Intel Speed Shift Technology... MIA?

[SuperJaw]

unclewebb, the author of Throttlestop has just updated it to v8.20 which enables Speed Shift support for Skylake.

Some interesting info;

Nice, find. It would be nice to have an simple app or utility that only enabled SST. Throttlestop is a power tool with a lot of options. But it does seem to enable Throttlestop. My main rig is laptop these days (Dell 7510 with the 6820HQ cpu), so I like to conserve battery life where I can.

 

[Free42]

Just a small update:
it seems ASUS has started to integrate CPPC v2 into their firmware versions. Don't know exactly when they added this feature but firmware version 1801 for the Z170-A didn't have the option while the current version 3007 does. And as expected SST is now enabled in Windows 10 for theses boards if you enable the CPPC v2 interface.

And again as I said before: enabling SST didn't have any positive or negative impact on my system as you would expect for a desktop Skylake system.

 

[mikk]

They started to add this with the recent updates, 3xxx.

 

[wpcoe]

unclewebb, the author of Throttlestop has just updated it to v8.20 which enables Speed Shift support for Skylake.

Some interesting info;

Nice, find. It would be nice to have an simple app or utility that only enabled SST. Throttlestop is a power tool with a lot of options. But it does seem to enable Throttlestop. My main rig is laptop these days (Dell 7510 with the 6820HQ cpu), so I like to conserve battery life where I can.

I've been following this thread, digesting as much as I can although some of it is above my head. I just installed ThrottleStop 8.30 to see if enabling SST/HWP does anything discernible on my Skylake (i3-6320) system.

What do I select/deselect in ThrottleStop to enable SST? With the default settings, when I start ThrottleStop, SST is still indicated in red in HWiNFO64.

 

[Free42]

They started to add this with the recent updates, 3xxx.

I can confirm that this is a new feature introduced in the 30xx versions, even though all the revision history shows is "Improve system stability". As usual the changelog ASUS provides sucks ... It seems ASUS made quite some changes under the hood in v30xx as AMIBCP shows.

 

[Dave2150]

They started to add this with the recent updates, 3xxx.

V3xxx is now removed for most Asus Z170 boards, it had many bugs.

My G752VY laptop and Maximus Hero VIII system have both lacked SpeedStep since day 1, rather disappointing.

 

[mikk]

I can confirm that this is a new feature introduced in the 30xx versions, even though all the revision history shows is "Improve system stability". As usual the changelog ASUS provides sucks ... It seems ASUS made quite some changes under the hood in v30xx as AMIBCP shows.

Yes there are lots of changes. Changelog is completely useless as usual. There is a a new option called AVX Ratio for example. Also several of the onboard stuff has been updated: Sata ROM, VBIOS, Management Engine, EFI GOP, CPU Microcode.

 

[Free42]

V3xxx is now removed for most Asus Z170 boards, it had many bugs.

I've only tested the new firmware on a ASUS Z170-A board yet and didn't have any problems so far with v3007. ASUS also hasn't removed this version yet from their support page for this particular model. But I'm also not surprised that there're problems with other models since this version includes so many changes.

Edit:
After playing around a little with SST and Windows 10 (1511) I found out that Intel's CPPC drivers disable HW Autonomous selection, i.e. explicit values are set for bits 23:16 to the IA32_HWP_REQUEST MSR. Which means that not all control over the P-states is handed over to the CPU. I.e. manually enabling HWP via MSRs won't lead to the same result as having a MB firmware with actual CPPC support.

Furthermore Intel's energy settings under "Intel CPPC Energy Efficiency Settings" don't seem to have any effect on the IA32_HWP_CAPABILITIES or IA32_HWP_REQUEST MSRs. I guess those settings only influence how the Desired_Performance bits are set.

I also noticed that the Kernel-Processor-Power log shows some strange values (750% for MaximumPerformancePercent and 23% for MinimumPerformancePercent) with SST enabled.

Edit 2:
A speciality about CPPC and Windows 10 performance plans is worth noticing - selecting the "High Performance" power plan while SST is enabled may cause all CPU cores to permanently run at full clock speed. That's because this plan sets the "desired performance" (bits 23:16) to the value of "maximum performance" (bits 15:8), i.e. this is not an error and the desired behavior. But this also means most of the energy saving technologies of your Intel CPU will be effectively disabled if you select the high performance power plan.

 

[wingman04]

How can I tell if speed shift is working in Hwinfo64?

 

[Free42]

How can I tell if speed shift is working in Hwinfo64?

Speed Shift is enabled if the SST label inside the Features box in the System Summary window of HWiNFO64 is green.

I unfortunately can't directly upload screenshots here and I'm too lazy to use another image provider ...

 

[wingman04]

My SST in HWiNFO64 is red, I have windows 10 is there a way to turn it on so it is green?

 

[Free42]

My SST in HWiNFO64 is red, I have windows 10 is there a way to turn it on so it is green?

If your BIOS supports CPPC simply enable this feature in your BIOS settings. Intel's drivers will then automatically enable SST in Windows 10.

Otherwise you could manually enable SST via the IA32_PM_ENABLE MSR (0x700) - which btw. works under Windows 7 / 8 as well. If you don't know what that means then you should probably not do it. Setting wrong values in MSRs will at least cause a blue screen. You can find some information about setting the MSR to enable SST in this thread.

 

[wpcoe]

Sorry to beat a dead horse, but above it is mentioned that ThrottleStop enables SST. How? What settings need to be specified in ThrottleStop? Just running ThrottleStop doesn't enable SST, at least not on my i3-6320 system.

 

[coercitiv]

Sorry to beat a dead horse, but above it is mentioned that ThrottleStop enables SST. How? What settings need to be specified in ThrottleStop? Just running ThrottleStop doesn't enable SST, at least not on my i3-6320 system.

Is SST enabled at BIOS level? All Throttlestop does is make it work on any Windows version, but does not compensate for the required activation at BIOS level.

 

[CHADBOGA]

All this drama getting SST to work, just reinforces my belief that whenever a new "technology" is announced by Intel or whoever, the first iteration is either useless or doesn't work and maybe in 2 or 3 years time it will be sorted out.

 

[coercitiv]

All this drama getting SST to work, just reinforces my belief that whenever a new "technology" is announced by Intel or whoever, the first iteration is either useless or doesn't work and maybe in 2 or 3 years time it will be sorted out.

It's also primarily aimed at mobile products, desktop users mainly gain experience

 

[Free42]

All this drama getting SST to work, just reinforces my belief that whenever a new "technology" is announced by Intel or whoever, the first iteration is either useless or doesn't work and maybe in 2 or 3 years time it will be sorted out.

One could say SST mostly did serve Intel's and Microsoft's marketing purposes if you read all those online articles about Speed Shift boosting performance of Skylake CPUs exclusively on Windows 10. I hope this thread helped to demystify this feature and that most of what you read online about SST is false. Neither is SST exclusive to Windows 10 nor does it boost the performance of Skylake CPUs. It might have a positive impact on the energy efficiency of (mobile) CPUs though.

However SST is the next technical evolution of technologies like Speed Step. It theoretically allows better control over the CPU's P-States through the OS - thus enabling a fine grained QoS without having to deal with sth. like CPU frequencies. And the OS can even hand over the full control over P-States to the CPU itself which will then try to automatically detect the best setting for the current situation.

So I partially agree to your statement, even though I wouldn't call SST "useless" - but I understand it might seem so from a user's perspective.

 

[WhoBeDaPlaya]

It's also primarily aimed at mobile products, desktop users mainly gain experience

Just another thing to disable when overclocking

 

[coercitiv]

I hope this thread helped to demystify this feature and that most of what you read online about SST is false. Neither is SST exclusive to Windows 10 nor does it boost the performance of Skylake CPUs. It might have a positive impact on the energy efficiency of (mobile) CPUs though.

SST having a positive impact on energy efficiency of mobile CPU equals to SST bringing a tangible performance benefit for mobile CPUs. For mobile chips operating on battery power any edge in power consumption will translate in better performance at normalized energy usage as long as thermals are not an issue.

Intel marketed SST as a performance boost for mobile, not desktop.

 

[wingman04]

SST just ramps up the Clock speed faster for CPU load, it does not use or save more power when getting the task done quicker instead of longer.

 

[coercitiv]

SST just ramps up the Clock speed faster for CPU load, it does not use or save more power when getting the task done quicker instead of longer.

When your CPU spends more time in sleep states because it ramped up faster and then ramped down faster than the OS would be able to tell it to, what do you think happens to energy usage over time? Imagine your electric car is already plugged in when your neighbor's car is still waiting for the garage door to open.

 

[wingman04]

When your CPU spends more time in sleep states because it ramped up faster and then ramped down faster than the OS would be able to tell it to, what do you think happens to energy usage over time? Imagine your electric car is already plugged in when your neighbor's car is still waiting for the garage door to open.

You cant compare SST to garage door. The same amount of work load has to be done when speed shift is being used or speed step, then it will use the same power.

Intel has told us that Speed Shift does not impact battery life very much, one way or the other, so to verify this, I've run our light battery life test with the option disabled and enabled.

This task is likely one of the best case scenarios for Speed Shift. It consists of launching four web pages per minute, with plenty of idle time in between. Although Speed Shift seems to have a slight edge, it is very small and would fall within the margin of error on this test. Some tasks may see a slight improvement in efficiency, and others may see a slight regression, but Speed Shift is less of a power savings tool than other pieces of Skylake. Looking at it another way, if, for example, the XPS 13 with Skylake was to get 15 hours of battery life, Speed Shift would only change the result by about 7 minutes. Responsiveness increases, but net power use remains about the same. http://www.anandtech.com/show/9751/examining-intel-skylake-speed-shift-more-responsive-processors

 

[coercitiv]

The same amount of work load has to be done when speed shift is being used or speed step, then it will use the same power.

This is fundamentally wrong: in the two cases we have, CPU operates at different average frequency and thus uses different amounts of energy. With SST enabled the CPU ramps up to max clocks while in the other case it may not even reach max clocks on task completion. Considering higher operating frequencies use exponentially more power, during task completion the SST enabled CPU might use a lot more energy to get the work done. However, this difference if offset by the savings done once CPU job is done and it enters sleep mode. So yes, I can compare SST to garage door. How do you think your car arrived first, if not by accelerating and breaking faster, thus using more energy to get from point A to point B.

The fact that SST is tuned with little power savings in mind in order to maximize responsiveness does not mean it does not offer both. Tune the OS/SST interaction at a different ratio for perf/energy usage and data changes. In fact even using different CPUs such as i3 or i5 might skew the perf/energy gains, since the lower top frequencies will limit the energy burn, while the initial jump from low frequencies to base frequencies still provides a tangible performance benefit. The Core i7-6600U might have been the perfect vehicle to showcase the performance gains SST has to offer, but not necessarily the one with the best power savings as well.

Intel claimed slight power savings. The Anandtech test fails to measure that while being a test tailored for consumers (which it should be), not for isolating CPU power usage. I don't see how this data invalidates Intel's claim, other than putting a bigger emphasis on slight.

 

[wingman04]

This is fundamentally wrong: in the two cases we have, CPU operates at different average frequency and thus uses different amounts of energy. With SST enabled the CPU ramps up to max clocks while in the other case it may not even reach max clocks on task completion. Considering higher operating frequencies use exponentially more power, during task completion the SST enabled CPU might use a lot more energy to get the work done. However, this difference if offset by the savings done once CPU job is done and it enters sleep mode. So yes, I can compare SST to garage door. How do you think your car arrived first, if not by accelerating and breaking faster, thus using more energy to get from point A to point B.

The fact that SST is tuned with little power savings in mind in order to maximize responsiveness does not mean it does not offer both. Tune the OS/SST interaction at a different ratio for perf/energy usage and data changes. In fact even using different CPUs such as i3 or i5 might skew the perf/energy gains, since the lower top frequencies will limit the energy burn, while the initial jump from low frequencies to base frequencies still provides a tangible performance benefit. The Core i7-6600U might have been the perfect vehicle to showcase the performance gains SST has to offer, but not necessarily the one with the best power savings as well.

Intel claimed slight power savings. The Anandtech test fails to measure that while being a test tailored for consumers (which it should be), not for isolating CPU power usage. I don't see how this data invalidates Intel's claim, other than putting a bigger emphasis on slight.

Speed step and Speed shift work with CPU load from idle to peak speed to get the work done faster, it does not mater if speed of the aplicatiion is on longer or shorter in time to get the same amount of work done, in a longer or shorter amount of time, both use the same amount of power in the end becase it is load + time of load.

Intel claimed slight power savings.

Post a link where Intel said there is any power savings. Anandtech said that Intel stated Speed Shift does not impact battery life very much, one way or the other

 

[coercitiv]

it does not mater if speed of the aplicatiion is on longer or shorter in time to get the same amount of work done, in a longer or shorter amount of time, both use the same amount of power in the end becase it is load + time of load.

Again, this is fundamentally wrong. The relation between frequency and power usage is not linear (due to voltage increase), but the one between frequency and task time is (or at least it's near linear). Therefore the same CPU operating at two different average frequencies will use different energy amounts to complete the task. The only reason the race to idle scheme works to generate performance with little or no additional energy cost is the energy savings obtained by putting the first CPU package into sleep mode and effectively shutting down chip parts, while the lower clocked system operates for a longer time and potentially wastes energy through static power usage.

Think about it, if what you said was true and systems would use the same amount of energy for task completion no matter the average operating frequency, we could simply briefly clock mobile CPUs to 4Ghz+ and have them finish tasks sooner while using the same energy. That does not happen because there's always a threshold where the exponential energy increase from high frequencies can no longer be compensated by the energy savings of turning the chip off after task completion.

Post a link where Intel said there is any power savings. Anandtech said that Intel stated Speed Shift does not impact battery life very much, one way or the other

From the review you provided:

Slight overall power reduction