 50 W growth of the total power consumption. We have seen almost the same gain during overclocking of the dual-core Core 2 Duo E7600 and Phenom II X2 555 processors.
The second graph has a few more surprises for us. It shows the changes in currents along the major mainboard power lines.
First, I would like to remind you that LGA1156 processors are powered not only via the 12 V line. Only the CPU’s computational cores are connected to it. The second processor core that contains the memory controller is fed via 12 V power line connected to the 24-pin power connector on the board. Therefore, the current going through these two power lines increases substantially during CPU overclocking. Moreover, it is especially interesting that in a number of cases, for example when the LGA1156 CPU works in its nominal mode, processor power line is not bearing the maximum load. This is a unique feature of all Intel systems equipped with Clarkdale processors. Unfortunately, since LGA1156 systems use a “distributed” voltage regulator, we can’t give a definite answer about the processor power consumption at this point. Nevertheless, even a quick glance at the graph lets us conclude that when we overclock our Core i3-540 from 3.07 to 4.2 GHz, its power consumption more than doubles.
We tested the power consumption of LGA1156 platforms with more than just a dual-core Core i3-540 processor. Intel offers not only dual-core but also extremely popular quad-core Lynnfield processors for this platform. Of course, we couldn’t disregard these CPUs that is why we included Core i7-860 into our today’s test session. Just like the Core i3-540, this processor also uses Nehalem microarchitecture, but is based on a monolithic processor die manufactured using 45 nm process. Moreover, this die contains not only four computational cores but also an 8 MB L3 cache, a dual-channel memory controller and PCI Express x16 graphics bus controller. The Lynnfield model we chose, Core i7-860, belongs to the upper mainstream price range that is why it supports Hyper-Threading technology and Turbo Boost. As a result, even though its nominal clock frequency is set at 2.8 GHz, it can overclock on its own up to 3.46 GHz depending on the operational load at a given moment of time.
In brief, the idea behind Turbo Boost technology implies that the processor clock frequency may be increased painlessly , if its heat dissipation doesn’t ever exceed the TDP. For our Core i7-860 CPU it is set at 95 W. However, when all four cores of this processor are fully utilized, the processor clock frequency is limited by 2.93 GHz.
Unfortunately, we can’t say how much power Core i7-860 processor actually consumes under maximum load, because its voltage regulator has the same peculiarities as the voltage regulators of all other LGA1156 processors. The total power consumption of a system equipped with this processor never exceeded 155 W in our tests. However, it is obviously more than the LGA775 system equipped with a quad-core processor consumed. Therefore, we have very serious concerns that Lynnfield processors in fact are pretty power-hungry.
During Core i7-860 overclocking we first of all disabled Turbo Boost, because the multiplier changes initiated by this technology lowered the maximum stable frequencies. However, once this technology was disabled, Core i7-860 multiplier could be increased by one point above the nominal value. We did use this option during overclocking, but later on we resorted to base clock adjustment. The maximum frequency for our processor to remain stable under maximum operational load was only 3.4 GHz (that was without any changes to its Vcore that stayed at the nominal 1.125 V). However, Lynnfield core responded very positively to core voltage increase, so that we could eventually push our overclocking maximum to the 4.0 GHz frequency, which is typical for 45 nm processor cores. All the settings used during our overclocking attempts are summed up in the table below:
All other voltages, which were not mentioned in the table, remained at their defaults. However, it is important to understand that these values aren’t universal. Processor dies differ from one another in their characteristics, that is why when you overclock different processors, you may need to play with slightly different settings.
The total system power consumption increased as the processor frequency grew higher. Take a look at the graph below:
It is a very illustrative picture, I should say. Up until 3.4 GHz, the system power consumption increases little by little, by about 4-6 W per every 200 MHz. It once again confirms our assumption that overclocking without involving processor Vcore adjustment doesn’t affect the system energy-efficiency that much. However, as we pass 3.4 GHz mark, things start to change dramatically. Every additional 200 MHz result into 30-40 W power consumption increase. And that happens only because we keep pushing processor Vcore 0.1 V higher every step of the way to ensure that the processor remains stable.
The currents graph shows an even more vivid picture:
When we hit 4 GHz frequency, Core i7-860 power consumption along the 12 V line reaches 180 W! But as you remember, this processor also receives some power from another 12 V line. In other words, Lynnfield processors can easily compete against Phenom II X4 965 in power consumption during overclocking. Here we also see some currents that exceed 15 A.
There is one more platforms that Intel positions for the upper price range – LGA1366. CPUs designed for this platform are built around the same Nehalem microarchitecture, but they still have a few peculiarities of their own. It is these particular peculiarities that determined our decision to include these processors into our today’s test session devoted to system power consumption during overclocking. We chose a relatively inexpensive solution - Core i7-950. This CPU is based on 45 nm Bloomfield core that should be considered the first Nehalem silicon incarnation. Just like other LGA1366 processors (except the latest Core i7-980X), Core i7-950 has four computational cores, built-in triple-channel memory controller and a QPI bus controller (this bus connects the CPU with the chipset). I have to say that unlike LGA1156 systems, the PCI Express graphics bus controller in LGA1366 systems is located in a more traditional spot – the chipset North Bridge rather than CPU.
Speaking of the specific CPU we chose, Core i7-950, we have to say that its nominal clock speed is set at 3.07 GHz, but due to Turbo Boost technology it can overclock to 3.33 GHz under partial load. It also supports Hyper-Threading technology. Each of the four cores has its own 256 KB L2 cache. There is also an 8 MB shared L3 cache.
One thing immediately catches our eye: the core voltage of our test Core i7-950 is set at 1.2 V, which is higher than the core voltage of the Core i7-860 with the same microarchitecture. The voltage of the North Bridge integrated into the processor was also higher: 1.2 V instead of 1.1 V. And there is a reason for these differences. The TDP set for Intel LGA1366 processors is 130 W instead of 95 W, which means that the manufacturer takes into consideration higher voltages. Of course, it affects the power consumption of LGA1366 systems. In nominal mode we registered 190 W power consumption in our system equipped with Core i7-950 processor, which gives us every reason to regard LGA1366 platform as one of the most power-consuming configurations.
Despite this high power consumption and heat dissipation, Core i7-950 handles overclocking pretty well. We managed to get this processor to run stably at 4.2 GHz frequency. Up until 3.6 GHz this CPU overclocked even without any voltage increase. To build graphs showing the dependence of power consumption and currents on the frequency, we once again took some measurements with a 200 MHz increment. The table below shows all corresponding settings:
Note that the clock frequency multiplier of our Core i7-950 is locked that is why we overclocked this processor by raising the base clock frequency. However, due to Turbo Boost technology, we can set the multiplier one point higher than its nominal value.
The graph showing dependence of system power consumption on the CPU clock frequency under maximum load is of pretty typical shape:
I have to say that LGA1366 platform consumes a lot of power not only in its nominal mode. Things get even worse during overclocking, which is in fact, not surprising at all, because we have to increase the processor Vcore in order to ensure stability at high clock frequencies. As a result, at 4.2 GHz our Core i7-950 consumes 127 W more power than in nominal mode. Note that during our overclocking tests we only changed the CPU settings, that is why we owe almost the entire power consumption increase to the processor.
To prove it we would also like to show you the graph for mainboard currents:
The current going along the 12 V processor line more than doubles during overclocking. At the same time you should remember that the CPU takes some power that goes to the integrated North Bridge from the mainboard. Therefore, the increase in power consumption along the 5 V line should probably also be assigned to the CPU. And by the way, even though the power consumption of the Core i7-950 working at 4.2 GHz frequency looks quite dramatic, the currents in the power lines do not really increase too much up until 3.8 GHz frequency. That is why even in LGA1366 systems the biggest power consumption increase occurs only when we start raising the voltages.
Now that we have introduced to you the processors to be participating in our today’s power consumption research and discussed the peculiarities of their power consumption during overclocking, it is time to move on to the next part of our session; power consumption measurements during work in real applications. We ran the tests in several typical modes:
Each of the participating processors was tested in three most interesting modes:
I have to say that during CPU overclocking for this part of our test session, we resorted only to raising the clock generator frequency and left the processor clock multiplier at its default value. We used this particular overclocking technique, because once you change the clock frequency multiplier Enhanced Intel SpeedStep and Cool’n’Quiet technologies stop working, because they are based on interactive adjustment of the multiplier. For the same reasons we set the voltages using relative settings, instead of absolute values, because in this case power-saving technologies are still capable of lowering the processor core voltage in idle mode. We disabled Turbo Boost during overclocking of Core i3 and Core i7 processors participating in our today’s test session, because as you should remember from our previous articles, it lowers the maximum stable CPU frequency.
As a result, during power consumption tests in real applications, we used the following system settings:
Unless indicated otherwise, the diagrams below show average power consumption readings of the entire system (including a mainboard, processor, memory, graphics card, hard drive and processor cooler with a fan) under different types of operational load.
It is especially interesting to check out the system power consumption in idle mode when there is no workload in place because contemporary computer systems are in fact in this mode for quite a long time. For example, while you are reading this article the system is most likely in idle mode unless there are flash banners on the screen. Most of the time a computer system is idle even in those applications, where things happen only in response to user’s actions. A typical example of an application like that would be office tasks, for instance. In other words, even though system power consumption in idle mode doesn’t affect the cooling system or power supply requirements, but it certainly plays its role in the amount of your monthly power bill.
If we put CPU overclocking aside, then in terms of energy-efficiency we will see a pretty transparent dependence in idle mode. LGA775 and LGA1156 systems consume the least power of all. AMD platforms require a bit more power, while the system with an expensive LGA1366 processor needs over 1.5 times more power for its idle mode needs. Overall, the same situation repeats also during overclocking. Only quad-core processors on Nehalem microarchitecture, Core i7-950 and Core i7-860, behave differently here. The idle mode power consumption of platforms equipped with these CPUs increases during overclocking a bit more than in all other cases.
The workload created during image editing in Photoshop is interesting for power consumption tests because of its diversity. Filters and operations used for color photo retouching are optimized for multi-core processors in totally different ways and load the memory bus with different intensity. Therefore, the utilization of computational cores and overall system resources is extremely uneven in this graphics editor. As a result, processor cores would go in and out of different power-saving modes, and Turbo Boost technology would constantly work in corresponding Intel processors.
Power consumption remains relatively low under workload of this kind. Even overclocked systems with high-end processors never get beyond 200 W. However, the best result in terms of power-saving belongs to a Clarkdale based platform. No wonder, since these processors are manufactured with the newest 32 nm process. Although, you shouldn’t forget that dual-core processors cannot outperform quad-core ones. Therefore, you should definitely take into account not only Core i3-540, but also Core 2 Quad that can boast low power consumption during overclocking even despite having four cores onboard.
As we know from our performance tests, video transcoding is one of those load types that scales as the number of computational cores in the system increases. Moreover, this load causes the CPU to heat up pretty seriously and consume a lot of power.
Taking into account everything that has been just said, it is not surprising that the power consumption of overclocked systems is significantly higher than that of systems working in their nominal mode. We have just seen in our preliminary power consumption tests under maximum load that during overclocking the CPU current may more than double. Therefore, when we increase the CPU frequency and slightly correct its voltage, 30-40 W higher power consumption is considered normal. Only quad-core Nehalem processors, Core i7-950 and Core i7-860, do not fit the profile. Overclocking causes a much greater power consumption increase of 80-90 W in systems with these processors.
In terms of processor load final rendering in 3D modeling suites is very similar to video transcoding. This task can also be easily paralleled and loads the CPU pretty heavily pushing its power consumption and heat dissipation to the maximum.
It is quite logical that the results of our power consumption measurements are very similar to those that we obtained during video transcoding. I would only like to point out the fact that in this case power consumption is still a little lower than in the previous tests.
Our preliminary tests showed that the power consumption increases significantly during overclocking only when we need to raise the processor Vcore to ensure system stability. In this case we see actual proof of this conclusion in real examples. When we overclock our processors at their nominal core voltage we add maximum 10 W to the nominal system power consumption. Once we attempt to get our CPU to work at its maximum possible clock speed by raising Vcore and Vtt voltages, total system power consumption may increase by a few tens of watts. Moreover, strange as it might seem, but CPUs with more computational cores get more power hungry, Core i7-950 and Core i7-860 from Nehalem generation being the most energy-demanding.
Until this moment we have been talking about applications that load primarily the CPU, but not the graphics sub-system. Of course, graphics card was used to display images on the screen, but in this case it worked only in 2D mode, and ATI Radeon HD 5870 graphics accelerator that we had in our testbeds is extremely economical in this mode. It consumes no more than 25 W. The performance of our test systems in 3D is a completely different story, as not only the CPU but also the GPU contributes a lot to the total system power consumption score. In order to check how overclocking affects the power consumption of gaming systems we included a popular 3D shooter into our today’s test session.
Here average processor power consumption is significantly low that during rendering or video transcoding. However, the graphics card does consume much more power. This relocation of the power consumption focus allows us to conclude that relative effect from CPU overclocking on the resulting power consumption readings has become lower. Yes, overclocked gaming systems do need more power in games. But if we compared the obtained results with one another we will see that even when we raise the processor core voltage it in fact contributes no more than 20% into the total system power consumption under gaming load. Overall, average power consumption of overclocked systems (with one contemporary graphics accelerator inside) is about 200-250 W during active gameplay.
Applications like 3D modeling or video transcoding and editing suites do load processors pretty heavily. However, there are even harder tasks for processors out there. For example, Linpack – a software library for solving linear algebraic equation systems. It was used to create LinX test utility, which we use to load processors to the utmost extent (in terms of power consumption).
Power consumption in this test does in fact turn out pretty high. For example, overclocked systems with the least energy-efficient quad-core Phenom II X4 965, Core i7-950 and Core i7-860 processors consume about 250-350 W of power. And these numbers are really high and they look even more impressive especially against the background of non-overclocked platforms with dual-core Core i3-540 and Core 2 Due E7600 processors that only require about 100 W.
The results of LinX are indeed very high, but they aren’t yet the maximum, because this utility only loads the CPU. in order to also load the GPU we launched another graphics test, Furmark, simultaneously with LinX. This test initiates maximum graphics card power consumption. This is what the average power consumption readings taken off this artificially created environment look like:
Well, there is nothing to comment on here. Overall, we see exactly the same thing as in the previous graph. Only the absolute numbers have become higher, which is in fact quite logical, since the graphics card did contribute its fair share this time. However, even though we managed to significantly increase the power consumption of our test platforms in this somewhat artificial test compared with the results we have just seen under more realistic operational loads, the power supply makers will obviously be disappointed. Even the most power-hungry LGA1366 system consumes no more than 500 W of power. As for the power consumption of all other tested platforms, it never exceeded 350-400 W.
Although we tried to recreate maximum possible load in our previous tests, we still can’t use these results to determine the required power supply capacity. The thing is that we have dealt with average results, which do not coincide with the peak ones even when we used special utilities loading the CPU and GPU to absolute maximum. It mainly occurs because the power consumption of the graphics sub-system is extremely uneven: it needs dramatically different amount of power for rendering a frame and then displaying it on the screen. That is why we decided to add one more diagram to what we already had: it shows absolute maximum power consumption of the platform in question registered over the entire test period. These numbers will allow us to conclude what PSU capacity we will need for our overclocked system in every given case.
Contemporary systems based on non-overclocked processors and equipped with a single-GPU graphics card, can in most cases be just fine with a quality 300-400 W power supply unit. Moreover, 400 W capacity will only be needed for the most resource-hungry quad-core processors like Core i7 or Phenom II X4. 400 W PSUs may also be quite enough even for overclocked processors, when its core voltage is not increased. However, if you intend to boost your system performance more substantially by overclocking it, then you absolutely need a serious power supply system. And as you can see from the obtained results even a 500 W power supply may turn out not good enough in this case. For example, when we overclocked our Core i7-950 processor to 4.2 GHz, the maximum power consumption hit 530 W. and that is also not the highest possible number, I should say, because this time we didn’t overclock our graphics card, which may make a significant contribution to the total system power consumption as well.
When we say that different processors working at different clock frequencies have different power consumption levels, we should keep in mind that their performance is different, too. For example, a processor may be extremely economical but at the same time so slow, that solving even the simplest tasks in systems based on it may take forever. Therefore, processor manufacturers have been actively lobbying the “performance-per-watt” complex parameter that describes the value of each watt of power for the overall system performance.
Estimating the energy-efficiency of an overclocked system is also a very interesting topic. Overclocking not only increases performance, but also affects the power consumption, as we have just seen. Therefore, we decided not to overlook it. however, instead of an artificial “performance-per-watt” parameter, we decided to study a different and more understandable one – the amount of electrical energy necessary for different systems to complete the same calculations. In other words, we decided to measure how many watt-per-hour we would need for typical tests. And the time it takes for these tests to be completed is directly dependent on the CPU speed. I also have to say that “power consumption” and “performance-per-watt” parameters are in inverse relation with one another. In other words, the results we obtained will allow us to draw conclusions about the connection between performance and power consumption.
Very unexpected, don’t you think so? It turns out that despite the growing CPU power needs during overclocking, we often gain rather than lose when it comes to energy consumption. The thing is that overclocking increases not only power consumption, but also performance. As a result, an overclocked system may cope with the same amount of work faster than a non-overclocked one, which produces certain savings right away. However, it is important to remember that if system energy-efficiency is the primary objective, then you shouldn’t get too involved with overclocking. By raising the processor Vcore, you trigger a sharp increase in power consumption, which won’t be compensated by the performance increase, as you can see from our test results. In other words, overclockers will hardly be pleased with their power bill, but it will be OK with those who overclocked carefully, didn’t raise the processor core voltage and kept the power-saving technologies up and running.
The new type of processor load didn’t affect the unexpected regularities that we had uncovered in our power consumption tests during photos retouching in the image editing application. Although Photoshop CS4 doesn’t always load all processor cores and video codecs on the contrary create pretty severe processor load, the correlation between the results of overclocked and non-overclocked processors remains the same.
And you shouldn’t be surprised with what you see. When the CPU frequency increases and the core voltage remains constant, the system power consumption grows linearly, and the constant of proportion in this case is relatively low. The time it takes the processor to complete its tasks also decreases linearly as the CPU frequency increases. However, since only the processor contributes to the system power consumption increase during overclocking, while other platform components do not change their power characteristics, the performance increases faster than the power consumption. The result in this case is quite logical: if the processor Vcore doesn’t change and remains at the nominal level as its frequency increases, overclocking ends up being beneficial in terms of power consumption minimization.
We see exactly the same thing during final rendering. By overclocking the processors without adjusting its core voltage you can save some power. More aggressive overclocking when we do increase the processor core voltage, on the contrary, causes higher power consumption despite higher performance.
Note that quad-core processors are almost always better than dual-core ones in terms of power consumption. Yes, these CPUs have higher performance-per-watt, but only in tasks optimized for multi-core systems. At the same time, the energy-efficiency leadership goes to Intel processors: the previous-generation ones based on Core microarchitecture, as well as some newer CPUs based on Nehalem microarchitecture, including Clarkdale and Lynnfield solutions, but not the older Bloomfield ones. It looks like Socket AM3 platforms are far not the most efficient choices.
3D games create a completely different type of load. Here, processor overclocking doesn’t speed up the calculations, because the time it takes for the game to run depends solely on the gamer’s skills, but absolutely not on the fps rate. Therefore, there is a different dependence between overclocking and power consumption in this case.
CPU overclocking, on the contrary, may increase the fps rate, namely, the number of frames that the CPU transfers to the graphics card for further rendering. As a result, the processor power consumption is not the only one increasing: so does the power consumption of the graphics card, because it receives additional load. Therefore, any type of overclocking applied to gaming systems always causes power consumption increase. That is why the only way you can save from overclocking is when the time it takes the processor to complete a certain task is directly connected with its performance. games are not one of those cases, so overclocking may only be interesting for the sake of better visual quality and lower response time to gamer’s actions.
In our articles we always talk about processor overclocking potential and even benchmark overclocked systems. However, it is important to understand that CPU overclocking is not only an affordable way of increasing the system performance without any additional investments. Although overclocking has gone down to the consumer level these days and doesn’t require the user to possess any specific skills or knowledge, there are a lot of hidden obstacles hiding behind simple adjustment of the BIOS Setup parameters. Today we tried to reveal one of these obstacles that many computer enthusiasts face: CU overclocking increases its power consumption. Moreover, it is fairly easy to push the processor power consumption far beyond its calculated TDP and everyone should be prepared for that. Processor cooling system should be able to dissipate a lot of heat. Processor voltage regulator circuitry on the mainboard should have more than twice the “current reserve” available. And the system power supply should have at least 1.5 times the power capacity of the system working in its nominal mode under maximum load.
So, it turns out that you can’t really overclock without any additional investments. Moreover, you have to make them beforehand, even when you are just putting the system together. And unfortunately, we can’t really neglect them: we always experience power consumption increase during overclocking, and it doesn’t matter what platform and what processor we have. However, we don’t want to overestimate the power consumption increase during overclocking, either. As our tests showed, there is always a cheaper” alternative.
If we look at the power consumption graph during overclocking of almost any processor, we will see, that it consists of two very characteristics parts. The first one is a relatively flat part of the graph that increases slowly, and the second one is the part where the graph starts growing very abruptly. The turning point on this graph doesn’t occur spontaneously, but happens at a very specific moment of time – namely, when it comes to raising the CPU Vcore. In other words, you don’t need t fear that the processor power consumption will skyrocket, until there is real need to adjust any of the system voltages to ensure stability. As for overclocking at nominal voltages, even though it doesn’t allow you to hit very high performance levels, it is not so useless after all. For example, several processors from our today’s test session worked at 3.6-3.8 GHz frequency and at their default core voltage. And by the way, overclocking like that doesn’t really stress the cooling and power systems in any way, but also allows to save some power when working on some resource-hungry tasks.
In conclusion I would like to stress that when you overclock different processors up to about the same frequencies, they still consume different amounts of power. It is quite obvious that quad-core processors consume more power during overclocking. However, this is not the only dependence that we observed. Namely, LGA775 CPUs proved to be the most energy-efficient during overclocking. Also, Athlon II processors demonstrated very modest power appetite. I would also like to point out Core i3-540 processor. This is the only CPU tested today that is manufactured using the most advanced 32 nm technological process, which had to affect its results. It obviously consumed the least power of all at its nominal frequency as well as after overclocking.
As for the power consumption “leaders” they are Phenom II and Core i7 CPUs. Overclocking these processors causes a much more substantial increase in the system power consumption than in all other cases. And I am not talking only about the absolute values here. Even in relative prospective, overclocking these CPUs may cause about 4—50% increase in system power consumption. Therefore, if you are looking for a suitable power supply for an overclocker system, make sure that you will have at least that much capacity in reserve.