Technology And Features To Reduce Power Consumption
A processor’s power consumption can be reduced in two ways. The first way involves the use of new manufacturing processes/techniques and the second consists of special architectural features.
Power Savings By Technology
Intel has been able to revamp its manufacturing processes roughly every two years as it continues to successfully output denser devices. The 65 nm manufacturing process, which came online in 2005, was replaced by the 45 nm process in 2007. In 2009, Intel will likely ramp up its 32 nm process. In recent years, every new generation offered increased performance with lower power requirements both in idle and at peak loads, which compensated for how smaller transistor sizes typically increased the risk of power leakage.
| Process | Code Name | Introduction | Main Feature |
|---|---|---|---|
| 90 nm | P1262 | 2003 | Strained silicon, copper interconnects, low-k dielectrics |
| 65 nm | P1264 | 2005 | 2nd-generation strained silicon, high-k gate dielectrics |
| 45 nm | P1266 | 2007 | Hafnium-based high-k gate dielectrics + metal gates for reduced leakage power |
| 32 nm | P1268 | Expected 2009 | ... |
| 22 nm | P1270 | Expected 2011 | ... |
Intel says that the transistor gate oxide leakage could be reduced tenfold, while processors produced with the 45 nm process had only one-fifth of the source-drain leakage compared to devices produced with the 65 nm process. While it’s hard to verify these statements, we can definitely confirm that the Core 2 Duo E8000’s power consumption is significantly lower than that of 65 nm E6000 series devices. Each new processor series has thus offered an increase in power efficiency.
Power Saving Features: CPU C-States
New features can decrease power consumption as well. The first way is to use different CPU power modes, which are referred to as C-states. While these can be applied to operating system C-states, the following table applies to an entire processor.
| C-State | AMD | Intel | Transition Time Back To C0 |
|---|---|---|---|
| C0* | N/A | ||
| C1* | Halt State for main CPU clocks | ~ 10 ns | |
| C1E | Enhanced Halt - Stops CPU clocks | Enhanced Halt - Stops CPU clocks and reduces Vcore | >10 ns |
| C2 | Stop Grant - Stops internal clocks via hardware, includes I/O buffers to save system Power | ~ 100 ns | |
| C2E | N/A | Extended Stop Grant - Stops CPU clocks via Hardware and reduces Vcore | > 100 ns |
| C3 | Sleep: Stops all internal CPU clocks Deep Sleep: Stops all internal and external clocks. | ~ 50 µs | |
| C4 | Deeper Sleep: Reduces Cpu Voltage | ~ 150 µs | |
| C4E / C5 | Enhanced Deeper Sleep: Further reduces CPU voltage and switches off cache | ~ 250 µs | |
| C6 | Deep Power Down: Further reduces CPU Voltage CPU context not preserved | Unknown | |
* required in all modern processors
Note that not all processors support all C-states, and that processors may only support some of these. Mobile processors typically support more of the higher C-states, while desktop processors are oftentimes limited to only a few C-states. Since marginal power savings are less useful on the desktop, but may be significant for laptop computers, this differentiation makes sense.
There is a reason why we list all these technical details: Switching from full-power operation mode into a low-power mode takes time. Please see the right column of the table above for details. While switching from C1 or C2 back to C0 mode can be done with hardly any delay, the deeper C-states require more system parameters to be modified, hence it may take up to 250 micro seconds to go back to full-power mode.
Power Saving Features: CPU P-States
Then there are AMD’s Cool’n’Quiet and Intel’s SpeedStep or Enhanced SpeedStep (EIST) power savings features, with which most users are already familiar. Cool’n’Quiet and EIST are available in almost all processors that are on the market today. While the C-states only apply when the processor is in true idle mode or is requested to switch into sleep mode (e.g. when putting the system into standby), the power-saving features make use of so-called P-states (performance states), which AMD and Intel implement in a similar way today.
As long as the CPU is in active mode (C0), Cool’n’Quiet or Enhanced SpeedStep may kick in. It is necessary to have BIOS and operating system support. With this requirement satisfied, Windows will trigger clock speed and voltage throttling for the processor, typically reducing AMD clock speeds to 1.0 GHz and Intel speeds to anywhere between 1.2 and 2.0 GHz, depending on the front side bus (FSB) speed. These features are crucial when it comes to reducing power for everyday applications.
Correction: Graphics card in set-up should be a 4850, never heard of a 485 it must be pretty fancy
Interesting.
So there's basically no real difference. Might as well save some power.
Thanks!
The difference was very small (2%) in performance, but the power consumption difference could have been much smaller.
The PSU that was used was WAY oversized and was only loaded 10-21% of capacity during the entire test. PSU efficiency in that range is terrible. A 430-500W PSU (like the Earthwatts series) would have been reasonably loaded and the efficiency would not have been so drastically different across the range of loads.
Efficiency example
The difference in efficiency between 10 and 20% is massive, compared to 30-40%.
You guys should do the same test on Phenom and Athlon (K8). Phenom's cores can run at different speed. CnQ also put CPU cores at lower frequency (1GHz or 1.25GHz) compare to Intel SpeedStep. Performance impact from energy saving on Phenoms is much obvious.
I agree, you need to do a follow up with the Phenom. I have experienced erratic performance with my X4 9650 @ 9950 in some cases. For example if a GPU limited games puts the CPU load of a core just at the threshold value for the power saving to kick in, that core will keep going in and out of power saving mode, causing stutters and lower overall framerates (>10% difference). I've experienced this in 3DMark 06, UT3 rolling demos and the X3 space sim among other games. The same thing might probably happen during other tasks such as file compression. Turning off power saving solved all such problems. I've now made it a habbit to always turn off power saving before launching games or compressing large archives.
i totally agree. this is why AMD is still in the CPU market, they make a much more efficient cpu when it come to energy efficiency, and the price is not that bad at all.
this is why AMD is still in the CPU market, they make a much more efficient cpu when it come to energy efficiency
BS
Stop perpetuating a lie. Intel took over the lead long ago. Athlon is much better than any Netburst, but Core 2 is much greater than anything AMD had ever put out.
i totally agree. this is why AMD is still in the CPU market, they make a much more efficient cpu when it come to energy efficiency, and the price is not that bad at all.
Only partly right. Those low power Brisbane Athlons are good, but Phenoms are still power hogs even under CnQ with highest possible energy saving option.
yep, core2duo's are better in efficiency than athlons since it finishes tasks in less time.
anyways, we started noticing power efficiency when that article came out of overclocking a pentium M that performs far better than a pentium 4 using less power and that was really amazing. that was the grandfather of the core2. hehe
I don't really understand your point because they were not looking at the power efficiency of the PSU or how different PSUs can effect CPU efficiency. They were trying to solely look at CPU efficiency and how power saving on CPU's affect performance. To do this I think they chose an oversized PSU so that the PSU was excluded as a variable, or in lay-mans terms: it had as little possible affect on the CPU comparison. This is because the cpus could draw as much power as they wanted to.
I don't think that the wattage of a PSU will change the efficiency of a CPU, it changes the efficiency of the PSU! It can effect the performance of a CPU by limiting its power supply, but it does not change its efficiency - it only restricts the CPUs potential.
PSUs are however very important in system efficiency as they are the main link between the external power supply and the system. Having a good PSU that is balanced to your system power needs will give you better overall efficiency as you stated.
A very interesting and informative article, thanks TOMs! I thought the power saving modes were simply gimmicks that made unattractive sacrifices in performance, seems I was wrong in this case.
I don't really understand your point because they were not looking at the power efficiency of the PSU or how different PSUs can effect CPU efficiency. They were trying to solely look at CPU efficiency and how power saving on CPU's affect performance. To do this I think they chose an oversized PSU so that the PSU was excluded as a variable, or in lay-mans terms: it had as little possible affect on the CPU comparison. This is because the cpus could draw as much power as they wanted to.
I don't think that the wattage of a PSU will change the efficiency of a CPU, it changes the efficiency of the PSU! It can effect the performance of a CPU by limiting its power supply, but it does not change its efficiency - it only restricts the CPUs potential.
PSUs are however very important in system efficiency as they are the main link between the external power supply and the system. Having a good PSU that is balanced to your system power needs will give you better overall efficiency as you stated.
Because they measured the power consumption at the wall (i.e. AC) not the DC the system actually uses, that's why.
Ideally they should have used a PSU sized so that the efficiency doesn't differ more than 2% during the entire scope of testing, to eliminate it's influence. They had a PSU sized that the efficiency differed ~8-10% during the measured values.
Ah, well in that case you are right.
What someone needs to do other than testing power efficiency is test how low in power AMD, and Intel can go. I have seen data on an Intel E2140 CPU based system that uses ~50W, and I have an older AM2 1210 opteron system (CPU TDP of 103W ) that uses less than 80W including a 19" WS LCD. My gaming rig on the other hand which is an Intel P35/ICH9R board, and an E6550 CPU, with an nVidia 9600GT *can* use up to ~220W while playing World in Conflict ( which I have found to be the most CPU/GPU intensive game that I currently own). Idle on this system is ~168W, which is where this system peaks with a 7600GT for discrete graphics.
Anyhow, my point here is that while efficiency may be important in a data center, lower system power can be more important at home. Maybe a lower powered system will take longer to do CPU intensive tasks, but at home a computer is going to be idle a lot. If you're a gamer perhaps not, but a person running off solar/batteries, definitely. Web browsing, and similar tasks barely use any CPU %. Even with this opteron I have, underclocked to 500Mhz, at .8v does this perfectly fine. Sure, and Atom based system could do this fine to, but does not have the potential to be bymped back up to normal when and if the need arises.
Anyhow, I am thinking with an AMD 740G based board, and an AM2 4850e, there is no Intel based system that can match it in performance, and power consumption. Even *if* an E2140 could match a 4850e in performance, I doubt the total system power consumptions could be matched. Also, let us not forget about the AM2 on CPU memory controller, which can improve performance greatly even while underclocked. Either way, I would like to see some data proving, or disproving my theory here.