Dell XPS Speed Shift

HI, I have enabled Speedshift in the bios on my 9560 and added the powercfg commands into windows. It changes the EPP successfully viewed in Throttlestop live view.
I have possessor energy performance preference set to 0 Plugged in and 100 on battery. If the power slider is on battery saver The EPP is 255, when I move it to better battery the EPP changes to 178 which is expected. But when I move it to better performance the EPP goes back up to 255! Then if I go on too best performance it jumps down to 84 which again is where id expect it to be.
how would I go about editing the better performance powermode? If I change the possessor energy performance preference values it still behaves the same but just with different EPP values, battery saver and better performance are always the same.

 

So all I've done with my 9560 is enable speed shift by following the guide from rockstar75 in the op. I can confirm it's enabled in hwinfo. Im on pretty old bios. I think like the 2nd or 3rd release after the laptop came out. So there's been many versions since. 1) should I upgrade the bios? 2) if I should, do I have to revert anything before I update the bios?

 

I cant say if its the correct thing or not but I updated my bios with speedshift enabled with the UEFI shell, I had no dramas and it stayed enabled after the update.

 

Thanks! Will give it a try tonight! Hopefully no issues....

 

No issues flashing the new bios last night! Whew!
Still has speed shift enabled but I haven't unlocked any controls via the windows power options so I don't think I'm utilizing it at all right now.

 

You can see SpeedShift variables using the following command. I think the 9560 comes out of the factory at 127 (50%) but can't remember.

powercfg -qh >C:\report.txt

Unlocking the controls in windows power options takes just a few minutes...

 

Yes. Under balanced mode speed shift is set to 50% on battery and 33% plugged in. I take it the lower the number the better? Does this also effect clock states? Meaning if I set it to 0% that it'll maintain the max clock of like 3.2ghz? Gonna need to do some reading today. How do I specifically know the difference when I change the speed shift setting? I'll see clocks jump quicker?

 

EPP=0 (or 0%) = max performance, clocks will spend a lot of time close to max available for all cores (I think 3.2GHz on your system, not 3.6GHz)

EPP=255 (or 100%) = max energy savings, clocks will spend a lot of time closer to 1GHz

A few guys noted EPP=127 (or 50%) does not allow clocks to ever hit max available so maybe something around 78 (or 30%) is a smarter choice. It allows you to use max clocks when necessary (for snapiness and executing tasks quickly saves energy per Uncle Webb); an EPP around 78 also reduces energy consumption when idle. Try 78 but you will need to experminent to find where that threshold is on your computer.

 

Is there any actual real world difference after enabling SpeedShift on a XPS 13 [8250u] ? For me, even after setting a value of around 175 the temps are still the same as with 0.

My laptop runs quite hot for some reason, touching around 90C at load after throttling and SpeedShift doesn't seem to make any difference either

 

From a thermal and performance perspective, SpeedShift is more efficient that SpeedStep or running with neither. SpeedShift allows corrs to quickly execute tasks and quickly return to low power sleeping states.

If you are looking to reduce power consumption or reduce heat, EPP closer to 255 will should run your CPU less aggressively. On my laptops, it will cap speeds to about 1GHz which should run cooler.

But the thermal impact of SpeedShift can "subtle" compared to a good undervolt or repaste.

90*c is hot. I don't know about the new XPS13 but Dell's thermal paste application generally has been poor on the recent XPS generations. Try an undervolt with ThrottleStop? Can you repaste the CPU?

 

But is this performance improvement actually perceivable in daily use or benchmarks? I don't mind turning it on, but I just hope there was a way to permanently keep it on than having to get it re-applied on restarts.

Yeah, I'm quite okay with the power consumption, I get crazy battery life on this any how and it doesn't quite make sense to cap a cpu which is designed to run at 3.4GHz to such low clock speeds. When I put it a more acceptable rate of ~175, its capped to around 2.9GHz but temperatures for me are still the same as the fans seem to come on only after the system is passing ~75-80C and it's later stuck at ~90C.

I'm not very interested in trying a repaste so early as I've heard that it's failed for a lot of people here. Undervolt yes, should try it, but I don't see it helping a whole lot. I generally like to keep my systems at around 60C [my PC], but this one seems to be hitting really high temps! Not sure if this is going to hurt the CPU in the long run

 

Figured it out after I had some time to read through the thread. Will do some tests this weekend.

 

I don't think casual users will notice increased performance with SpeedShift. Maybe longer battery life and slightly lower temps. Running EPP=255 is very laggy and very noticable. For my CPU-hungry, 40GB virtural pianos, latency performance improves with SpeedShift. A benchmark which runs CPU "flat out" all the time probably won't show improvement but that is not a realistic usage scenario.

On a desktop, disabling c-states, SpeedShift & SpeedStep should provide optimal performance. But on most laptops, that is running the CPU at 100% all of the time and creates massive heat so most laptop companies will disable turbo speeds if you disable c-states. Best performance for me on 9550 & 9560 is running SpeedShift with access to full turbo speeds.

There are a few ways documented in the early pages of this thread to enable SpeedShift.

Could you check the BIOS to see if Dell provided a SpeedShift switch? A prior XPS13 had that available. Switching it on might also open the windows>power options selection of EPP.

Alternatively, you can use ThrottleStop and have that launch every time you turn on your computer.

Alternatively, there are a few more technical ways documented in the thread but CAUTION as nobody has confirmend the 8th gen registers as far as I know.

There are more stats for the XPS 15 where you will see lots of successful thermal tweaks. On my 9550, improved temps on benchmarks are not atypical:

undervolt___-11*c
repaste_____-5*c
Total______ -16*c

The 7th & 8th generations will probably have less benefit from undervolt and more benefit from repaste. I would expect thermal improvement from SpeedShift to be signifincantly smaller than offered by undervolt & repaste but have not seen any testing there.

EDIT - simple undervolt, repaste, (plus a small registry tweak for TDP) stops all thermal throttling on my 6300HQ. More radical measures are required for the 6700HQ and 7700HQ.

EDIT2 - Install DELL COMMAND--POWER MANAGER and choose "thermal management-cool". That will run fans at higher speeds and may adjust processor via ways we can not normall access

Average lifespan of typical electronic components drops exponentially with a rise in heat. Whilst your CPU may be rated at say 100*C by Intel, it will last much longer on average with temps below 70*C. Regardless of what Intel or Dell or Apple are selling, they can't conquer basic physics...

 

Hey, first off, thanks for the elaborate response, really helpful!

There's no switch on my 9360-R for SpeedShift, only have the switch for SpeedStep in BIOS. I've turned SpeedShift ON now via ThrottleStop [EPP set to 128 - battery+power], however I'm not sure if Windows or another program is interfering with these settings as the laptop seems to be pinned at 2.4GHz on battery [50-60C] and is only switching to 3.4GHz on power while running the TS bench.

I've also made sure it starts automatically with Task Scheduler and made appropriate changes on ThrottleStop as well.

I've tried the Dell Power Manager settings and none of them seemed to help a lot, and the Cool mode throttles the CPU quite a bit it seems, which kind of invalidates it.

As for undervolting, after a couple of tries I've settled at -100mv undervolt on the CPU and cache, -50mv on GPU [can probably go lower on gpu but haven't bothered to try]. I think that's a pretty good undervolt considering my chip's average seems to be around 80-90mv. Now with the undervolt i do see slightly lesser temperatures on load and idle [~5C] with lesser throttling [3.1GHz+] on load.

I've got a Geekbench score of 13.8k and Cinebench score of 674 which are pretty good but temperatures do briefly touch the 90s. So overall, yes, undervolt has helped but I believe there's still scope for improvement with thermal paste being the main culprit.

Unfortunately I don't see many XPS 13 owners post their temps and clock speeds for me to figure out if this issue is a one off on my machine or if it's across all 8th gen XPS 13's.

 

Dell did not enable SpeedShift in 2018 XPS13 (9370) BIOS 1.1.3 (Intel 8250U).

ThrottleStop can enable SpeedShift on some 8250U processors so good chance it will work here on 9370 also (see below images which include SpeedShift enabled. That post and subsequent show efforts to boost TDP!)

 

My xps13 9370 has SST enabled

 

Which processor and which BIOS version are you running?

Where in BIOS do you select SpeedShift - same section as SpeedStep?

Thanks

 

It may be possible to mess with it using Dell Command stuff where support for SpeedShift was supposedly added for older models (not as a setting configurable in the BIOS menu)?

 

I didn't do anything its already enabled

 

Nice. Is there a switch in BIOS to enable and disable?

 

In the bios only option i see it to enable/disable Speedstep not speedshift

 

Good video! The funny thing is that Dell spoke so much about their new thermal design but the end goal seem to be the same? Temperatures are still as they were on the 9360. However the testing results expressed in the video was slightly off, as the temperatures on these machines don't hold at 90+, they just touch 90 after which the fans come on, throttling happens and settle in ~80 or below.

 

Hi, I have a 9350 and have not been able to use c-states since I got the laptop. It is enabled in the bios and in throttlestop, but the laptop does not use ANY c-states. As suggested in the very first post, I downloaded the latest Intel RST driver for the SATA controller, but that didn't work.

 

I have a 7700HQ in my Dell 7577 that won't go beyond C2 package state but I run in pure AHCI mode, so I don't have that problematic IDE driver pointed to as the culprit by @Che0063 . Is there a way to address this issue on my machine? I'm off to keep looking for some info. Thanks!

PS.) Anyone know the deepest C state a 7700HQ can go?

 

7700HQ should have an idle package power of less than 0.4W. I got my friends' laptops with 7700HQ down to 0.3W.

7700HQs can go down to C10, but manufacturers rarely do so. Most laptops (and I mean 99.999%) only go down to C7 or C8. C2 is terrible. Try to get the Samsung Nvme controller or something. The ACHI can still be a culprit. @pressing I'm planning to do a Windows Optimization guide to improve battetry life, which will include C state management. I'm hoping to get the complete guide done in a week's time, before school holidays start for me. Try enabling HIPM+DIPM to lowest, which enables HIPM, DIPM, and DevSleep if your device supports it. @Vasudev got quite a big battery life boost from that. I also have the PM961 Nvme SSD and I found I needed the Samsung NVME controller . I used to have a Samsung 850 Pro Sata, and that required either an Intel or Samsung controller.

 

Thank you for this information. I already have the Samsung NVMe driver installed actually. I may see if booting in Windows 7 presents this same problem. I got my stock 7577 tuned really nice today and hitting 9500 on FireStrike w/undervolt and OC on Win10 w/newest 1.4.2 BIOS and am happy with the high end of my performance but now I need to dial in this thing to sip power when it's off its leash.

Also when it is off its leash it throttles at 15W package power. I have been trying to study up on ThrottleStop so I can have a battery sipper profile using clocks between 0.8GHz and 1.9GHz ish. Now my only issue left to tackle is this lack of anything deeper than C2 package state.

 

The developer of ThrottleStop noted optimal battery life will happen with fast clocks; that way the CPU completes tasks quickly and can go into deep c-states quickly. So running at 0.8GHz may not be optimal for battery life. I think you should have full turbo speeds available.

As a first step, UncleWebb recommended focusing on the C0% (main screen of ThrottleStop). That should be less than 1% at idle with all programs closed. If not, find out what processes are out of control.

SpeedShift is a nifty power consumption feature; set EPP at 255 (or 100%) for max energy savings. Set around 78 (or 31%) for average power savings (this will allow max turbo clocks, your system might not be exactly 78 so you will need to check).

My Dell 9550 6300HQ now hits C8 both package and core; I'm pretty sure I did it on the 7700HQ also. It appears Intel permits C9 or C10 for Skylake & Kaby Lake, but due to either instability and/or lack of testing, Dell locked those out and I have not seen anyone hit above C8. Also note, once you are out of idle and doing light tasks, the Package C-States seem to drop out of C8 pretty quickly.

Check out the posts on c-states from the ThrottleStop Guide; start searching for related content from @Che0063

 

Yeah my profile clock speeds are subject to change but I want to experiment for times when my laptop sits idle while I read or peck some code out slowly. Iirc the sweet spot is more like 1.6 - 2.4 GHz ish from a study about power consumption vs compute speed. I will have to see if I can find it again or not.

When I get home I can share some screenshots of my setup with throttle stop, I'm using EPP and SST and getting good results. I wish there were more than 4 profile slots myself. At idle I have around 150 processes running. With task manager open I'm seeing 2% ish or less use which I consider pretty normal. I'm sure there is more fat I can trim but even with EPP at 255 I do not get anything beyond C2.

 

http://forum.notebookreview.com/threads/why-does-the-8250u-require-such-a-high-voltage.815154/

I took an hour of my life trying to find the optimal states already. It seems that the maximum turbo speeds are not optimal. When TurboBoost is in effect, the CPU rapidly ramps up its voltage resulting in a power consumption spike. And whilst I do believe what unclewebb says, I still believe it is better to have the EPP set to a more energy-saving-ish value, such as 180 or even 200. This is just what I've noticed in my testing. I have one profile that limits the clock to 1.8GHz, and that allows me to undervolt the CPU down to -200mV. TurboBoost requires a high voltage. Even then, on my laptop with the 8250U, the performance is slightly better than a 6500U.

 

Everyone's system and usage will be different. That is the challenge generalizing over a lot of CPUs, over several generations, with a lot of undocumented information, and laptop makers not 100% transparent on their tweaks.

Do that!

BTW - The Dell Command-Power Manager has access to some power registers and fan controls you might not otherwise be able to change. I suspect the Quiet setting would provide best battery performance but you need to read the descriptions and decide for yourself.

That is a monster undervolt for a newer generation CPU.

I got the 6300HQ to -170mv and the 7700HQ to -120mv with rock-solid stability. The first CPU is a lucky bin. When I was testing thermals for those laptops, undervolting made a massive difference in heat production, even an extra -10mv (so energy consumption). Cooler thermals mean less energy used by the fans also. The dGPU can be undervolted or disconnected on my laptops.

 

Well if I enable TurboBoost I can only have a -140mV undervolt.

 

Also, could you elaborate on this or point me in the right direction?



EDIT: (the results are coming in slowly):

https://www.tenforums.com/tutorials...er-management-power-options-windows-10-a.html

 

Host-Initiated Power Management (HIPM), Device-Initiated Power Management (DIPM), and Device Sleep (DevSleep/DevSlp) are power saving features of almost all modern hard drives. DevSleep is only available on SSDs, and generally puts the SSD into a sleep function than uses >5mW. @Vasudev got quite a bit of battery boost from this. By default, Windows only enables Host-Initiated Power Management. To change this;

Win+R > Regedit > HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\0012ee47-9041-4b5d-9b77-535fba8b1442\0b2d69d7-a2a1-449c-9680-f91c70521c60

There should be a Description that says ‘Configures the LPM state’, as well as an ‘Attributes’ DWORD with the value 1. Change ‘Attributes’ to 2


Now go to Control Panel > Hardware and Sound > Power Options > Change plan settings (of your active power plan) > Change advanced Power Settings > Hard disk > ACHI Link power Management and set it to Lowest. If that is not an option, change it to HIPM+DIPM

As a side effect, on some systems, such as my older Aspire V Nitro, enabling DevSleep and DIPM allowed the Package C State to enter C3, using 2.1W of power on idle. Previously it idled at C2, using 2.4W. Modern mobile CPUs should idle at less than 0.5W, however. (Tested on U, Y, and HQ CPUs) More about C states below.

That's directly from the guide I'm writing.

 

Yeah I think mine likes to stay around 2.7W at idle. Would love to get it to deeper package states. The cores do read high C7 values most of the time tho.

 

no no no 2.7W is way too high. the cores may be in C7, but your uncore (cache, etc, things that handle the other system components) will be in something like C2/C3.

Ugh lord knows what happened to my avatar. I don't have the time to fix it lol.

 

At idle, mine does 0.2-0.6W on C6.
For those who want to get advanced power options in W10 and fine grained control run this .cmd script https://gist.github.com/Nt-gm79sp/1f8ea2c2869b988e88b4fbc183731693

 

Thanks, just turned on all the extra options. Gotta read up on how exactly to set it all still.

 

I just left it at defaults and simply installed Power Plan Switcher UWP app to switch between Balanced/High Perf and Powersaver modes based on AC source.

 

Well, I'm not 100% sure what exactly it was, but I was in fact able to set my AHCI link to the "Lowest" setting (in the Advanced Power Options for Win10) which supposedly includes the DevSleep along with the HIPM+DIPM and restarted. I was getting wattages around 1.2W - 2.4W so I proceeded to end task on anything extra I didn't absolutely need (except for TS itself) and noticed I was finally getting <1W readings so I clicked the C7 button on the main TS window and sure enough my Package States are reading correctly now with C7 sitting around 80% at idle. My TS Package Power now reads 0.3W to 0.4W at idle.

 

In my opinion, there should be 0 background tasks. Great to hear you optimized Windows so much. So few people know nor care about this

 

The only extra things running in the background for me are programs I want running that I installed myself because I use them. They are all solid, but contribute to keeping the CPU from being able to sleep as deeply as I would like.

OTOH keeping TS set to my 3.4 - 3.8GHz profile with SpeedShift set to 0 finally made my laptop feel as fast and responsive as I expected it to be since day 1.

I always spend time tuning my OS and fixing issues that aren't working right.

 

Are you recommending 50% on A/C and 65% on battery? Or just that you were able to reach max CPU speed on either settings?

 

Battery is 50 and plugged in is 40 now and tbh I can't remember any details why as that was a long time ago

 

This Intel document will help you understand some of those registers. Some are rather esoteric. Some are rather subtle. Some will not work without additional user tweaks (eg speed shift).

http://www.intel.com/content/dam/ww...eveloper-system-programming-manual-325384.pdf

I think you had battery EPP at 50% (128) as that is Intel's general recommended SpeedShift setting for "average" performance.

I think you had plugged in EPP at 40% (~101) as that was the highest EPP you could reach max turbo clocks.

 

That is both an awesome PDF and somewhat terrifying. Who the hell writes that stuff and moreso who the hell actually uses or knows how to use it all? Lol, could you imagine if everyone who wanted to write a program had to understand and read all of that?! Haha.

PS.) It looks like it only goes up to 6th gen CPUs... I wonder how many changes there were from 6G to 7G. I was under the impression they are quite similar architecture wise.

 

Makes sense

 

Sorry for being lazy last night, but my brain was mostly off duty already. Anyways, just double checked and changing that setting from HIPM to Lowest absolutely was the cause of my troubles preventing sleep deeper than C2. I just tried toggling back and forth in real time and TS responds accordingly (without needing to reboot Windows).

Side note, looks like NBR is still having image uploading issues for me...

Here are the pics externally:

https://mega.nz/#!YSZ2EboK!WqZb7O1bKEr95rQODx-5DH1N4Qn4LKpOjs3EVgXbI1A

https://mega.nz/#!wO4BjJbL!olUgJjvNhgJxKuHOroqAgjo95imDFakBqJ481MPXJKs

 

Btw, just curious how many of you guys have seen this Dell-related post:

http://forum.notebookreview.com/threads/dell-inspirion-7577.808514/page-118#post-10707389

 

This laptop would melt if you upped the tdp!


And why did you upload pics to mega.nz, it doesn't work as it is file hosting not image.
use imgbb, imgur, flickr if you need hosting m8.

 

Yeah yeah I know... It's just where I collect all my Dell7577 files in one place for easy sharing.