Power Savings and Voltage/Frequency Drop
Power savings consist of dropping the frequency, voltage, or both at idle or less-than-maximum load conditions to help save power and reduce temperatures. To ensure the frequency drops at idle, you need to do two things.
1. Make sure EIST and C-States are enabled in the UEFI. In the motherboard section of this guide, I have highlighted where the power savings options are located for each board.
2. Make sure you do not have "High Performance" chosen as your power plan in the "Power Options" menu in Windows Control Panel. Conversely, you can set "High Performance" to engage the maximum multiplier at all times on many boards even with EIST enabled in the UEFI.
If you want the voltage to drop depending on the CPU frequency (not load), then you will need to use Auto, Offset, or Adaptive mode for the VCore (and, of course, enable frequency power savings).
I wanted to see the impact of dropping the frequency without dropping the VCore (frequency power savings engaged, but VCore drop not engaged), dropping both frequency and VCore, and not dropping either. Maximum power, temperature, and voltage are equal, which is what I expected. Idle power is really what power savings is about, and there is a noticeable drop in CPU power draw when utilizing frequency drop and voltage drop, even with frequency drop alone you save about 10 Watts.
The temperature does not drop as readily as power. Without any power savings maximum idle temperature are 27C, and dropping frequency alone yielded 25C, while dropping both yielded 21C. Compared to the 35-40C increase from idle to load these gains are small. I should also note that dropping frequency and voltage together should help in terms of the life of the processor, but your processor will probably become obsolete before it dies considering how many CPUs from previous generations are still going strong.
Serial Voltage Identification
Technically, you can use CPU Auto VCore (SVID) and LLC to control your VCore instead of manually setting it, but your SVID results will vary from mine, and are different from CPU to CPU.
I wanted to see how high the VCore went on its own when I increased CPU frequency. The upper limit on SVID is 1.52v, which is very high on air and watercooling. Please realize that each CPU will pick a much different voltage at different frequency levels. In fact, the voltage the CPU reports for itself can be used to find better CPUs (usually lower VCore at stock frequency is a better chip). You can see the large drop from idle to load voltages. Also, the CPU is quite optimized at 4GHz (this is a 6700K).
CPU Temperature and Power Scaling
Now let's take a second and look at how frequency, temperature, and CPU power consumption are related. On the left, I kept the voltage fixed at 1.35v and increased frequency to check out power and temperature, and on the right I fixed the frequency and changed VCore in small increments and measured power and temperature. For those of you wondering how I measured power, I used an inline hall effect sensor on the 8-pin CPU power plug (all CPU power goes through it), and logged the maximum power draw and multiplied it by the measured 12v voltage. I usually see a much steeper exponential slope, but Skylake looks much more linear than previous platform results which is a good thing for overclocking. I think that this might be because of reduced leakage, which is also a good thing.
How much more VCore is needed for faster Cache?
I tested Cache frequency at different levels (1:1 with the CPU frequency, fixed at 40x, fixed at 45x) and then scaled the CPU frequency. I then recorded the lowest voltage required to not crash in Windows with CPU-Z open; I corrupted Windows 10 more than once by doing this. A few trends do pop up. For starters, the Uncore/Cache/Ring seems to have its own required voltage level that doesn't change depending on the CPU core frequency, but it's small.
On average I required 0.03-0.05v more to run the Cache at 1:1 with CPU frequency. As we saw earlier the benefits of running faster cache are minimal, but higher VCore produces higher temperatures. I would say that you do not need to keep the Cache ratio at 1:1 with the CPU for 24/7 use, instead try to keep it at least 500MHz away, and only closer once you find your optimal CPU frequency and if you can afford a few extra degrees.
Maximizing DDR4 Frequency with VCCIO and VCCSA
By default, the VCCSA is higher than the VCCIO, but both are equally important. I took a closer look and tried different CPUs to see what I had to change, and I came across this one CPU, which showed the largest differences between VCCIO and VCCSA required voltage levels for highest 4GHz XMP overclock.
For this test, I did maximum CPU-Z validations using the same timings and DRAM voltage (1.375v), but different VCCSA and VCCIO. The results at the top of the list are the voltages that the "Auto-Rules" on the motherboard provided. I found that lowering the VCCSA too low causes a failure to even boot, but increasing it too high causes a negative effect to memory overclocking. I found that VCCIO just showed steady gains as it was increased, but these gains were very small for very large amounts of voltage. I would keep the VCCSA below 1.3v but above 1.2v and thinker with VCCIO to see if it helps your specific CPU sample. However, for normal usage, I would keep both of them at or below 1.25v.
PRICING: You can find products similar to this one for sale below.
United States: Find other tech and computer products like this over at Amazon's website.
United Kingdom: Find other tech and computer products like this over at Amazon UK's website.
Canada: Find other tech and computer products like this over at Amazon Canada's website.
- Page 1 [Introduction to Skylake Overclocking]
- Page 2 [Disclaimer and Before You Begin Overclocking]
- Page 3 [Overclocking Flow Chart, Non-K SKUs, BCLK, and Multipliers]
- Page 4 [BCLK, Core, Cache, Memory, and FCLK Scaling]
- Page 5 [Skylake Overclocking Voltages]
- Page 6 [Power Savings and Voltage/Power Analysis]
- Page 7 [The Durability of Intel's 14nm Node]
- Page 8 [Skylake Memory Overclocking: Corsair and Frequency Scaling]
- Page 9 [Skylake Memory Overclocking: G.Skill and Memory Timings]
- Page 10 [ASRock Z170 Overclocking: Z170 OC Formula]
- Page 11 [ASUS Z170 Overclocking: Maximus VIII Extreme]
- Page 12 [EVGA Z170 Overclocking: Z170 Classified 4-Way]
- Page 13 [GIGABYTE Z170 Overclocking: Z170X-SOC Force]
- Page 14 [MSI Z170 Overclocking: Z170A XPOWER GAMING TITANIUM EDITION]
- Page 15 [Stability Testing, Delidding, Crashing, and Throttling]
- We at TweakTown openly invite the companies who provide us with review samples / who are mentioned or discussed to express their opinion of our content. If any company representative wishes to respond, we will publish the response here.
Latest News Posts
- AMD trademarks Kyzen, Aragon, Promethean & CoreAmp
- Oukitel announces K10000 Pro with 10,000 mAh battery
- ZOTAC GeForce GTX 1080 Ti Mini is super-small, powerful
- Sony to launch two new Xperia flagship devices this year
- MSI has the first graphics card with a USB Type-C port
- ADATA AI910 Lightning Card Reader Plus Review
- PlayerUnknown's Battlegrounds Benchmarked: CPU & GPU War
- NIC not working after BIOS corruption (ASUS X202E)
- Logitech Harmony Elite Advanced Remote Control Review
- C2750D4I won't get passed initializing 55
- Qualcomm fuels IoT growth by currently delivering more than 1 million chips a day into a wide range of connected applications
- Team Group announces theme for COMPUTEX 2017 showcase: go beyond the limit and reach for the top
- SAPPHIRE announces PULSE Radeon RX 560 graphics card
- ELITEGROUP computer to stand out at Computex for its smart campus deployment, robotic technology, education laptops, tablets, mini PC, and motherboards
- ADATA Shares a Symphony of Technology at Computex 2017