Contribute
Register

[Guide] Dell XPS 13 9360 on MacOS Sierra 10.12.x - LTS (Long-Term Support) Guide

Status
Not open for further replies.
Excellent find - VoltageShift is a safer (non-permanent) way to test UV settings. It correctly lists the already-undervolted components and provides a continuous monitoring option

I've also found another component to undervolt (Analog I/O) which seems to be working well (though unsure what the real-world difference will be). Still, every little (undervolting) helps....

What are your current successful undervolt values for CPU & GPU?
 
-75mV is stable on high loads for both CPU & GPU. This is for the i7-7500u
What are you seeing on Coffee Lake?
 
I'm pretty sure voltageshift is not working, or maybe because the bios needs to allow overclocking first?

Code:
------------------------------------------------------
   VoltageShift Info Tool
------------------------------------------------------
CPU voltage offset: -500mv
GPU voltage offset: -100mv
CPU Cache voltage offset: 0mv
System Agency offset: 0mv
Analogy I/O: 0mv
OC mailbox cmd failed
Digital I/O: 0mv
CPU BaseFreq: 2000, CPU MaxFreq(1/2/4): 4000/4000/3700 (mhz)
CPU Freq: 0.8ghz, Voltage: 0.5704v, Power:pkg 3.70w /core 0.89w,Temp: 47 c
 
I'm pretty sure voltageshift is not working, or maybe because the bios needs to allow overclocking first?

Code:
------------------------------------------------------
   VoltageShift Info Tool
------------------------------------------------------
CPU voltage offset: -500mv
GPU voltage offset: -100mv
CPU Cache voltage offset: 0mv
System Agency offset: 0mv
Analogy I/O: 0mv
OC mailbox cmd failed
Digital I/O: 0mv
CPU BaseFreq: 2000, CPU MaxFreq(1/2/4): 4000/4000/3700 (mhz)
CPU Freq: 0.8ghz, Voltage: 0.5704v, Power:pkg 3.70w /core 0.89w,Temp: 47 c
I've been using voltageshift for some time and it appears to work properly. Currently am stable at -120 -50 -120 on i7-8550u, with anything higher causing crashes during geekbench or prime95 runs. Validated that I see the same behavior in Windows using throttlestop with similar testing. With voltageshift, immediate change is achieved with offset but is not persistent, good for finding your specific CPUs limits.
 
I've been using voltageshift for some time and it appears to work properly. Currently am stable at -120 -50 -120 on i7-8550u, with anything higher causing crashes during geekbench or prime95 runs. Validated that I see the same behavior in Windows using throttlestop with similar testing. With voltageshift, immediate change is achieved with offset but is not persistent, good for finding your specific CPUs limits.

And you haven't set any BIOS UEFI vars for unlocking? Its just weird that voltageshift allows me to set -500 (with the --damage parameter) and the PC works fine. I would expect it to hang, hence I believe it's not working.

Also I don't see the CPU voltage drop when doing this.
 
And you haven't set any BIOS UEFI vars for unlocking? Its just weird that voltageshift allows me to set -500 (with the --damage parameter) and the PC works fine. I would expect it to hang, hence I believe it's not working.

Also I don't see the CPU voltage drop when doing this.
The only bios changes I have in place are from your git repo :) I also didn't see an appreciable voltage drop, but at load I did see less severe throttling (2.5ghz vs 2.2 without voltageshift)
 
Surely you need to enable overclocking first, as per my settings. Otherwise the settings may appear to be passed to the MSR register but will be blocked. I've pasted the relevant excerpts here:

Code:
*Overclock, CFG, WDT & XTU enable
(0x4DE) -> 00
(0x64D) -> 01
(0x64E) -> 01
 
Surely you need to enable overclocking first, as per my settings. Otherwise the settings may appear to be passed to the MSR register but will be blocked. I've pasted the relevant excerpts here:

Code:
*Overclock, CFG, WDT & XTU enable
(0x4DE) -> 00
(0x64D) -> 01
(0x64E) -> 01

This is correct, I installed XTU on Windows to Go USB to see what was happening.
Interestingly enough XTU changed the 0x653 variables by itself.

Some testing learns that the machine is stable (UEFI var configurations) with the following:
  • CPU, -100mv (0x64 for 0x653)
  • GPU, -30mv (0x1E for 0x85A)
With -120mv, Windows would bluescreen. With -150mv the laptop goes into orange / white flashing LED error mode.
GPU at -50mv causes Windows to boot to a black screen. -30mv seems stable, but I haven't been able to use the XTU GPU stress test, only the CPU one.

In Voltageshift this reflects as follows:
Code:
------------------------------------------------------
   VoltageShift Info Tool
------------------------------------------------------
CPU voltage offset: -102mv
GPU voltage offset: -31mv
CPU Cache voltage offset: -102mv
 
Yep those numbers seem more in line with what I'd expect.

With regards to stress testing - for GPU use furmark in windows (or linux), and CPU use Prime95. Protip: If you really want to fine-tune your stress test (and make a bit of money on the side), hand-compile cpuminer and gpuminer respectively and baseline the respective hash rates.
 
Yep those numbers seem more in line with what I'd expect.

With regards to stress testing - for GPU use furmark in windows (or linux), and CPU use Prime95. Protip: If you really want to fine-tune your stress test (and make a bit of money on the side), hand-compile cpuminer and gpuminer respectively and baseline the respective hash rates.
Not sure what to tell you guys. With only the efi updates from darkvoids repo, I am able to run stable at -120 -50 -120 via voltageshift(same values as in windows using throttlestop and realbench testing). I am mostly stable in macos at -130 -70 -130 as well, but since that failed in realbench on windows i went with the highest stable settings from there.

Your comments did make me curious if the voltageshift values weren't actually applying in macos, so I just tried bumping offset to -150 -80 -150 and immediately locked my system ¯\_(ツ)_/¯

As for the actual offsets, I've tested on about 5 9360s at this point and have seen the stable values all over the map with older builds maxing out much closer to what darkvoid is reporting. I'm happy to run any other tests you guys would like as I've been very happy with voltageshift so far.

Thanks as always for the awesome guide and work guys.
 

Attachments

  • IMG_20180721_024424.jpg
    IMG_20180721_024424.jpg
    5.5 MB · Views: 284
Last edited:
Status
Not open for further replies.
Back
Top