Dell XPS Speed Shift

UPDATE MAR 2019

*** HOW TO CONFIRM SPEEDSHIFT "EPP" VALUE IS STICKING
http://forum.notebookreview.com/threads/dell-xps-speed-shift.796891/page-37#post-10889194

*** DELL HAS ENABLED SPEEDSHIFT VIA BIOS IN 9570

SPEED SHIFT BACKGROUND​

One of the exciting features of Skylake processors was Intel upgrading SpeedStep with SpeedShift. This could make your CPU snappier and more efficient. I am seeing a significant jump in CPU performance & latency in extreme situations (with minor caveats).

A good summary of Speed Shift is published by Anandtech:

http://www.anandtech.com/show/9751/examining-intel-skylake-speed-shift-more-responsive-processors

Recent XPS 13 & 15 laptops (Skylake & Kaby Lake) should all be ready to run SpeedShift:

- 9570 ==> Dell enabled SpeedShift by default in recent BIOS update​

- 9350 ==> Dell enabled SpeedShift by default in recent BIOS update
- 9560, 9360, 9550 ==> Dell disabled SpeedShift by default (see how to enable below)

==> XPS KABY LAKE users update <<==
read this guide & special section below. Then read thread before beginning research

HOW TO ENABLE SPEEDSHIFT ON XPS
Dell Enabled SpeedShift on new 9570 via BIOS.

http://forum.notebookreview.com/thr...-its-predecessors.816824/page-4#post-10731554

Whilst Micro$oft enabled SpeedShift November 2015, Dell has not yet supported SpeedShift via a BIOS update in all systems. And in March 2017, Dell Engineering stated they will not officially enable SpeedShift on the 9550 & 9560:

==> Regardless of Dell's official support, there are a few ways to enable SpeedShift:

1. Download ThrottleStop 8.40 or newer software
The ThrottleStop Guide

2. Apparently write 1 to MSR 0x770

HOW TO SEE IF SPEEDSHIFT IS ENABLED
You can download HWiNFO64, which indicates SpeedShift is enabled when the letters "SST" are green (not red).

You can also open up the ThrottleStop TPL window and see if the Speed Shift option has a check mark beside it. If Speed Shift is checked and can no longer be toggled off, that means Speed Shift is enabled

*** Note, after my laptop is powered down I need to relaunch ThrotteStop 8.30 to re-enable SpeedShift.

PERFORMANCE IMPROVEMENTS W/ SPEEDSHIFT
ThrottleStop 8.30 was released this week and I am seeing a significant jump in CPU performance using the new SpeedShift features.

There are now three key variables to adjust on SpeedShift in ThrottleStop 8.30:

EPP: 0 to 255
Min: 1 to 32
Max: 1 to 32

* EPP is "energy performance preference" (which some quote in range of 0-100 or 0-64 etc so be careful). Low (0) is higher performance, high (255 here but 100% or 64 on some systems) is energy savings

* EPP seems to be the feature that boosts the performance; I only saw minor improvements in ThrottleStop 8.20 (which did not have EPP)

* For min/max ranges, my CPU only goes to 3.2ghz so you may adjust the "32" upwards to 35 for 3.5ghz processors (etc) I suppose.

* Setting Min & Max both to 32 seems to disable EPP and does not provide me with "best performance"

The "best performance" settings for me (on 3.2 ghz turbo max on i5 6700HQ) were:
EPP: 0
Min: 1
Max: 32

My testing is using a digital piano that calls up very large samples in real time. Latency is critical. When the CPU falls behind, one typically hears clicks and pops from the CPU stuttering. The typical solutions include getting a faster CPU, slowing down the data, shrinking the data packages; or try SpeedShift...

Using ThrottleStop 8.30 with above "best performance" SpeedShift settings, I could run very aggressive samples with no stuttering. However, after ~10 minutes the stuttering returned so still investigating there...

WILL EPP~50% REDUCE CPU SPEEDS?
Intel seems to recommend laptop makers to set default EPP in the 50% (or 128 [of 255]) range for a balance of power and energy use.

- 9550 i7 - GoNz0 confirmed max CPU speeds can be reached with EPP=50% and EPP=65% (p.14).
- 9550 i5 - my CPU hits max speeds with EPP=50% but has slightly reduced CPU speeds at EPP=65%. I use about 30% (or 78/255) for a performance tilt based on UncleWebb's recommendations
- 9560 i7 - Rockstar confirmed max CPU speeds can be reached with EPP from 0% to 75% (p. 14)
- 9560 i7 - GoNz0 confirmed max CPU speeds can be reached with EPP from 0% to 40% (http://forum.notebookreview.com/thr...5-9560-kaby-lake.802345/page-21#post-10580298)
- 9360 - EFI hack reported to work by Hakujou Sep 2017 (see below)


We had better results with SpeedShift EPP than Uncle Webb did with a different Kaby Lake computer:

9560 & 9630 - KABY LAKE?

9560 users have reported that Dell has disabled SpeedShift in Kaby Lake machines. I have enabled SpeedShift with ThrottleStop on a 9560 i7 7700HQ.

Use caution editing MSR-BIOS on Kaby Lake as there are some differences.

A. That said, RockStar posted success activating SpeedShift on 9560 (7700HQ) via RWEverything:

B. For BIOS fixers

1- RockStar used Goodwin's codes to provide step-by-step instructions below. He also confirmed this worked on his 7700HQ

2- GoNz0 posted similar step-by-step instructions (linked below)

3. Hakujou posted EFI specifically for the 9360 (linked below)

4- Goodwin did the ground work to figure out and document the 9560 codes for us below (thanks!); his post and comments are below

5- tiger501 provided a tutorial for some advanced options in 9550 BIOS (post and link below)

==>GonZ0's Step-by-Step post
http://forum.notebookreview.com/threads/xps-15-9560-owners-thread.800611/#post-10440059

==>Hakujou's EFI variables for 9360

OTHER INTERESTING SPEEDSHIFT REFERENCES
http://www.pcper.com/reviews/Proces...sted-Significant-User-Experience-Improvements

Direct way to enable SpeedShift manually on XPS 9550
https://forums.anandtech.com/threads/intel-speed-shift-technology-mia.2463727/page-2
http://rweverything.com/
>CPU 1
>WRMSR 0x770 0x0 0x1

Powercfg command line options - parameters from M$
https://msdn.microsoft.com/en-us/library/windows/hardware/mt644885(v=vs.85).aspx

Task scheduler primer
http://www.digitalcitizen.life/use-task-scheduler-launch-programs-without-uac-prompts

Sample powercfg and GUI instructions by jntfoster & technical detail by fo3nx (for SP4)
http://answers.microsoft.com/en-us/...e/a20f481a-fe3d-4e15-b0b5-ea11f8c62795?page=8

Intel SpeedShift technical doc (around page 474)
http://www.intel.com/content/dam/ww...eveloper-system-programming-manual-325384.pdf


**Some references help with writing MSR. Also see UncleWebb's detailed comments below

Package C-State fix for improved battery life & thermals?

For those interested in potentially improving battery life and thermals. It appears that some laptops (including the XPS 15) can not enter deep Package C-States due to issues with IDE ATA/ATAPI control driver from M$.

One way to check is via ThrottleStop (free monitoring software):

- exit all other programs

- click the button at the bottom of the panel labeled "C8" or similar

- Look at section labeled "Package C State Percentage"

--- Before, with M$ driver, my system was showing say 80% stuck no higher than C2

--- After, with updated Intel driver, my system was showing say 80% in deeper C States (e.g. C8)

Special thanks to @Che0063 and @CraftyClown who identified issues and found a simple solution. So hopefully they will draft a guide as they do more testing and have some time. More details on issue and fix are linked below:

 

I did some simple tests on my i7 XPS 9550 using TS 8.30 with the latest Windows 10.
With EPP set to 0, i noticed that the CPU is stuck at a quite high frequency; with EPP set to 255, i got an opposite result; if i set EPP to a value above 128, the CPU won't hit the highest frequency.
I'm an Android programmer, so i measure the possible performance improvement using Android Studio 2.2, observing the cold start time. I didn't notice the start time had significantly decreased.

FYI: I set min to 1 and max to 35

 

Nocchi, is "cold start time" a good measure of CPU performance or a better measurement of SSD performance?

 

Speed Shift allows the CPU to go from minimum speed to maximum speed approximately 100 ms faster compared to the Windows Balanced power profile. During some short tasks, that might make a significant difference. If you use a benchmark test takes 20 seconds, 100 ms only represents 0.5% of the duration of the benchmark. For the other 99.5% of the time, the CPU will be running at the exact same speed whether Speed Shift is enabled or not. This is why the majority of traditional benchmarks do not show any measurable difference in performance. For longer benchmarks, there isn't any difference.

To see if Speed Shift is enabled you can use HWiNFO but you can also open up the ThrottleStop TPL window and see if the Speed Shift option has a check mark beside it. If Speed Shift is checked and can no longer be toggled off, that means Speed Shift is enabled.

With the latest versions of Windows 10, you should be able to go into the Control Panel Power Options and toggle Speed Shift on and off as well as adjust the Energy Performance Preference value. The register in the CPU can be adjusted from 0 to 255 but most things in the Power Options are scaled so the range there will be 0 to 100. If these options are not visible in Power Options, they might be part of the hidden settings.

@pressing - Thanks for posting those links. Speed Shift disables some common throttling methods that manufacturers use so some of the performance improvements you are seeing might be because of that. In theory, you should be seeing maximum performance and minimum latency when your CPU is running at its maximum speed. If it has some throttling issues, Speed Shift might not cure the problem but perhaps it can delay the inevitable slow down. Why not turn on the ThrottleStop Log File option while testing so you have a record of what the CPU multiplier is doing.

@Dufus used to use a graphing tool that he wrote that I think could measure the CPU multiplier in 1 ms intervals. That data would be interesting when testing out Speed Shift. I might have to build something similar.

 

Thanks for visiting UncleWebb!

With SpeedShift enabled, I ran a ThrottleStop log file for 5 minutes until the crackling started. Didn't see much in variance; the multipliers were pegged around 30-31x, temps were 35-40*C, etc

I did try all sorts of combinations of BIOS (disabling c-states, turbo, SpeedStep) disabling SpeedShift in ThrottleStop to no avail. One discovery was that disabling the turbo (only) resulted in a major increase in crackling

Also, LatencyMon seemed to show decent results. Except driver Wdf01000.sys (kernal mode driver framework runtime) may be a problem (high TOTAL execution time). I don't think I can disable that unfortunately so will keep hunting around.

 

CPU performance matters. When i had "Power Saver" option checked in TS, the start time of Android Studio would significantly increase (of course with SST disabled).
And after reading some articles and unclewebb's post, i'm aware that SST won't bring any performance improvement if CPU has already been running at a high frequency. Based on my tests and your post (set EPP to 0), i guess that you feel SST boosts the performance because your settings keeps the CPU running at the highest frequency.

 

Thanks for your post Nocchi...

I spent some time trying to figure out what the buttons on ThrottleStop were doing but am still working them out. It seems that "Power Saver" might be suboptimal for modern processors. That said, the developer UncleWebb, can't test everything and Intel has a lot of secrets that are not public. So he recommended users spend some time testing ThrottleStop to find optimal settings for their systems:

Good point on setting EPP to 0. Not really sure what is going on...
- The virtual piano software CPU Usage spikes when notes are played but is otherwise quite low.
- On the face of it, if we keep the Idling at a high frequency, it seems SST should not provide improved performance.
- Maybe SST is allowing the CPU to fully ramp up to "higher wattage" Usage spikes more rapidly (?)

Regardless, the virtual piano software performs significantly better with SpeedShift enabled (for a while!)

 

> With the latest versions of Windows 10, you should be able to go into the Control Panel Power Options and toggle Speed Shift on and off as well as adjust the Energy Performance Preference value.

I thought I am on the 'latest' version - I can't find such a setting anywhere. And wouldn't it mean that Dell does not need to update their Bios anymore to make the 9550 Speed Shift ready? Confused...

 

Windows provided support for SpeedShift in Nov 2015. For the SpeedShift setting to be enabled on the the 9550 via control panel power options, Dell needs to provide SpeedShift support (presumably via a BIOS update we are waiting for).

You can get around Dell's bad user support by writing 1 to MSR 0x770 or using ThrottleStop 8.30, for example...

 

Adding Speed Shift Control to the Windows Power Options

Before starting, I would check to see if these options are already part of your Windows Power Options. In Windows 10, to open up a command window with Admin privileges, hold down the Windows key on the keyboard and then press X to pop open the window. Let go of those keys and press A for the Command Prompt (Admin) option.

To query your power options type in

powercfg -qh >C:\report.txt

This command creates a lot of information so I redirect the output (>) to a file called report.txt which will be located directly on my C: drive. You can redirect the output wherever you like. The -q option stands for query and the h part of that option will show hundreds of hidden settings. Search that file for the 3 Speed Shift related entries.

(Processor energy performance preference policy)
(Processor autonomous activity window)
(Processor performance autonomous mode)

If any of those 3 are missing, copy and paste each powercfg command into the command window and press enter. After you are done adding those, I would reboot.

Balanced Profile

(Processor energy performance preference policy)

Code:

powercfg –setacvalueindex SCHEME_BALANCED SUB_PROCESSOR 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 50
powercfg –setdcvalueindex SCHEME_BALANCED SUB_PROCESSOR 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 50

(Processor autonomous activity window)

Code:

powercfg –setacvalueindex SCHEME_BALANCED SUB_PROCESSOR cfeda3d0-7697-4566-a922-a9086cd49dfa 0
powercfg –setdcvalueindex SCHEME_BALANCED SUB_PROCESSOR cfeda3d0-7697-4566-a922-a9086cd49dfa 0

(Processor performance autonomous mode)

Code:

powercfg –setacvalueindex SCHEME_BALANCED SUB_PROCESSOR 8baa4a8a-14c6-4451-8e8b-14bdbd197537 1
powercfg –setdcvalueindex SCHEME_BALANCED SUB_PROCESSOR 8baa4a8a-14c6-4451-8e8b-14bdbd197537 1

High Performance Profile

(Processor energy performance preference policy)

Code:

powercfg –setacvalueindex SCHEME_MIN SUB_PROCESSOR 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 50
powercfg –setdcvalueindex SCHEME_MIN SUB_PROCESSOR 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 50

(Processor autonomous activity window)

Code:

powercfg –setacvalueindex SCHEME_MIN SUB_PROCESSOR cfeda3d0-7697-4566-a922-a9086cd49dfa 0
powercfg –setdcvalueindex SCHEME_MIN SUB_PROCESSOR cfeda3d0-7697-4566-a922-a9086cd49dfa 0

(Processor performance autonomous mode)

Code:

powercfg –setacvalueindex SCHEME_MIN SUB_PROCESSOR 8baa4a8a-14c6-4451-8e8b-14bdbd197537 1
powercfg –setdcvalueindex SCHEME_MIN SUB_PROCESSOR 8baa4a8a-14c6-4451-8e8b-14bdbd197537 1

These options are hidden by default in Power Options. After you have rebooted, open up the command window again with Admin privileges and enter the following 3 commands so these new entries show up in Power Options.

To show these new settings in Power Options

Code:

powercfg –attributes SUB_PROCESSOR 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 -ATTRIB_HIDE
powercfg –attributes SUB_PROCESSOR cfeda3d0-7697-4566-a922-a9086cd49dfa -ATTRIB_HIDE
powercfg –attributes SUB_PROCESSOR 8baa4a8a-14c6-4451-8e8b-14bdbd197537 -ATTRIB_HIDE

After this, open up your Power Options, scroll down to Processor power management and there should be some new options in there. I believe that "Autonomous mode" is what Microsoft calls Intel Speed Shift. If you ever need to hide these options just change the minus sign (-) before ATTRIB_HIDE to a plus (+) sign in the above 3 commands.

With Autonomous mode enabled, check with HWiNFO to see if SST is displayed in green. In theory, you will have to set this option to Disabled and then you will have to reboot to disable Speed Shift.

I do not yet know if a bios update is needed or if Windows Power Options can be used to set MSR 0x770 to 1 to enable Speed Shift. Depending on what you guys find out, it would be easy to create a small .bat file so adding all of these options to Windows could be done in one simple step.

 

I have had this set for a while and it doesn't enable SST, that only happens when MSR 0x770 is run.

 

Thanks Uncle Webb for the detailed post. I successfully was able to add the 3 new settings in power options. Unfortunately, they seem to display dummy variables:

- changing data in power options window does not have any effect on system (per HWiNFO64 and powercfg query)
- displayed data in power options window may or may not be correct (per HWiNFO64 and powercfg query)
- displayed data in power options window does not change (even with change in ThrottleStop, powercfg, rweverything - even after full shut down)

Similarly, powercfg commands and queries are not 100% reliable around SpeedShift

I did manage to enable SpeedShift via RWEverything but I couldn't get the other nifty SpeedShift options working with any certainty (I used powercfg so there is probably a better way via MSR 0x771 but that is above my pay grade).

So until a Dell 9550 BIOS update comes down the pike, ThrottleStop is an easy way of accessing these settings. . .

 

From the "horses" mouth, followed by my reply.

me

 

Thanks GonZ0. I will harass Dell on the BIOS upgrade today.

I would encourage everyone here to call Dell and request SpeedStep be enabled via the upcoming BIOS update - Thanks!

 

Thats funny,

I asked Dell in May about SpeedShift and the answer was, that SpeedShift ist already enabled, but not showing up in HWInfo64...They told me in May, that this will be solved by another BIOS Version. Now they say that the feature will not be available. Funny. I can provide my emails with my conversation if necessary but it is in german.

 

Keep harassing them and escalating to senior engineers... This is a nice performance upgrade that is easy for Dell to implement in the next BIOS...

 

Try making noise via dell forums also...

http://en.community.dell.com/support-forums/laptop/f/3518/t/19994644

 

I wrote Dell Support to ask them which answer of them is right...As I said, they told me SpeedShift is supported but not shown right by HWInfo and you are told, that it will never be supported. I think somethings not right in their answers

 

Because they don't know sh!t

Sent from my SM-G920F

 

Just for the record, HWiNFO works fine. If Speed Shift is enabled and you start HWiNFO, SST will be green. This confirms that Speed Shift is enabled. You might need to restart the HWiNFO System Summary page after enabling Speed Shift to update the SST flag. If you have a laptop with a bios that properly enables SST, make sure to test that SST is still green after resuming from Stand By.

Be careful how much noise you make. Dell banned me from their forums for being a little too vocal. Sadly, the cheapest way to fix some problems is to just not talk about them and hope that no one important notices.

 

My 9350 i7 iris has speed shift enabled by default

 

Thanks for the info, thats just what they told me . I enabled it with a script and RWEverything. Then I used throttlestop and HWiNFO to check if it is working and everything seems to be fine.
It would be great to be able to get it as an Option in BIOS and not via script. Lets see...

 

Thanks for the confirmation!

Thanks for the recommendations. We are walking a fine line so need to be diplomatic. The Dell team tend to ignore selected threads. But a little positive motivation via active forums and polite service calls should encourage the engineers to see that users are excited about a very cool feature (that is easy to implement).

 

I find it hard to be polite to the tech dealing with me at Dell after telling me they had a firmware update for the Samsung SSD then sending me a frigging driver.

 

That is great news Sebbe.

I. On the 9350 do you have the ability to ramp up EPP to max performance (0) via BIOS? That is a very powerful feature.

II. Alternatively, out of the factory, was there an adjustable EPP dropdown menu? a. Go to "Control Panel>Hardware and Sound>Power Options>...Advanced Power settings>"

b. Navigate to "Power options" window and the "Advanced Settings" tab

c. then you scroll down and click the "+" sign on "processor power management"

d. then there should be a row titled something like "processor energy performance preference policy". Clicking on the "+" sign there you should be able to adjust from 0% to 100%. I assume 0% gives you max performance and 100% will be max efficiency (but super laggy performance). Try adjusting as the difference is immediately obvious to the user experience.

On my 9550 the "Processor energy performance preference policy" option did not show up on the "power options". I "unhid" the dropdown but there are just dummy variables that do not work...

 

In case anyone is interested I automated the enabling of SST through TaskScheduler and ThrottleStop so that SST is enabled after startup and coming back from sleep.

More information can be found here:
https://gallery.technet.microsoft.com/Automatically-Enable-ad9c2208

Let me know if there is anything I can change/add.

Cheers.

 

I. No i can't do anthing like that.
II. No, i can't see any preference policy, only min processor, max processor and cooling policy...

 

It appears Dell is not providing Speed Shift BIOS support on the brand new 9360 Kaby Lake XPS 13 : (

 

http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/page-514

The above post shows how to add ThrottleStop to the Task Scheduler so you can make sure that Speed Shift is enabled whenever your computer is running. It also shows why you need to restart HWiNFO after the status of Speed Shift has changed.

@pressing - Enabling Speed Shift can interfere with many of the throttling schemes that manufacturers have dreamed up over the last six years. There must be some reason why Dell is still hesitant to turn on Speed Shift in their recent laptops. They have had over a year to think about this and still nothing.

 

Agreed.

But there are some important caveats here:

0. Speed Shift was an important selling point of Intel's Skylake product line (importantly, there weren't many compelling selling points for Skylake HQ chips)

1. Speed Shift has enjoyed full support of Intel and Micro$oft for over a year

2. Speed Shift works fine via ThrottleStop 8.3x

3. Dell's review of drivers and firmware for the 9550 has been cursory (at best) so there is clearly no corporate mandate to provide anything beyond minimal testing or QC for Speed Shift (or anything for that matter)

With respect to item #3, you may not know that Dell released the 9550 in beta form a year ago. The BIOS and many of the drivers were absolutely atrocious. After about 6 months, most of the bugs were worked out. After a year, there are still a few major issues (e.g. flickering problem with the 4k screen, challenges with some nvme drives).

Nobody appreciates being sold X but receiving X-Y

 

So I tried using ThrottleStop to enable Speed Shift on my i5 9550. I realised that in battery mode, I was getting poorer endurance i.e. the battery runs down a lot more quickly than without ThrottleStop. I left everything at default, no overclocking, no throttling, nothing. Anyone else has the same observations?

 

Edit - see comments below from Uncle Webb, the engineer behind ThrottleStop who was kind enough to drop by.

Also read through the ThrottleStop thread for more info. A few tips to get you started
- Make sure you have power options (in Windows control panel) set at high performance when using ThrottleStop (not balanced)
- See benefits of undervolting below
- Don't click on the "disable turbo"

 

I only have one in my script, what's the 2nd one for?

 

MSR 774 sets a variety of parameters. I think the most interesting option is allowing one to set energy performance preference (EPP) from 0 for max performance to 255 (?) for max energy savings. You can also set min and max CPU performance. And other stuff I don't understand. There is also MSR 772.

EDITED FOR CRITICAL TYPOS

Check out the official details included in the following Intel tome before going further (particularly around p 474):
http://www.intel.com/content/dam/ww...eveloper-system-programming-manual-325384.pdf

Caution - don't try this at home kids, high risk of bricking computer forever! I'm serious about this warning so any problems are 100% on your own shoulders. I like to tinker but have zero formal computer expertise. I haven't seen anyone brick their computers with ThrottleStop so that might be a better option if you want to experiment with Speed Shift. Also the ThrottleStop team actually understands these parameters and obviously does some stress testing...

Preliminary draft notes below include what I wrote into the MSR Editor in CrystalCPUID for your comments and corrections (EDIT - TYPOS & ADDED MSR 772!):

A. MSR 0x770 is IA32_PM_ENABLE (Enable/Disable HWP [Speed Shift])
0X00000770
0x00000000
0x00000001

B. MSR 0x772 IA32_HWP_REQUEST_PKG
[EDIT - Didn't know about this one. You can figure that out! first try...]
0X00000772
0x00000000
0x00002001

C. 0x774 IA32_HWP_REQUEST (Conveys OSPM's control hints (Min, Max, Activity Window, Energy Performance Preference, Desired) for a single logical process)
WHAT I USED FOR 9550 i5 (I hope this sets EPP to max performance 0 (00) and max processor speed to 3.2ghz (20):
0X00000774
0x00000000
0x00002001

SO MAYBE (???) 9550 i7 would be something like (setting EPP to max energy efficiency 255 (FF) and max processor speed to 3.5ghz (23):
0X00000774
0x00000000
0x FF002301

[EDIT - the FF was in the wrong place - sorry!]
_________
FYI - User f02nix etal at the Surface Pro forum linked below recommended the following for those laptops (NOT for the Dell 9550). I used it as a starting point:
"The MSR at 0x774 should read 0x000000007F002201 for i7 (0x000000007F001E01 for i5 or 0x000000007F001601 for m3).

000 Activity_Window (bits 41:32, RW) - Hint on the duration of the sliding window used to optimize performance/frequency. Intel default is 0x000 to let the hardware decide itself, while Microsoft set 0x19E indicating a 30ms window.

7E Energy_Performance_Preference (bits 31:24, RW) - Hint on how to balance perfomance vs energy savings. The Intel default value is 0x80, whereas Microsoft uses 0x00 for maximum performance leading to nonstop max turbo

00 Desired_Performance (bits 23:16, RW) - Anything but 0x00 indicates Speed Shift to be disabled or overruled by the OS."

http://answers.microsoft.com/en-us/...e/a20f481a-fe3d-4e15-b0b5-ea11f8c62795?page=8
_________

 

You are putting the FF in the wrong position. It should be 0xFF002301. Same with your 7F. It should be 0x7F002301. The Intel default value for EPP is 0x80 which is equivalent to 128.

There are 2 separate registers that control EPP. One for the individual cores and then a second register controls the entire CPU package.

I find doing scientifically valid battery testing to be very difficult. There are so many variables including over 1000 background threads running on a typical Windows 10 PC that it is almost impossible to come up with a controlled testing environment. When you are running ThrottleStop and your PC is totally idle, what does ThrottleStop show for the percentage of time the CPU is in the C0 state? When you open up the C State window, what sort of percentages are you seeing? The biggest problem I see is that most users have a whack of stuff running in the background, sometimes without their knowledge or consent, which can significantly interfere with battery run time. ThrottleStop itself is the least of ones problems compared to some of the background bloat that is running.

Another problem is that some manufacturers are not using all of the power saving C States that Intel has made available. I have seen many newer laptops that are limited to the Package C2 state which consumes a lot more power compared to the deeper package C States. Here is an example of how my 4700MQ looks from a few generations ago.



When idle, less than 0.5% in the C0 state and individual cores spending on average 99% of the time in C7 are good numbers to shoot for. Intel has introduced package C States like C8, C9 and C10 but I have not yet heard of any manufacturers using these.

To improve battery run time, I would definitely be using ThrottleStop to under volt the CPU. This is where energy savings can be found. Intel's default voltage settings for Skylake CPUs seem to be much higher than needed.

 

I want to open one big super secret. There is a way to enable SST permanently But as usual, dragons are inside, so it is possible to totally f**ck up you laptop (what i did with my experiments, so used hardware programmer to restore my bios eeprom). In short - for just enabling SST it is enough to change UEFI variable 0xD8 from value 0x00 to 0x01 (did on my laptop, so after each reboot hwinfo64 shows SST flag as green, and i have no ThrottleStop even installed) . If somebody brave is interested in details i can make short how-to manual somewhere after next Monday.

 

Thanks for taking the time to read over this and correct some critical errors. I edited the post per your comments. Really appreciate your mentoring here and quite enjoy peeking under the ThrottleStop hood to learn a bit.

Didn't know about that!

Corrected earlier post regarding energy savings with ThrottleStop.

I would add on the performance side, undervolting via ThrottleStop is easy. It can have a massive impact on thermals, particularly for a small laptop like the 9550 which has limited cooling ability by definition. On my i5, benchmarks with a -170mv undervolt showed ~12*C drop in temps. Repasting the CPU/GPU and fixing thermal pads on the VRAM only provided another ~5*C drop in temps. Those deltas are at the limit so are not entirely representative. But others have reported similar stats. The results speak for themselves.

 

Hi goodwin - I think that would be a fantastic learning opportunity. Please feel free to document your process here or on a new thread as you like. Kindly add a bunch of disclaimers and warnings to scare off the novices (like me)!

 

I was thinking a lot, and due to too high risk of breaking laptop i will not describe whole process for modifying nvram variables. But i will share key information for thus who is brave to do that. As for me - i did this on two XPSs and both are working great.
So, first - intro. We already have option for enabling Speed Shift in our laptop, it is inside of our BIOS and it is actually working. And works fine. But for some reason (who know what body-part they are using for thinking process) - Dell engineers decided to remove this option from BIOS interface. But there is a way to change it. Ok, i won't be so rude - there is one of common guides to change variables through UEFI shell - but be carefull, one mistake and you will break your lovely laptop! Just skip steps from B to I from that guide - i will tell you exact variable to modify.
So, first identify that your variable has correct value:

Code:

setup_var 0xD8

It should return value 0x0. Next modify value:

Code:

setup_var 0xD8 0x1

Now you can reboot and check in windows that your Speed Shift is now working just as it should without manually enabling it.
P.S. I'm not taking any responsibility agains broken laptops. So just be careful!

 

So I gave ThrottleStop another shot, mainly because I didn't like the bloat that Intel XTU had, as well as XTU's handicap of app based undervolting. It's been a few weeks since I've been on throttlestop and it works flawlessly. I basically used it to enable SpeedShift, as well as undervolt CPU/Cache and GPU in both plugged in and battery modes with the following respective settings: -100mV, -50mV & -150mV, -80mV.

These settings work great on my i5 and although I could have gone lower, I decided to leave a little buffer for stability.

 

Thanks for this but I think it should be down to the user to decide if they want to risk bricking a laptop and should only be done if you have a valid warranty so Dell can come swap the mobo!

A link to the guide would also help.

 

That's why i decided to post it there. People have right to decide. Just want to filter newbies that will break laptops and than will start crying everywhere. About guide - just forgot to insert link. On evening will find it again and edit my post.

 

Blurblur,

As you probably know, there are over a dozen undervolt results posted here at NBR and one user posted a spreadsheet summarizing some of those results. Each chip is different so you need to stress-test your system to know where your CPU stands but user reported data provides a good baseline. The i5 has a lot of undervolting potential (vs i7).

I run -170mv on the CPU-cache and it provides massive thermal benefits with zero instability. You might try testing a more aggressive undervolt. Interestingly, the related new Kaby Lake chips take advantage of this delta, and the "increased performance" corresponds to very little undervolt head-room.

Also, you should run throttlestop on power plan "high performance". But on battery you might test running power plan "balanced" without throttlestop. Unrelated (I think), in Dell Command-power manager, under thermal settings "Quiet" probably provides better battery life than "ultra performance" does.

 

It seems that enabling SpeedShift overrides Windows built-in "High Performance" power plan CPU governing.

Using windows "High Performance" power plan, without SpeedShift, the CPU is fixed at the highest frequency. With SpeedShift enabled, the CPU freq is no long fixed and jumps quickly. And this causes performance degrade. I ran geekbench W/ SpeedShift and the results were:

Without SpeedShift
Single-Core: 4293
Multi-Core: 13042

With SpeedShift
Single-Core: 4167
Multi-Core: 12812

Of course, when using "Balanced" or "Power Saver", enabling SpeedShift will give better performance, because SpeedShift is faster in increasing the frequency.
But is it possible to automatically disable SpeedShift when switching to "High Performance" power plan?

 

it is probably due to a missing setting or windows performance being able to communicate properly with the BIOS, unless the dell xps13 users with proper speedshift support can confirm the same?

 

Those geekbench results are pretty close. Are the differences repeatable?

With SpeedShift, in the ThrottleStop "Turbo Power Limits" window, did you set EPP=0 and max to your processor's max MHz (i5=32)?

I'm still trying to figure how to enable absolutely highest performance, regardless of heat (as my music software program needs real-time, glitch-free processing of large amounts of data and does not have heat issues). What works for me now is basically:

- Disable SpeedStep in BIOS
- Disable C-States in BIOS
- Use Windows "High Performance" power plan
- Use Dell Command-Power Manager with thermal management at "Ultra Performance"
- Use SpeedShift with EPP=0 & undervolt via ThrottleStop

Under that scenario, the gitches disappear (well with networking disabled). However, the i5 doesn't seem to ramp up to full 3.2 GHz, rather peaking around 2.6GHz. Without SpeedShift, it seems to peak at 2.2GHz. Still trying to fine tune to get this running flat out at 3.0GHz +

Some people argue that the processor should be running CONSISTENTLY fast so things like SpeedShift should all be disabled...

This makes sense

The new 9360 XPS13 does not have proper SpeedShift support either : (

 

Could you please provide a link to the guide. Thanks.

 

Thanks uncelwebb.

May I ask one thing? In ThrottleStop, can we apply different EPP values (in TPL) for different ThrottleStop profiles(Performance, Game, Internet and Batter)? For instance I would like to set EPP to 0 in Performance and Game profile and 255 in Battery profile. I found that single EPP value is applied to all profiles? You know I mean ThrottleStop profiles, not Windows power plans.

 

That is correct. Speed Shift allows the CPU to manage itself so the Windows power profiles become obsolete when it comes to managing the CPU speed.

A different EPP value for each ThrottleStop profile is an excellent idea. Unfortunately, no time at the moment to add this feature. Maybe someday.

 

Can't find that guide again, so i will describe it in short:
1) format flash drive with FAT32 filesystem
2) download that archive https://www.firewolf.science/wp-content/uploads/2015/04/EFI-shell.zip
3) put efi file from archive into Boot/bootx64.efi on your flash drive
4) boot from flash drive. you will get uefi shell
If it is not booting - go to bios, boot sequence - and add custom entry pointing to that efi file. than reboot, F12 and choose your custom entry
After that in shell

Code:

setup_var 0xD8

It should return value 0x0. Next modify value:

Code:

setup_var 0xD8 0x1

After reboot windows will be happy to enable Speed Shift for you. No more need to enable it manually from Throttle Stop (even who knows if it is working properly if you are enabling it after system boot)