Overclocking: Can Sandy Bridge-E Be Made More Efficient?

Overclocking: Procedure, Details, And Log

Let's get down to business. Our first question is whether we should be enabling Turbo Boost or not while we overclock manually. Although power management and automatic tuning mechanisms are great behind-the-scenes features for folks who don't like to mess around under the hood, they generally aren't idea for generating clear, scientific results.

Nevertheless, Turbo Boost turns out to be one of the best ways to fine-tune performance at any given load level. So long as you're able to manually specify clock rate based on however many cores are active, you can control how much voltage is required to maintain a stable system in any given workload.

Overclocking With Second-Gen Turbo Boost

And so, we made the decision to overclock with Turbo Boost enabled. In fact, per Intel's guidance to motherboard vendors, overclocking on platforms wielding unlocked processors is achieved through Turbo Boost ratios. We increased our maximum TDP and maximum current limits to create additional headroom, and went about increasing the multipliers.

We also enabled power-saving features wherever possible. In practice, you want to cut back on consumption as much as you can during idle periods, even as you push performance up under load.

The multiplier combinations we used are listed below. Turbo Boost is able to specify multiplier settings based on processor activity. Using a base clock of 100 MHz and a default multiplier of 33x, the following combinations are then allowed to modify that 3.3 GHz product.

Overclocking Table

Swipe to scroll horizontally
Header Cell - Column 0 Stock ConfigurationOverclock #1 (Mild)Overclock #2 (Mild)Overclock #3 (Moderate)
Row 0 - Cell 0 33x33x33x33x
Five or Six Cores Active36x38x40x42x
Three or Four Cores Active37x39x41x43x
One or Two Cores Active37x41x43x45x
Idle Voltage (Measured in AIDA)0.846 V0.846 V0.846 V0.846 V
Load Voltage (Measured in AIDA)1.241 V1.286 V1.316 V1.336 V
Firmware Voltage SettingDynamicDynamicDynamicOffset 40 mV
Stable?YesYesYesYes
Swipe to scroll horizontally
Header Cell - Column 0 Overclock #4 (Moderate)Overclock #5 (Moderate)Overclock #6 (Aggressive)Overclock #7 (Aggressive)Overclock #8 (Aggressive)
Row 0 - Cell 0 33x33x33x33x33x
Five or Six Cores Active43x44x45x46x47x
Three or Four Cores Active44x45x46x47x47x
One or Two Cores Active45x46x47x47x47x
Idle Voltage (Measured in AIDA)0.846 V0.846 V0.846 V0.846 V0.846 V
Load Voltage (Measured in AIDA)1.321 V1.326 V1.331 V1.331 V1.336 V
Firmware Voltage SettingOffset 40 mV1.34 V1.355 V1.37 V1.385 V
Stable?YesYesYes Mild V-DroopYes Low V-Droop PLL Override: Enabled Idle State: High Performance

Voltage Overview

The following table shows which voltages were used for the other core components of the test system.

Swipe to scroll horizontally
Processor Core (V)Dynamic
Memory (V)1.65 V
System Agent Voltage (V)1.05 V
Processor I/O (V)1.05 V
Processor PLL (V)1.8 V
PCH Core (V)1.1 V

Overclocking Log Book

Turbo Boost Multiplier Settings: 42-43-45 and 43-44-45

The platform froze at 4.5 GHz using dynamic power in single-core mode. After applying a 40 mV offset, the system stabilized. The same problem surfaced when we tried the next-highest multiplier ratio, though. We prioritized multi-core performance and reduced the third ratio setting from two bins to one. This allowed stable operation without further changes to the test configuration.

Turbo Multiple Settings 44-45-46

Changing the offset to 120 mV allowed the machine to boot; only a manual voltage adjustment helped. But this allowed us to reach the speed at which energy efficiency fell off sharply.

Turbo Multiple Settings 45-46-47

This case also didn't allow for simple voltage increases. We had to switch from High V-Droop to Mid V-Droop, which again led to increased power consumption. This is a good way to dial in more stable performance, but not to improve efficiency.

Turbo Multiple Settings 46-47-47 and 47-47-47

We weren't able to stably reach 4.8 GHz, although all the cores eventually reached the 4.7 GHz mark. For this step, we had to disable the rest of the energy-saving functions, which accounts for the really bad efficiency results, as we had to:

  • enable Low V-Droop
  • set PLL Override to High
  • set Processor Idle State to “High Performance“

After these adjustments, the test system consistently ran at 4.7 GHz. Interestingly, the single-threaded performance was influenced by raising the Turbo Boost setting for 3 + 4 cores, due to non-optimal scheduling in Windows.

  • mayankleoboy1
    This article appeared on tomshardware.de weeks before.
    Reply
  • Combat Wombat
    Good to know!
    Reply
  • Yargnit
    What about trying to under-volt it at slight under-clocks to slight-overclocks. How much room is there to reduce it's stock voltage to gain better efficiency?
    Reply
  • billj214
    Was there an efficiency chart made for the Core i7 2600k or 2700k?
    Nice to know Intel doesn't just set the stock clock speed for just performance!
    Reply
  • Marcus52
    And then there's the Core i7-3820, which only sports four cores, but operates at a base clock rate of 3.6 GHz. Although this less-complex chip could probably hit higher Turbo Boost frequencies, Intel limits it to 3.9 GHz to keep it from outshining the top-end Core i7-3960X in single-threaded tasks.

    Did someone at Intel tell you that was the reason for a lower Turbo Boost limit, or did you just assume it?

    I think we should be careful of this kind of guess at another person's, or company's, reasoning. There could be some other cause for the limit - for example, they will obviously sell it for a lower price, so wouldn't a possible reason be they have looser binning specs to allow for chips that wouldn't make it under more strenuous tests through? (Remember, Intel, or any CPU manufacturer, doesn't warrant the product based on what it can be pushed to, and is generally going to provide it at a clock rate they feel is safe over time to guarantee.)

    I'm certainly not saying it is a bad assumption, what you said makes sense to me, but I do think there are enough other reasonable possibilities that I wouldn't have stated it as a fact unless I knew it to be.

    ;)
    Reply
  • Marcus52
    Thanks for the analysis!

    I do think articles like this are very important; those of us who overclock, especially when we turn off all the power-saving features in hopes of reaching that max stable a CPU can do, should be aware of how much money we are spending if we keep said OC. It's more than just the high end cooling solution.

    The people that bash higher capacity PSUs could also stand to learn a thing or two, here. An overclocked CPU can require a huge amount of peak power over and above what a stock CPU needs (349W measured here). An overclocked Sandy Bridge-E and an overclocked GTX 580 could require a peak power of 650W just considering those 2 components!

    A Kill A Watt or similar device is a great way to measure how much you actually spend a month operating your computer. You might be surprised.

    ;)
    Reply
  • lahawzel
    "Intel Core i7-3690X Extreme Edition"

    Tom's Parallel Universe Hardware.
    Reply
  • giovanni86
    Just a thought, so at 4.7Ghz the performance increase was only 16%? For being such a High overclock i was hoping for more then that. You guys literally upped the bar from stock clock to the OC clock by 1.4ghz, seems like a small increase in performance if you look at the amount of watts it takes.. Well at least its good 2 know my future billing of electricity will sure be expensive.. =P
    Reply
  • Naxos
    Does anyone spending 600-1k$ on a cpu really care about efficiency??
    Reply
  • cangelini
    Marcus52Did someone at Intel tell you that was the reason for a lower Turbo Boost limit, or did you just assume it?I think we should be careful of this kind of guess at another person's, or company's, reasoning. There could be some other cause for the limit - for example, they will obviously sell it for a lower price, so wouldn't a possible reason be they have looser binning specs to allow for chips that wouldn't make it under more strenuous tests through? (Remember, Intel, or any CPU manufacturer, doesn't warrant the product based on what it can be pushed to, and is generally going to provide it at a clock rate they feel is safe over time to guarantee.)I'm certainly not saying it is a bad assumption, what you said makes sense to me, but I do think there are enough other reasonable possibilities that I wouldn't have stated it as a fact unless I knew it to be.Hence the "probably." Of course, we don't know for sure, nor would Intel ever admit as such, but it's an educated guess nonetheless. =)
    Reply