Intel Speed Shift Technology... MIA?

[ELopes580]

So I figured I would add to this thread by mentioning that Asus has plans to add Speed Shift support in an up coming firmware for the ROG Maximus VIII Gene and possibly others but I am not sure.

I had been contacting Asus since late June for a few weeks asking for them to enable it. At first they pawned it off at first saying it was a mobile only feature. Another time saying it was supported and it was up to Intel and Microsoft to enable it; telling me to contact Intel. So I did, but figured it needed to be UEFI support just like VT-x and SteepStep/EIST is; sure enough that is what Intel stated to me and it was up to the board manufactures to enable it, back to Asus. (I'm not one to give up!)

I finally got ahold of a livechat rep who fully understood the issue and stated it shouldn't be to hard as it was enable in their laptops. He said he would escalate it to their R&D dept and see what he can find out. Days later, I am told it would be enabled in a future UEFI update, the current version of ROG UEFI (1902) doesn't enable it. so I am hoping the next one will.

Hopefully it will be a full across the board support update for Asus owners.

 

[TheRyuu]

I believe bios support is no longer necessary with the release of W10 AU. The anniversary update should contain the proper code to make use of the skylake features. IIRC the bios stuff was just to hack around the fact the operating system didn't support it. If you have a Skylake CPU it should be using the new technology if you're on the AU.

 

[ELopes580]

This is news to me as It should have been supported in the Win 10 Nov 2015 v.1511 update. I was holding off upgrading to the v.1607 update but I am installing it now and will report back in a bit.


EDIT: the AU v.1607 still has it disabled. So here's to hoping on that UEFI support soon. I guess I could mess with the MSR but why bother and I think someone was claiming stability issues doing that way. We should be keeping the motherboard makers accountable to have things working and supported. Especially something like this that can enhance responsiveness to multitasking and workload demand.

 

[Free42]

[...]Another time saying it was supported and it was up to Intel and Microsoft to enable it; telling me to contact Intel. So I did, but figured it needed to be UEFI support just like VT-x and SteepStep/EIST is;

It is true that it's actually the responsibility of the OS and its CPU driver (i.e. MS / Intel) to enable Skylake's Speed Shift / HWP and not the UEFI firmware. So no, there's no firmware support needed for this specific feature to be enabled. That's why HWP just works fine under Linux and can be manually enabled by modifying the MSRs.

[...]sure enough that is what Intel stated to me and it was up to the board manufactures to enable it, back to Asus. (I'm not one to give up!)

Thanks to the screenshot chinobino posted I was able to identify what MSI did and why this firmware seems to cause HWP to be enabled by Win 10. The setting we see there doesn't actually directly enable HWP but makes the CPPC extension available as defined by the ACPI 5.0 specification (section 8.4.5).
It seems the relevant ACPI / CPU drivers of Win 10 enable HWP on Skylake systems if this ACPI extension is made available to the OS by the firmware.

Which of course means that MB manufacturers could indirectly enable Skylake's HWP for Win 10 users by integrating / implementing the CPPC extension. But that's actually sth. beyond simply supporting HWP, CPPC is not even specific to Intel's Speed Shift technology.

Especially something like this that can enhance responsiveness to multitasking and workload demand.

As said before, you should not expect HWP to have any positive impact on your Win 10 system performance or responsiveness if you're using a desktop Skylake platform. But nevertheless I agree that it would be nice to have an (easy) option to enable this feature, no matter if its use is limited or not.

Edit: forgot to mention that you should ask ASUS & Co for CPPC v2 support rather than HWP support. This should make it a lot clearer to them what they actually could do on their side.

 

[ELopes580]

Thanks for the insightful information. I guess I'll just have to wait and see what Asus does. Hopefully they are aware of the update v5.0 to the ACPI spec and will implement it. That was definitely an interesting read. Hopefully they will have it fully supported for future desktop platforms from the get-go.

 

[chinobino]

I just want to add something else interesting that I found out.

I have a 2nd system with an i3 6320 and Windows 10 x64 1607 (Build 14393) overclocked using a non-k OC BIOS to take it up to 4.6 GHz.

I have HWinfo monitoring the CPU socket temperature and with SST enabled the idle CPU temperature drops to ~32°C which is close to my case's ambient temperature, even at v1.35 V.

With SST disabled the CPU temp hovers around 40°C to 50°C when idle.

Also, SST is NOT automatically enabled with Windows Anniversary Update - you still need BIOS support or to switch the MSR manually.

 

[SuperJaw]

A while ago someone requested that I share my Gigabyte SST enabled BIOS. Here it is... It would be awesome to get a newer F6 Bios modded with SST. Gigabyte says the don't have plans to add it.

https://www.dropbox.com/s/sj4gm063reki2hd/Z170XUD3GA.zip?dl=0

 

[chinobino]

A while ago someone requested that I share my Gigabyte SST enabled BIOS. Here it is... It would be awesome to get a newer F6 Bios modded with SST. Gigabyte says the don't have plans to add it.

https://www.dropbox.com/s/sj4gm063reki2hd/Z170XUD3GA.zip?dl=0

I requested it - thanks for uploading it.

It is different from the F5g Beta BIOS that is publicly available at Tweaktown.

I will see what I can do...

[Edit] 17 modules are different, will need some time to examine the changes properly.

 

[chinobino]

Well it was surprisingly easy to enable SST (HWP) in the BIOS, although there is no option to turn it off once enabled, so I can only use Windows 10 (Build 10586 or later) or Linux (no big deal for me).

This is the setting I modified in my BIOS using AMIBCP v5.02.0023 (picture shows SuperJaw's custom made Gigabyte GA-Z170X-UD3 BIOS F5g);

Setup --> Main --> (Expand folder with no name)

(0712) HardWare P states (HWP) set Optimal to "Enabled" then save the BIOS.

I have a SPI programmer to recover in the case of a corrupted BIOS, so I had no hesitation to flash the modded BIOS - if you don't have one then you flash the modified BIOS at your own risk!

 

[wpcoe]

I'm confused. (What's new?) From this thread, I gleaned that Win10 support for Speed Shift was non-existent, and BIOS support was needed for Speed Shift to be active.

But from http://www.anandtech.com/show/10610...-january?_ga=1.65931630.1382271800.1470761504 :

The first iteration of Speed Shift reduced the time for the CPU to hit peak frequencies from ~100 milliseconds down to around 30. The only limitation was the OS driver, which is now a part of Windows 10 and comes by default. We extensively tested the effects of the first iteration of Speed Shift at launch.

So, does Win10 now actively support Speed Shift by default?

 

[chinobino]

So, does Win10 now actively support Speed Shift by default?

HWP or Hardware Performance States (AKA SST or SpeedShift Technology) is only supported in Windows 10 1511 (Build 10586) and later.

It is not enabled unless the PC/device has HWP/SST enabled in the UEFI.

Sometimes there is no visible option to turn HWP/SST on/off in the UEFI, such as with many MSI Z170 motherboards (although it is enabled).

Asus & Gigabyte Z170 motherboards seem to have HWP/SST disabled by default with no UEFI option to turn it on.

 

[ShintaiDK]

Quite odd to see the lack of Speedshift support still from some mobo makers. I am just happy both my NUC and MSI board got it right away.

But I guess its a prioritization from those board makers.

 

[Free42]

HWP or Hardware Performance States (AKA SST or SpeedShift Technology) is only supported in Windows 10 1511 (Build 10586) and later.

HWP is not exclusive to Win 10. Intel CPPC drivers are available for Windows 8 / 8.1 as well, at least for some systems (like some Lenovo Thinkpads for example). The standard (non CPPC) Intel / MS processor drivers for desktop Skylakes however do not enable HWP, no matter which version of Windows you're currently using.

MSI btw. doesn't directly enable HWP but gives you an option to enable the CPPC interface as you can see in your own screenshot

Enable/Disable Intel(R) Speed Shift Technology support. Enabling will expose the CPPC v2 interface to allow for hardware controlled P-states.

I.e. enabling this option doesn't directly activate HWP but exposes the CPPC interface to the OS. The OS's CPPC drivers are then responsible to actually enable HWP.

Manually enabling HWP via the MSRs in Win (10) without CPPC drivers means that your system won't use the full potential of HWP like application specific QoS.

 

[SuperJaw]

Quite odd to see the lack of Speedshift support still from some mobo makers. I am just happy both my NUC and MSI board got it right away.

But I guess its a prioritization from those board makers.

Motherboard reviewers and tech news sites should put a little more pressure to support major hardware features like SST, (doesn't seem to get mentioned as a Pro or Con in reviews). I wouldn't have purchased my Gigabyte board had I known the their lax attitude towards SST. It seems most reviews just focus on the spec list and cosmetics. Would've definitely gone MSI this time around.

 

[wpcoe]

I've got an ASRock Z170 Gaming-ITX/ac motherboard, and don't see any obvious Speed Shift options in UEFI/BIOS.

Then again, there are UEFI settings that I'm not sure about, as I asked in another thread. Such as:

PACKAGE C STATE SUPPORT -- "Enable CPU, PCIe, memory, graphics C State Support for power savings" -- is disabled by default. Is there any reason to NOT enable it?

PCIE ASPM SUPPORT
PCH PCIE ASPM SUPPORT
DMI ASPM SUPPORT
PCH DMI ASPM SUPPORT
-- all are disabled by default. Again, any reason NOT to enable them?

DEEP SLEEP -- "We recommend disabling Deep Sleep for better system compatibility and stability." Most of the articles I find Googling deep sleep deal with Realtek network controllers. On this motherboard (with Intel Ethernet LAN and Broadcom WiFi) what does deep sleep refer to?

Would any of those settings enable Speed Shift?

 

[chinobino]

@wpcoe

Unfortunately not.

@Free42

Some very good info there, keep it coming.

 

[chinobino]

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

Some interesting info;

Microsoft decided to market Speed Shift as some sort of Windows 10, latest and greatest version, only feature but that is not true. It is an Intel Skylake feature and Intel's public documentation shows how to enable Speed Shift no matter what operating system you are running. No bios support is needed either. All software has to do is write 1 to MSR 0x770. Now the Linux guys can check out Speed Shift too.
Once enabled, it stays enabled and yes, the Windows power profiles will be ignored when Speed Shift is enabled. I have not tried this in Windows 10 v10586 yet so not sure what happens in that situation when Microsoft is also using Speed Shift. On my desktop Skylake, going into Stand By mode and then resuming disables Speed Shift. That's why I included the ThrottleStop option - Enable Speed Shift when ThrottleStop starts. If this is checked, Speed Shift will be enabled when TS starts or when you resume from stand by mode. Depending on the motherboard and what Microsoft is doing, you might be forced to reboot to disable Speed Shift.

If Speed Shift is enabled, the ThrottleStop Set Multiplier feature and the various Windows power profiles will be ignored. In this state, the CPU gets to decide what is best for performance vs power consumption.

The Speed Shift register has another variable that should be able to offer some more control. It can be adjusted from 1 to 255 which varies the aggressiveness of Speed Shift. The Intel recommended default is 128. I will find some room and add this extra variable to ThrottleStop so you can play around with it. I left this out for initial testing because I could see users setting this too low, their CPU would get stuck at 8 and then they would blame ThrottleStop for making their laptop so sluggish. Writing software that makes everyone happy is a tough gig at times.

Setting this variable above 128 should allow your CPU to reach the maximum multiplier when stress testing.

 

[ShintaiDK]

Motherboard reviewers and tech news sites should put a little more pressure to support major hardware features like SST, (doesn't seem to get mentioned as a Pro or Con in reviews). I wouldn't have purchased my Gigabyte board had I known the their lax attitude towards SST. It seems most reviews just focus on the spec list and cosmetics. Would've definitely gone MSI this time around.

I kinda feel lucky I went MSI. I am surprised so many of the "top makers" are so ignorant to actually doing proper support. On the other hand they are also busy doing auto overclocking out of the box to get a 5-10% "performance" boost over others in motherboard tests as pure BS that even caused problems previously due to stock coolers.

When reviews do performance tests on boards with virtually no difference. Yet doesn't go into the features that its all about. Its really just sad.

 

[Dave2150]

I kinda feel lucky I went MSI. I am surprised so many of the "top makers" are so ignorant to actually doing proper support. On the other hand they are also busy doing auto overclocking out of the box to get a 5-10% "performance" boost over others in motherboard tests as pure BS that even caused problems previously due to stock coolers.

When reviews do performance tests on boards with virtually no difference. Yet doesn't go into the features that its all about. Its really just sad.

'Auto overclocking' such as Asus's multicore enhancement options makes far more difference than enabling SST. SST is designed for mobile devices, where battery life is critical. It provides next to no tangible performance increase, since the majority of desktop users don't let their CPU's downclock in the first place. A 200Mhz+ automatic overclock will provide performance increases, no matter which way you try and spin it.

FYI I personally manually overclock my systems, I've had my 6700k @ 4.7Ghz @ 1.28v since the 7th of August 2015 on air, Asus stability ftw

 

[Phynaz]

the majority of desktop users don't let their CPU's downclock in the first place

Majority? How about < .01%.

Only an anal enthusiast mucks around with their processor P-states, the vast majority of people don't even know they exist.

 

[TheRyuu]

Majority? How about < .01%.

Only an anal enthusiast mucks around with their processor P-states, the vast majority of people don't even know they exist.

Lets not forget that messing around with processor P-states is also probably useless. Modern processors are pretty good at the whole power saving thing and sometimes you should just let it do its thing.

 

[Phynaz]

Lets not forget that messing around with processor P-states is also probably useless. Modern processors are pretty good at the whole power saving thing and sometimes you should just let it do its thing.

Exactly. Intel knows how their CPU's function better than just about any DIY'er does.

 

[Dave2150]

Majority? How about < .01%.

Only an anal enthusiast mucks around with their processor P-states, the vast majority of people don't even know they exist.

I obviously meant 6600K/6700K owners, we're not talking about business desktops here after all....

I think a good % of 6700k owners are running their CPU's at maximum turbo all the time. Look at any overclocking guide and you'll see it recommended to disable the P states.

The post I was replying to was still absurd - implying that speedstep is more of a performance increase than Asus' auto overclock facility. A 200Mhz+ overclock will do much more than some mobile speed step gimmick.

 

[SuperJaw]

I think this test proves that there are benefits to both mobile and desktop. 50 or 60ms is a relative eternity. I don't drive my car at 6000rpm all the time just in case , but I want it to get there ASAP when I need it.

https://www.pcper.com/reviews/Proce...sted-Significant-User-Experience-Improvements

 

[Dave2150]

I think this test proves that there are benefits to both mobile and desktop. 50 or 60ms is a relative eternity. I don't drive my car at 6000rpm all the time just in case , but I want it to get there ASAP when I need it.

https://www.pcper.com/reviews/Proce...sted-Significant-User-Experience-Improvements

The difference between 0% load 800Mhz and 0% 4600Mhz is very little TDP wise, I prefer running mine at 4.6Ghz 24/7. No delay at all then