• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Intel Speed Shift Technology... MIA?

SuperJaw

Junior Member
Has anyone messed around with Speed Shift Technology? It's supposedly a key feature for Skylake, but only seems to be enabled in a couple laptops and the new Microsoft Surface's.

I've got a Gigabyte z170x-UD3 and can't seem to get it enabled in Win 10 according to HWinfo. I can't seem to find any switches in the BIOS (latest beta). I guess I'm waiting for Gigabyte to add support.

For those of you who haven't heard of it. Speed Shift basically is a feature that lets the CPU control the multiplier and power states directly instead of relying on the OS. This supposedly results in much faster ramps ups that are noticeable in the UI. It should help quite a bit for bursty tasks like web page rendering.

It seems like there was a lot of hype back last November about this, but then nothing really happened after the big Windows 1511 update.

Also, HWinfo seems to be the only piece of software that will tell you if SST is enabled or not. Anyone know of any other tests?

https://youtu.be/YeyT9rUcwtE
 
Last edited:
Its enabled on MSI and Intel NUCs. And its quite good. Essentially its equal to running the CPU at full speed (100% CPU max performance in power profile) all the time in terms of snappyness.

sst.png
 
Its enabled on MSI and Intel NUCs.

That's frustrating that others haven't supported it yet. Your report makes me more determined to get it working. It shows as Red for me in HWInfo.

I've read some comments that it only helps for mobile, but I disagree. My i3-6100 idles at 800mhz in the balanced power plan. So the faster it can crank up to 3700mhz the better. Not sure if Speed Step (SST's predecessor) is any more responsive on desktops versus mobile.

It does feel better (in web browsing) when set to Max Performance (stays at 3.7Ghz), but would rather have some turn down for power and heat.

Ironically features like this probably benefit lighter desktop users more than power users rendering video and such. Huge workloads will keep the CPU pinned SST or not, lighter users should appreciate the speedup and responsiveness.
 
Thanks to Magic Carpet, you can actually just run the 100% plan when you got Haswell/Broadwell/Skylake. My own 6700K changes 2-3W from 800Mhz to 4000Mhz idle. So its not a terrible loss as previous generations.

But I agree, its pathetic some motherboard makers haven't supported it yet.
 
Not enabled yet on Asus boards. Though I found a forum post saying they hadn't completed testing yet since the feature was just released in November.
 
I guess I'll keep it on Max Performance until Speed Step works if it's only a couple watts. It will probably be less since I'm running a dual core. What is magic carpet, I did a quick google search

Is there a chance it will never work? Does it require more than a supporting bios to work, or it is there some special hardware requirement like special power circuitry that Gigabyte and Asus "forgot" to include on there MBs.
 
Is there a chance it will never work? Does it require more than a supporting bios to work, or it is there some special hardware requirement like special power circuitry that Gigabyte and Asus "forgot" to include on there MBs.
From what I read all you need is software support. Hardware-Controlled Performance States a.k.a Speed Shift requires a software initialization through the use of a Register.

Host OS also needs to provide the CPU with some information regarding desired performance or energy usage, and hints based on load history that may affect rate of performance increase/decrease. In other words OS describes the situation to the CPU in raw terms, then CPU makes decisions based on current load.

A comprehensive description can be found here, Volume 3 - Chapter 14.4
 
Last edited:
Hwinfo64 from the NUC in case anyone wishes.
sst2.png


Note that Intel haven't even the AVX2/FMA3 microcode fix ready yet for their own NUCs. As bad karma as those mobo makers without SST support.
 
Last edited:
Just an update... You missing how if you're running the "balanced" windows power plan (with Speed Shift). After running "high performance" for a week, it really does help the responsiveness for bursty loads. I haven't measured the watt difference yet.
 
Hm, one reason why I plan on buying a Skylake system was for testing Speed Shift. Everything that potentially lowers DPCs is interesting to professional audio users who want to use small buffers.
 
Asus Maximus Hero VIII owner here, patiently waiting for SS support.

Glad I chose Asus though still, rock solid overclocks, Intel NIC, fantastic aesthetic combined with windowed case = pure joy 🙂
 
After running "high performance" for a week, it really does help the responsiveness for bursty loads. I haven't measured the watt difference yet.
For an undervolted i5 @ 4Ghz difference is ~5W in estimated package power. (3W for 800Mhz vs 8.5W for 4Ghz @ 1.08V) However, this difference increases with voltage and frequency, if I remember correctly it went up to around 15W for 4.7Ghz @ 1.35V

One can always choose the middle ground, set minimum speed in power options at 50%. Idle power usage sees only a negligible increase (1W) while system responsiveness seems nominal to me.
 
That's odd, considering we have the same MB. However, other components might matter: I'm on iGPU and RAM voltage is 1.35V

No IGP and 1.2V memory. I may have a lower voltage as well with 1.154V at 4000Mhz idle(CPU-Z says 0.956V tho). 800Mhz is 0.832V.
 
No IGP and 1.2V memory. I may have a lower voltage as well with 1.154V at 4000Mhz idle(CPU-Z says 0.956V tho). 800Mhz is 0.832V.
The plot thickens.

When you wrote CPU-Z is a different voltage value than what BIOS reports, I was quite surprised, since in my case the software was reporting the exact same values as BIOS did.

Which leads us to another difference between our systems - your CPU voltage was likely set on Auto for both multiplier and voltage, while mine was set to 40x and undervolted with a -100mV offset. I just set it back on Auto for both multi and voltage and my system also began showing the same characteristics as yours:

  • very small package power delta between 800Mhz idle and 3.9Ghz idle (1-2W)
  • CPU-Z no longer reports the same voltage as BIOS while CPU is idle
PS: it's the CPU multiplier - as soon as change it to a fixed value, software reported voltage and frequency values match. If I change it back to Auto, even if frequency is reported as nominal (3.9Ghz), voltage value oscillates around the minimum value, the one used for 800Mhz.

So even if you set the High Performance profile, CPU is likely hovering around 800Mhz. Whether that means a performance loss or not, it depends entirely on how effective Speed Shift really is.
 
Last edited:
Very interesting. So under volting in your case may have increased power consumption in a lot of cases and only reduced it in more heavy loads.

I also looked twice at CPU-Z and hwmonitor64 and thought hmmm 🙂
 
Very interesting. So under volting in your case may have increased power consumption in a lot of cases and only reduced it in more heavy loads.
It depends on whether one uses a fixed multiplier and also on the choice of power profile. I was on balanced, since it's just a home server, so everything was ok power wise even when system was idle.

This does open op some possibilities though: it seems to me certain combinations between Speed Shift, BIOS settings and Windows power plans may produce unexpected results, at least from a user perspective (Intel engineer may just nod head).

Maybe there's some material in here for a future Anandtech article, cough cough, or just a rainy weekend project.
 
Don't forget that OS power-management always takes priority over Speed Shift! The Balanced profile may take over more often than not.
 
Back
Top