GS63VR's i7-6700HQ running slower after undervolt?

Hey!

So, I decided to get a refurbished Kaby Lake MSI GS63VR since it was about half the price of a Gigabyte Aero 15 or any fancy new laptop. I'm really happy with the product, I don't think the screen is nearly as bad as I've heard it is in the -001 model (maybe that's what is refurbished?) and so far the battery seems petty okay for non-gaming use. (~4 hours of web browsing on wifi while downloading a file and/or watching a YouTube video in 1080p.)

I want to get under the hood and tune this thing as much as I can in order to get the very most out of it. So, last night I started running a benchmark to get some numbers in before I changed anything. I ran Dolphin's benchmark 5 times on battery power (without needing to recharge, mind) and then I ran it 5 more times plugged in with the AC adapter. After getting those numbers, I undervolted the CPU -0.010v and tried running Dolphin 5 more times.

I learned a few things here:

• Temperatures go up when plugged in with the AC adapter. I thought this was really strange, but then I decided it must have something to do with 'permanently' turning the GPU on while plugged in. Since the GPU and CPU share heatpipes, it would make sense that my CPU performance wouldn't change in spite of the temperatures rising a little bit.
• While running Dolphin's benchmark (which I know isn't the most intense test), the fans don't get so loud that it was distracting from watching a TV show on a TV that was across my living room from me.
• And... with a -0.010v undervolt, I saw a pretty significant performance hit. To my understanding, changing the voltage shouldn't change the CPU's performance at all, but this was a drastic change.

My worst score while at the stock voltage was 467 seconds, my worst score at -0.010 was 585 seconds. That's about 25% slower!

My best score while at stock voltage was 442 seconds, my best score at -0.010 was 539. ~22% slower.

Is there some automatic tuning that goes on in the i7-6700HQ based on voltage? Is there a way to turn it off? My thermals did improve by a few degrees C, but I've seen those results on desktop CPUs without changing my performance at all with an undervolt.

I used Intel Xtreme Tuning to make these changes on this laptop, historically I've mostly used (desktop) motherboards' BIOS.

 

-0.010v undervolt?
Um....
Do you mean -100mv?
or -10mv?

I think you mean -0.1v
At 1.10v, a 0.010v undervolt makes this 1.09v...that's nothing.

1.10v with -0.1 undervolt is 1.0v.

You also should undervolt the cache as well as the CPU.
And are you doing this on battery power?
Not sure if you should undervolt on battery, as I think idle /power state voltages on battery are lower than on AC....but maybe I'm being dumb...

 

If I'm being completely honest, I don't know what I mean by "-0.010v", but this is what I changed in XTU:

 

Yeah, do -100mv (or -0.1v).

 

I'll try that in a bit, but should that really have anything to do with the 20+% drop in performance I'm having?

 

I don't know. Maybe power limit. I don't know how many threads Dolphin uses. And did you see the loss of performance on AC or battery? You didn't clarify which.
HQ processors will throttle if they go above 45W. No way to avoid it.

 

I'm almost done with my lunch and I'll head over to the laptop. (Also waiting on a download on it to finish.)

Before undervolting, I tested both on AC power and on battery power, unplugged. After undervolting, I tested only off of battery power.

Before the undervolt, I noticed no appreciable difference between benchmarks. The temperatures were ~3C higher when plugged in to AC power, but I chalked that up to the GPU being more active when the system is plugged in. (The CPU and GPU share heat pipes.) Here's a google sheet where I was recording my tests.

 

It's possible heat had built up from the first two runs of tests, which meant the undervolted testing was starting from a "pre-heated" state.

 

I tried to account for that by waiting a few minutes for the laptop to return to ~35C on the cores between tests. I just realized I'm running on the "Balanced" power plan and I should definitely be doing my initial testing on High Performance. Some time this weekend I'll get around to re-doing my testing on High Performance.

There definitely might have been some sort of anomaly, though. I got to do my five runs at -0.010v with AC power connected and my results were much closer to my runs before the undervolt and they were much more consistent. The temperatures were a lot closer to my (lower) battery temperatures at stock voltages than my (higher) AC temperatures at stock voltages.

Perhaps Windows update or something silly like that messed up my first set of undervolted results.

 

Try using Throttlestop instead of XTU. When i used XTU i found it used more resources than throttlestop, which means it takes resources away from Dolphin, which may be the issue here.

 

I'll keep testing with XTU for now and then later tonight I'll see if I have better results with Throttlestop.

While we're on the subject, MSI's software sucks total booty... I tried setting a custom fan profile and it didn't seem to change anything. Is there another program I can try to do the same thing? I tried Speedfan, but it couldn't detect any of the system's fans.

 

HWinfo seems to work for me across many systems, so try that out. However, i haven't been able to change fan speeds for the GPU fan in HWinfo, so keep that in mind

 

GPU fan speeds only work in HWinfo on a certain laptop. I think it was Asus. Might have been alienware but I'm pretty sure it was Asus only.

 

MSi software is only good for triggering max fanspeed.


Auto works fine for most situations if you're not gaming.

 

Ugh, I was working on a long post and somehow it got tossed. Anyways...

I didn't see anywhere in HWiNFO64 to control my fans, and I couldn't make sense of Throttlestop as there wasn't an option that simply said "Voltage Offset."

I've updated the spreadsheet to where I am with my testing so far. It seems like my original performance hit was some sort of anomaly which is either Google Sheets somehow distracting the CPU enough to cause that hit or something more discrete like Windows Update sneaking around in the back.

That said, the new curiosity is that I can't seem to get to the lower 460 times (or even lower) like I originally did on the Balanced power plan... I'm using High Performance and seem to be getting lower scores consistently.

 

Okay, so those unusually high temperatures haven't come back in all of my testing. The dropped speeds had to have been some strange anomaly.

I still haven't given Throttlestop a shot because I haven't taken the time to digest its interface, but my system seems stable at a -0.160V undervolt. At -0.160V, I'm able to run Dolphin's benchmark and PCMark's 'Creative Workflow' benchmark without crashing, and I can't imagine I'd do anything more strenuous than that.

 

Bear in mind that an undervolt that's stable under load may not be at idle. Further testing may be needed (or if you start seeing freezes or BSODs, back the undervolt off by a hair).


Sent from my iPhone using Tapatalk

 

For anyone interested, I reapplied thermal paste on the CPU and GPU. This is a refurbished unit and the paste looked pretty bad...

After going from the 'stock'/refurbished paste to Thermal Grizzly, my CPU temperatures under Dolphin benchmark have dropped at least 6C. I think this is as much due to the nicer paste as it is to me probably doing a better job applying the paste in general. With the new paste and on battery power, even with my undervolt, I got my lowest Dolphin benchmark score yet. I did this benchmark in a cafe so I can't guarantee that the ambient temperatures aren't drastically lower (???) but if I confirm that is the case I'll post here again.

 

When you say "Lowest sore" is that good or bad?
Usually benchmark scores= higher=better.

Unless you're talking about seconds to complete.
If you mean you're getting better scores now after repaste, then good job. Kryonaut works extremely well and should be quite durable up to 80C, although it's guaranteed to work up to PROCHOT (even beyond that; it isn't just intended for CPU's only).

 

It's time to completion, lower is better.