Overclocking and Intel’s Power Management Settings (EIST, C-States, Turbo Boost)
Does enabling EIST and the C-States after you are done with overclocking cause any performance impairments or other problems? Should they be enabled or not? We’ve managed to reach a conclusion by measuring performance with both purely raw performance oriented (Cinebench, PerformanceTest etc.) and real-world oriented (Final Fantasy XIV, GTA V) benchmark tools. But first, let’s elaborate.
I have to point out something before I begin, in order to eliminate any possibility of confusion. You should disable features like EIST, Turbo and C-States in your BIOS while you’re overclocking (while you’re playing with your CPU’s frequency, voltage and so on). This work focuses on said features’ advantages and disadvantages after the overclocking is done and a frequency/voltage balance is set. If you try to overclock with these features on, you might have certain stability issues, especially in the voltages department. Let’s get you familiar with these features and what they do before we get to testing.
1) Idle Temperatures
3) AIDA64 GPGPU
6) Final Fantasy XIV Online: Heavensward
7) GTA V
E. Conclusion and Thoughts
B. What are EIST, Intel Turbo Boost, Enhanced Turbo Boost and the C-States, and what’s their function?
- EIST (Enhanced Intel SpeedStep): This is a feature that Intel has been using since Pentium III. It saves power and cools your processor by dynamically changing the clock frequencies. Working together with the C-States, EIST tries to figure out the right frequency that you need at that exact moment through raising or lowering the clock multiplier and also adjusts the voltage accordingly to save even more power. This is why it can result in inconsistent results and system crashes while overclocking. For example; EIST adjusts the clock multiplier of the processor I used in this experiment, the i7 4790k, somewhere between 7-47 and the voltage somewhere between 0.10-1.27 V according to system load.
- Intel Turbo Boost: It raises the clock frequency of the processor to a manufacturer defined turbo speed. System load, active cores, estimated current, power consumption and the core temperature are taken into account in the boosting process. So it can, just like the EIST, cause inconsistent results while overclocking. Also usually, it raises the frequency of just the first two cores instead of all. For example; the i7 4790k I used in this experiment operates at a default frequency of 4.0 GHz, and the turbo boost raises the first two cores’ frequencies to 4.4 GHz when it’s necessary. If the core temperatures hit somewhere above 65°C, the turbo boost deactivates itself. This feature has to be enabled in BIOS for it to function. By default, it is enabled.
- Enhanced Turbo Boost: This is a feature that MSI brings with the mainboard I used in this test, but other brands have similar features as well. Like I just mentioned, as the Intel Turbo Boost raises the clock frequency of just the first two cores, Enhanced Turbo Boost forces the processor to raise all core frequencies to the manufacturer defined turbo frequency. For example; as I mentioned the i7 4790k I used in this experiment has a stock frequency of 4.0 GHz and a turbo frequency of 4.4 GHz. While by default the turbo raises the first two cores’ frequencies to 4.4 GHz, the Enhanced Turbo Boost just makes them all go 4.4 GHz. This feature has to be enabled in BIOS for it to function. By default, it is enabled.
- C-State: C-States are the “states” that your processor comes to, in order to lower its power consumption and temperature, and their number varies between processors. All CPUs have the C0, C1, C2 and the C3 states by default. Also different cores of the processor can be in different C-States at the same time.
C0: The default working state of a CPU.
C1: The state of the processor when the CPU is idle, but it’s able to instantly revert to its working state. C1E (C1 Enhanced) is the updated modern version of the same state.
C2: The state of the processor when the CPU is idle, and it can revert to its working state with some delay. Most of the current CPUs don’t use this state as it causes lags.
C3: The default sleeping state of a CPU. It’s the most power efficient state of a processor, and only the actual shutdown state can pass its efficiency. All cores’ frequencies scale down to the bare minimum. The L1, L2 and L3 caches are flushed. Not all CPUs use this state, and some of them have newer, upgraded versions of it.
C6-C10: In these states the Intel CPUs respectively flush their L3 caches, start parking their cores and lowering the voltage:
In the C6 state, the CPU prepares to lower its voltage and takes “one last picture” by saving its current state.
When the CPU comes to the C7 state, the CPU starts to flush its L3 cache. If the L3 cache is able to be entirely cleared, the CPU cuts its power to save energy.
In C8, CPU saves its state one more time and lowers the voltage even more. The power to the PLL is cut.
In C9, the VCCIN (VCC Input Voltage) gets lowered to a minimum.
When the CPU hits C10, the single phase core management system developed by Premier Farnell and adopted by Intel, the VR12.6, goes into a low-power state. The CPU is almost shut down.
The i7 4790k I used for this article is a fourth-generation Intel processor which has all the aforementioned C-States. This feature has to be enabled in BIOS for it to function. By default, it is enabled.
Right about now you might be thinking “Why, though? Why all these states and stuff exist? Can’t the processor just shut itself down and restart when it’s needed?” The answers of these questions lie in the very basis of processor technology. Processors cannot continue to operate from the same state after they’ve been shut down because of their nature. So your computer’s processor has to be busy while it’s on, even if you don’t use it at all. The system must be booted again if the processor is shut down, in order for it to operate again. These facts force features like low-power or sleep modes to try and save power through slowing down the processor (by lowering the voltage and frequencies) or through shutting some of the cores down, but without shutting the processor off. Even the pseudo-process named as the “System Idle Process” in your Task Manager, which seems to be eating 99% of your CPU from time to time, is there because of this. Even if you’ve hanged yourself in your room and are long gone from this petty world, -on the assumption that your computer isn’t set to automatically go into some sleep or low-power state- this system process will make sure your processor stays awake by signaling it that it’s busy. So this is why all these complex technologies are developed, and why manufacturers are trying to lower the power consumption and temperatures of their processors without actually turning the damn thing off.
1. The Computer
OS: Windows 7 Ultimate 64-Bit
Mainboard: MSI Z97 GAMING 5
CPU: Intel i7 4790k @4.7 GHz / 1.252 V
CPU Fan: Cooler Master Hyper 212 EVO
Memory: G.SKILL RipjawsX 2×8 GB 1600MHz DDR3 @ 2000MHz / 10-12-13-25-208-1T
Graphics Card(s): 3x AMD HD 7950 (1x ASUS DirectCU II, 2x Sapphire Dual-X) @ 800/1250 MHz
SSD: Samsung EVO 840 250 GB
HDD: 1x 2 TB Western Digital Caviar Black, 1x 2 TB Seagate Barracuda, 3x 1 TB Seagate Barracuda
PSU: Xigmatek Vector S 1050 W 80+
Case: Corsair Obsidian 750D Full Tower ATX
Additional Fans: 3x Corsair AF140, 7x Corsair AF120
2. Benchmarking Tools and Details
Temperatures were measured with OCCT and Open Hardware Monitor. The room temperature was 23°C±1.
Benchmarking tools: WinRAR, AIDA64 GPGPU, PerformanceTest CPU, PerformanceTest 2D Graphics, Cinebench CPU, Final Fantasy XIV Online: Heavensward, and GTA V with two different settings (1440p and 1080p).
For the sake of consistency, all the benchmarks were done three times for each power management configuration and the mean results were calculated and are presented. All the tests were performed in a freshly booted, Aero-disabled Windows 7 which was clear of any unnecessary background processes. All benchmarks were installed on the SSD in order to eliminate any lags that could’ve caused by the hard drives.
The two power management configurations:
a) Power Management ON:
Intel Turbo Boost: Enabled
Enhanced Turbo Boost: Enabled
Clock Frequency: 4.7 GHz (With Automatic Multiplier Regulation)
Voltage: 0.1 – 1.27 V (Automatic, System Controlled)
b) Power Management OFF:
Intel Turbo boost: Disabled
Enhanced Turbo Boost: Disabled
Clock Frequency: 4.7 GHz (Constant)
Voltage: 1.252 V (Constant)
Let’s see if Intel’s power management technologies hinder the performance of their processors!
1. Idle Temperatures
Idle temperatures were measured with OCCT and Open Hardware Monitor after the system was booted up and 15 minutes have passed. The load temperatures are not included in the tests simply because they don’t change between the two configurations.
As expected, the i7 4790k runs a lot cooler when the power management settings are enabled. There was a mean difference of 12°C between two configurations. While the idle temperatures without the power management features seem a tad high, they’re certainly fit for daily usage and don’t pose a threat.
2. WinRAR (60 Seconds)
I ran the built-in benchmark of WinRAR 64-bit for 60 seconds and took the average resulting speed in terms of KB/s.
While the processor does perform better with the power management features off, it certainly doesn’t create a serious difference for daily usage.
3. AIDA64 GPGPU
Although this benchmark was designed to measure the raw performance of graphics cards, it was used because the CPU always has some impact on the results.
a) AIDA64 GPGPU – 3x AMD HD 7950 (Mean Score)
b) AIDA64 GPGPU – Internal Graphics Adapter (Intel HD 4600)
c) AIDA64 GPGPU – x64 CPU (Intel i7 4790k)
In all three of the tests, the i7 4790k without the power management features continues to be ahead by a tiny bit.
a) PerformanceTest CPU Score
b) PerformanceTest 2D Graphics Score
The differences continuing to be this low seems to reinforce my belief that these tests won’t show any actual difference.
Two power management configurations draw their swords again, and the i7 4790k with disabled power management settings manages to win by “1” points yet again.
Let’s get to what’s actually important: The real-world benchmarks! I couldn’t have chosen anything better for this than the CPU-hungry Final Fantasy XIV Online and GTA V, the game who forced people to buy new hardware the most in 2015. Shall we begin?
6. Final Fantasy XIV Online: Heavensward
Resolution: 1440p (2560×1440)
Wet Surfaces: Enabled
Occlusion Culling: Disabled
LoD Streaming: Disabled
Real-time Reflections: High
Edge Smoothing (Anti-aliasing): FXAA
Transparent Lighting Quality: High
Grass Quality: High
Background Tessellation: Standard Quality
Water Tessellation: Standard Quality
Shadows (Self): Display
Shadows (Other NPCs): Display
LoD (Shadows): Disabled
Shadow Resolution: High – 2048p
Shadow Cascading: Best
Shadow Softening: Strong
Texture Filtering: Anisotropic
Anisotropic Filtering: x16
Movement Physics (Self): Full
Movement Physics (Other NPCs): Full
Limb Darkening: Enabled
Radial Blur: Enabled
Screen Space Ambient Occlusion: Strong
Depth of Field: Enabled
7. GTA V
I wanted measure GTA V, a game which is famous for making computers sweat, in two different set of options, namely 1440p and 1080p. The settings that both the presets share are listed below and the differing settings are listed separately in their sections.
Population Density: 10/10
Population Variety: 10/10
Distance Scaling: 10/10
Texture Quality: High
Shader Quality: High
Shadow Quality: High
Reflection Quality: Ultra
Reflection MSAA: x8
Water Quality: High
Particle Quality: High
Grass Quality: Ultra
Soft Shadows: AMD CHS
Post FX Quality: Ultra
Depth of Field: Enabled
Anisotropic Filtering: x16
Tessellation Quality: High
Advanced graphics settings:
Long Shadows: Enabled
High Detail Streaming While Flying: Enabled
Extended Shadow Distance: 10/10
a) GTA V 1440p Preset
Resolution: 1440p (2560×1440)
Advanced graphics settings:
Extended Distance Scaling: 4/10
b) GTA V 1080p Preset
Resolution: 1080p (1920×1080)
Advanced graphics settings:
Extended Distance Scaling: 7/10
At last, some difference! As you can see in the graphs, there’s a considerable amount of difference in lowest and average FPS rates, and a particular one in the maximum FPS rate especially in the 1080p preset.
But let’s not forget that the actual reason the gap is so wide here is the system itself is capable of high frame rates as seen above. So as one gets lower and lower performance in GTA V, the gap’s going to get tighter and tighter.
E. Conclusion and Thoughts
Well… Is the difference worth running your CPU 12°C hotter when it’s idle? That’s the question. Unfortunately, I’m in no position to tell you exactly what to do here. But personally, yes! Especially if you’re not using a budget CPU cooler like me -the Cooler Master Hyper 212 EVO-, you have no reason to enable power management settings, because your system’s already going to be running cooler than mine. You wouldn’t have gotten a “k” series CPU if you were that worried about your bills and power consumption, would you now?
Until another adventurous test or comparison, so long.
6 thoughts to “Overclocking and Intel’s Power Management Settings (EIST, C-States, Turbo Boost)”
I have the same motherboard. Using i5 4690k and an EVGA Supernova G2 850W PSU. Overclocked the CPU to 4.6GHz and the Uncore clock to 4.4GHz. I have EIST enabled. The rest all disabled. If I try to enable C-states, I get a system lock after some time. No BSODs. It will freeze. Sometimes the reset button doesn’t even work and I have to hard shut down. With it disabled, no problems. I am thinking since with C-states enabled, the voltage fluctuates, it is not providing enough voltage or not switching to higher voltage in time when the clock speed changes. Did you have similar issues?
do you have the same issue when you’re on stock clocks?
With stock clocks and C-states enabled I don’t have any issues either. It happens mostly when there isn’t too much load. Generally when doing light tasks such as browsing. That’s why I think the problem is that the voltage doesn’t change quickly enough when clock speed changes. Currently the vcore is set to 1.25V in the BIOS. Will try at 1.265. C-states are not a must for me. The fixed voltage does not increase temperature but would like to know what could be the cause.
MAK, I’ve seen this also, when my i7-4790K was in my Dell XPS 8700, most of the time, all was well, then when what seemed to be an idle period, the mouse & whole screen would freeze, no BSOD. No errors other than ‘Windows was not shut down properly’.
How the heck is one supposed to diagnose based upon that? The tool used to collect & analyze any dump files couldn’t find anything, because it wasn’t there.
So I took it at the time, that it was freezing because the XPS 8700 MB has only one 4 pin CPU power socket, whereas aftermarket boards has two. Since being in this ASUS Z97-PRO Gamer, no more freezes. However, once I set all cores to 44x, my benchmark numbers on the Intel HWBOT site took a nosedive. At the same time, User Benchmark rates my PC at all time highs, two categories are very deep in UFO territory & the desktop rating is just a couple of points from it at 99% (Nuclear Submarine). Note that I haven’t manually messed with voltages nor C-State settings.
Therefore, I have two sets of benchmarks telling me two different stories. Maybe I need to start over, used the ASUS tuning Wizard to auto select a profile, based on my cooler & type of usage. 12% CPU increase & 2% DRAM up. NO freezing, all is OK, just the HWBOT scores (which may or may not mean anything) are in the lower 900’s, when using the simple ASUS Optimal profile was up in the 980 range. Yet at that time, the Desktop score of User Benchmark was only 95%
I believe that HWBOT can be VERY addictive & hard on the hardware over time if ran a lot, pusing temps to the 70C mark, even with the $79 CPU cooler purchased. Maybe I need to rely more on a different benchmark, even CPU-Z shows that my CPU is edging the i7-6700K narrowly, however that the PassMark rating also, just 200 points apart in favor of the i7-4790K.
Intel i7-4790K Quad Core CPU
32GB GSkill Aegis RAM (two sets of 16GB) PCE-12800, 11-11-11-28 (2T Command Rate)
EVGA GeForce GTX 1070 FTW 8GB GDDR5
ASUS Z97-PRO Gamer MB
512GB Samsung 950 PRO NVMe SSD (PCIe 3.0 x4)
EVGA G2 Supernova 650W PSU (plenty for a single GPU system)
Seagate 1TB ST1000DM003-1CH162 (SATA-3, 64MB cache, 7200 rpm)
Noctua NH-D15S CPU cooler (at first glance, a huge cooler)
Fractal Design Define R5 Case (very heavy, tons of open space, makes the Noctua look small)
I believe component wise am OK, just need to learn more about overclocking to meet my needs. Not hardcore, rather a stable build to show on forums, am not a gamer, though do run virtual machines.
Probably a little late, but I guess your PSU has problems specifically with the c6 state. Older PSUs might not support the c6 state and cause the system to crash when entering that state. Look into your bios and see if you can disable the c6 state only.
Something I would like to see in addition to what you have tested, just to cover all the bases, is whether setting windows power management to “high performance” results in the same performance benefits that turning off the energy saving features in the bios does.
As I understand it, “high performance” mode does essentially the same thing, and could be a quick toggle for getting the same performance benefits you’ve shown here without restarting and entering the BIOS.
Excellent testing you’ve done here. Delving into things that other benchmarking and hardware sites don’t really examine at all. I did similar testing with windows system timers (hpet, tsc-acpi, etc) and found that having hpet enabled had worse performance than the default windows timer (tsc-acpi), which was surprising given all the articles that tell you that fully turning on hpet in windows gives you a performance boost. My testing showed the opposite: More dpc latency, input lag, and worse CPU performance, although GPU performance/fps was just a tad higher. Definitely better leaving it off (in bios and windows), imo. I didn’t test it in nearly as many different areas as you did in this article though, just a few games and cinebench.. so it might be a good follow-up test to this article seeing as you like to dig into intricacies like this.