*** The Official MSI GS65 Stealth Owners and Discussions Lounge ***

My Aida trial ended a couple days ago. Also, I know if I lower the cache past 180, I start to get crashes. -280 is gonna be an instant bsod. Do you have a different program you want me to try? I can post average temps from prime 95. I’ll put them up tomorrow, I gotta go to bed

 

Yeah. Run it exactly as the guide mentioned in my post but switch to P95 and Heaven. Just follow the guide

Edit. Maybe I missed info due I’m on my phone. People most part usually undervolt Cashe higher than the Core domain. Aka the opposite vs. you

 

Just for grins, you might try disabling ThrottleStop and see if you can run the same undervolt settings in XTU. I'm wondering if something else in TS is affecting this undervolt range, as it's a bit unusual to put it mildly.

Also there were BIOS hacks that skewed the undervolt / overvolt settings quite a bit, you didn't happen to play with that?

I guess this could be a new batch of laptops with even higher BIOS voltage settings in the BIOS on the CPU, which allow for more Core undervolt range, but that would show up for others too if they updated their BIOS - what rev # is it?

There have been other -170mV to -180mV reports, but usually they edge downward after longer stability testing - and after running a variety of apps and games.

It's quite exciting

It would be interesting to get some comparisons on power, thermals, performance differences between stock and your final best tunings to see how much cooling effect this gives. @ -100mV at stock -10c is the usual reduction at 100% all Core CPU load, so this large undervolt might be measurably more.

So you mentioned someone else was getting undervolt in this range too? Did you sync up your TS / BIOS settings so you are playing by the same "rules" in the same environment? Again it would be interesting to see XTU running with the same settings on their laptop too for comparison - while TS is disabled or uninstalled so it's other effects aren't in play.

Thanks for all the updates, I look forward to your posts.

 

Interesting, I would have expected TS to block this differential setting then, I wonder if he see's the same thing in XTU.

@Buffalord try syncing up the cache/core in TS then, and/or try XTU and see what that offers, try the multiplier settings in XTU while you are there too.

 

Thanks for posting this mate

 

I called one of the MSI resellers in Canada. They said warranty voided if the GS65 is opened up - whether to add more ram, 2nd SSD, etc

How are you guys dealing with this?

I am interested in the laptop but warranty is important for me in case of motherboard issues, etc.

Thanks

 

What shift do you guys use when undervolting and gaming?

 

It's not up to the reseller, you can call MSI directly to confirm.

 

running heaven and prime 95 at the same time is much less stable than running realbench

 

Odd. While I don’t see major temp differences when cpu and gpu at 100% for both combinations, the -300 one gets bsods when running firestrike and heaven together

 

Although I cannot complete the test, the -330 one shows less throttling and throttles later than the -175 one. I cannot complete the test, because although realbench and firestrike are fine, I get bsods when running prime 95 small ftt and heaven benchmark together

 

-330mv vs. -175mv undervolt should in reality means a lot lower temp and Cpu package power. The voltage Settings (-330/185mv) you use don’t works as you think they do.

+rep for trying

 

It’s strange. If I watch the voltage I’d in throttlestop, I can see it dropping. It dips into the high .5’s on -330, and only goes as low as .62 on -180. And running firestrike, I see 3C temp drops, which, I agree, are not as lowas I expected, but there is still a difference. I think I’m just going to leave it at -180, -180, as I get bsods in rare conditions if I raise the core higher, and bsods commonly if I raise the cache higher. Also, is it possible to undervolt my graphics card overclock? I tried the traditional method, overclock, then flatten curve, but I didn’t notice a difference

 

So the most stable UV you got is ~ -175mv?

What temps and CPU clock speeds do you maintain (lets say after about 10 - 15 min) with that UV in prime95 + heaven?

I presume that the CPU package power doesn't fall below 45w during that combined (CPU+GPU) test?

 

Edit: never mind, did some more testing, at max temps, they’re equal. That was quite the rabbit hole. Any idea on my how to undervolt an overclocked gpu question?

 

As expected But here is an guide for you for the graphics http://forum.notebookreview.com/thr...rvolting-overclocking-and-overvolting.817033/

 

Thanks for the guide man! There goes about 12 hours of my lif,e but at least he mystery was solved
Edit: hmm. It seems I can undervolt from stock, which results in modestly improved thermals, or do a regular overclock, which results in no change in thermals but much higher rpeformance (about 1000 points in firestrike). I can’t seem to undervolt an overclock. Any idea which I should go with?

 

Wierd, there seems to be no temp difference between my undervolt and my overclock

 

If your target was to have considerably lower gpu temps, then I don't think that you'll achieve that.

By raising the clocks for the same voltage points, you're just allowing the gpu to maintain higher clocks at the same power draw.

You could try to setup the curve so it has higher clock speeds (than the default ones) at the same voltage points), but you flatten the curve before it reaches higher clocks.

Try to experiment more with the voltage curve and as long as the graphics score is at least the same or higher, then the clock to voltage ratio is good.

If you go too low, you can bump into the pascal problem, where your clocks are maintained at a high level, but the performance (the FPS and scores) are considerably lower.

 

-180 each. Running small ftt on prime 95 and unreal engine for 30 minutes, average cpu temps are 88 degrees, with one core throttling slightly, in comparison to everything throttling. If I dropped the core clock from 36 to 35, I don’t think anything would throttle

 

For 45w 35x is the sweet spot for sustained load, that's where I have a custom setting for batch jobs on 45w CPU's and it gets the best undervolt and performance.

Interesting that TS let you put in a different undervolt for cache / core, and weirder it picked the cache undervolt of -180 as the one to use vs -330 for core as you set them.

I wonder if TS isn't internally averaging them or some other side effect of balancing having different values in each core / cache setting, even though what ever undervolt is set by TS is lock in sync to both cache and core once applied to the CPU.

I wonder how the XTU interface would treat this on the 8th gen CPU's?

It's not so much a waste of time as a learning experience - we've all done similar at some time or another. At least you know for next time.

You jumped well ahead of comments there. You undervolt / OC each - CPU and GPU - independently first, and then when doing combined tests they are fighting for internal power distribution.

You need to pick one or the other OC - CPU or GPU - to maintain it's OC and sacrifice the other to free up power budget.

So if your primary goal is CPU OC, reduce the GPU OC, or in this case reduce the undervolt to reduce GPU performance. Or reduce the CPU OC to keep the GPU performance.

It's not a surprise that a thin and light laptop won't have power budget to support full performance on both CPU and GPU at the same time.

I didn't notice, did you undervolt or tune the iGPU as well? I don't buy Optimus / iGPU enabled laptops so IDK how it currently is set up to support tuning iGPU, but if you lock everything you can to dGPU only (Windows and Windows apps are locked to iGPU) there might be some power budget you can take away from iGPU as well.

If you can spare the time, it would be great to get a summary for tuning on the GS65 after you are done, including fan curve + RTSS and whatever else you find to tune for better cooling and performance.

Thanks again for posting your tuning experience along the way.

 

Why is 35 a sweet spot?

 

All core 35x at 100% draws ~45w on all the 45w laptop CPU's I've tried (4c/8t). You could try to edge out a little more performance @36x, but the frequency drop usually negates any performance over time. Or you could try 34x and get a little more undervolt. All fun stuff to try out on each new CPU.

Depends on the CPU, cooling, and work load, but try it for yourself and watch the CPU package power draw with hwinfo64, here's an example




Edit: @Buffalord knows he has a 6c/12t CPU, I didn't need to tell him that, but I did add 4c/8t to my description above of what I've tested.

Let's see what he comes back with before getting critical and telling him what he can't do. Being positive about trying different things when helping someone just learning and exploring tuning for the first time is important. He needs to feel empowered to try settings higher than possible as well as lower than possible to see the range of results for power, thermal, and performance.

 

Why do you want to draw exactly 45 Watts?

 

Your locked CPU will hit higher power draw for the Power Timeout Window setting, in seconds, usually a limit of up to 96 seconds at most, but has been as low as 8 secs to 28 seconds, but after that timeout it is limited back to 45w and the CPU can over correct and drop the core multipliers well below 35x, averaging to less performance over time than just locking it to 35x to start.

It's best to start at and hold 35x for long running jobs - longer than Power Timeout Window (name has varied over time). Check out the XTU settings in Yellow for settings. Each CPU and CPU generation can have different names for options and can have different settings locked or unlocked, or not available for change - that's why I suggested comparing what is available in XTU for what is there available for change or missing compared to TS.

For unlocked CPU's like the HK models you can set to max power available, and even unlock maximum power limits if you are willing to hack the BIOS - but usually the power available is limited anyway to near the BIOS set limits.

Locked CPU's have the same goals as unlocked, tuning CPU voltage to minimum, but unlocked CPU's have more knobs to twiddle.

 

So in applications where the cpu doesn’t hit 100% or close to it, is there a benefit of using 35 vs 36, other than about 2 degrees celsius and battery?

 

For job mixes that vary, it's "best" to set for maximum OC available under thermal throttle point - and run with fan at 100% if environment allows (you and others in the room have headphones), and let the CPU handle the on the fly load tuning.

On Windows Balanced power plan with Advanced CPU performance set to 0% minimum and 100% maximum to allow the CPU to adjust CPU frequency as load demands, allowing the idle time between load demands to cool the CPU down.

You could also live on 35x/36x with more undervolt and cooler running on average, probably won't be too noticeable in games or most apps, but will show lower maximum performance test numbers - numbers don't mean much unless for bragging rights - and for getting to know the limits of your hardware, which helps in knowing how to tune for your special use case(s) vs every day use.

Also of note, on thin laptops it's popular to set Maximum CPU performance setting in Balanced to 99% which disables Frequency Boost completely, setting maximum frequency to Base Clock. That's a bit too much of a "de-tune" for me, but some people seem to enjoy it. It can reduce temps enough to allow auto fan to work well enough, and keep the fan noise down. For thin laptops without enough cooling for sustained high performance this can be a useful setting.

Lots of setting variances to test and try, to get to know new CPU's and laptop limitations.

 

Can’t compare (locked down) 4 cores vs. 6 cores Coffee. Run 6 cores Coffee chips on 4 cores and it will use a bit more juice vs. due more Cashe.

Btw. Awful reading

 

I called MSI Canada but the guy I spoke to said he couldn't give me warranty coverage info without laptop serial number. I told him I couldn't buy without knowing warranty info Catch 22.

 

hmscott

From what I've seen so far, the sustained clocks for 8750H in continuous 45w loads, are about 3.0 to 3.2Ghz (this is with an UV, mostly higher negative values than -100mv, without UV, I've seen the clocks drop to 2.5Ghz) that's why I was puzzled when I saw he mentioned 3.6Ghz, but with one core that was slightly lower probably due to thermal throttling (I remember a previous post where it is mentioned that one core gets hotter).

Papusan

That's why I've expected lower sustained clocks (than the 4 core CPUs) at 45w, since these coffee lake CPUs don't have a noticeable increase in efficiency over kaby lake CPUs (CFL is mostly just KBL with 2 more cores and higher cache) or even compared to the broadwell CPUs such as the 5950HQ from the above example.

He probably has a higher UV value than the avr. for most 8750H users, but I don't think that alone could be enough to allow sustained 3.6Ghz clocks at 45w (it's also unlikely that he has a golden 8750H).

Something doesn't fit properly in this scenario (or is it just me missing something or wrongly interpreting data?)!

 

You can disable 2 cores but you can’t disable the higher Cashe level and run it exactly as with previous gen 4 cores chips. Already tested. Use more Power. And there ain’t any efficiency over previous chips family. The results is put in Stone.

Edit. Only lower voltage aka better undervolt and lower heat for Coffee can make them more similar.

 

Yep, that's what I thought too.

Also, to maintain similar clocks under the same circumstances, he had to disable 2 cores and have a better voltage to clock ratio than the 4 core CPU.

Don't know if it's possible to have similar clocks with all 6 cores active and a considerably lower (as in a higher negative value) undervolt?

I mean, how much lower does that UV have to be to compensate for the increase power draw of the additional cores + higher cache?

Regarding the heat, a difference (as in lower temperatures) of about 15C (or a higher value) is probably needed in order to have a perceptible difference in efficiency.

 

@Buffalord knows he has a 6c/12t CPU, I didn't need to tell him that, but I did add 4c/8t to my description of what I've tested to my previous post - he can't assume that - I need to tell him.

Let's see what he comes back with before getting critical and telling him what he can't do. Being positive about trying different things when helping someone just learning and exploring tuning for the first time is important. He needs to feel empowered to try settings higher than possible as well as lower than possible to see the range of results for power, thermal, and performance.

3.6ghz on a new process with higher undervolt even on 6 cores might be possible, if that is what he is seeing I can believe it. It is also dependent on workload.

That's why 35x on several generations older Broadwell process was the setting to stabilize right at 45w - check the hwinfo I posted, it's actually well over 45w after 2 hours running - and that 5950HQ CPU had an extra 128MB of L4 cache, which would take more power as well.

I think we need more users testing and posting that have the 8750H and tune for best undervolt, which is getting better as well, over -150mV -> -180mV +, then we can know what the right all core multiplier will be to sustain frequency at constant 45w, 32x-36x, or?.

 

I agree that we need to see more test under the same circumstances to draw a better conclusion, but even if that's what really happened during that previous test (the UV was high enough to allow a 6 core CPU to maintain slightly higher clocks (3.6Ghz for 5 cores and one core slightly slower) than a 4 core at 3.5Ghz, at the same power draw (45w, those 0.251w could be included in the margin of error), then I would surrey consider that to be one of the exceptions (but yeah, so does having very high and stable UV).

I'm in no way criticizing the user, I just think that it's possible that we don't have enough data on that equation and it is possible that there is a misintepretation somewhere.

 

@Buffalord - I assumed you would do this anyway when you were ready - you're already moving pretty fast - but at some point please post some screen shots of 100% all core in progress tests of hwinfo64 for CPU info, showing multiplier, temperature, power, similar to what I posted earlier for the 5950HQ @ 35x running that FAH batch program - which is another good test for CPU and GPU load testing BTW.

When / if you get time to do some fine tuning as @raz8020 suggests, running 4 or 5 cores at a higher multiplier but de-tuning 1 or 2 cores to limit power to just over 45w to allow the clocks to sustain, that would be cool too.

I would have been happy to slowly reach these ideas and suggestions as you report progress, but the rest of the thread is itching to see if their guesses are "right". Some people just gotta know right now.

Take your time and have fun, if it gets to be too tedious take a break, I'm fine taking it slow.

 

Canada has the same laws as the US when it comes to warranties. MSI can't void your warranty for upgrades as long as you don't damage anything while doing the upgrade. Obviously they won't cover the upgraded components though.

 

Call the toll-free support number, they don't need serial number to tell you what kind of warranty they have.
Also, the warranty policy is pretty much the same for the entire North America region.

 

Can anyone with Ubuntu installed on their GS65 send me a pm? I'm having some issues with the wifi hardware when waking up from suspend. Thanks.

 

No one can fix this. Contact MSI.
This bug existed in windows on their wifi adapters on their Titan laptops for months (or years), and was only recently fixed in a Bios update.
Sounds like the bugfix was only tested in windows, and not in Linux.

Good luck actually reaching someone at MSI who will both listen AND care AND who will develop a fix for you. If they actually send -you- a fix bios compile directly, I will eat crow. Because I KNOW they won't.

 

I've bee doing this for like, 3 days straight for 8 hours a day, but I'm game for one more test. What kind of performance are you expecting, though? I get more fire strike from 36 vs 35. Are you expecting to see no down clocking?

 

https://imgur.com/a/qKdLTpg thats a run of realbench for ten minutes at 3.6 ghz, max power draw, according to throttlestip, was 44.4 W
https://imgur.com/a/iLI2rXF thats a run at 3.5 ghz, max power draw was 42.9 W

holy ****! So I dropped the undervolt to 340 for ****s and giggles, and there was a 5C drop in temperatures at max load. The voltages on the cores were ~.900, insteat of ~.950. Of note, however, this did crash when I tried to turn off realbench But it does seem that cache and core are seperate. Also, I'm pretty sure the crash only happens when gpu and cpu are being stressed at the same time

Edit: I found something... odd. Under a CPU only load, the change in core past -180 (where the cache is) makes not difference. no matter the under volt, the voltage stays around .9 per core. BUT, if the CPU and GPU are being stressed at the same time, there is a .050 difference in voltage on each core, and the average temps are 5C lower! Under gaming circumstances, however, this makes no difference. It only seems to matter when GPU and CPU are both being heavily stressed. It might make some difference in well optimized games that use most of the GPU and ~60% of the CPU. Can someone else verify if this is a thing?

 

Just to clarify, do you have Ubuntu installed on your laptop? Or are you just referencing issues with older models?

 

No and it's irrelevant whether I do or not anyway. I've only booted to ubuntu on a flash drive, not as an installation.
The bug about the wifi adapter disappearing when you wake up from sleep was a longstanding bug on the GT73VR that was only fixed a year later after people kept complaining about it.
The fact that you experience the same bug on a newer model means that their fix clearly only worked in windows.

 

About the last part first, I think it's a matter of the system not having enough power to feed 100% load to both CPU and GPU at current settings. There wouldn't be a difference in support for CPU settings split cache / core voltage adjustments, it either works or it doesn't, but it sounds like it doesn't and you are seeing coincidental effects, not actual effects.

Core 2 jumped up to 41x and just over 90c thermal throttle point, probably just for an instant, and some other cores hit 39x. Not sure if it's due to your CPU not supporting multiplier setting for cores - TS could be taking the setting but not affecting the CPU. If you set 36x on all cores then the other cores showing higher multipliers / frequency 39x / 41x shouldn't be happening.

And, it could be that you are making changes on the fly and not restarting / resetting hwinfo64 so min / max / averages for multiplier settings aren't pure to the current setting. There's an Icon on the bottom of the hwinfo sensor dialog, hover over the icons to get a tooltip text describing function, you click on Reset to Zero out all the readings after making a change so it clears the average, maximum, minimum - do a reset or exit and restart hwinfo64 after making any setting changes to get pure readings from that point forward.

You'd need to set High Performance power plan, Advanced Settings => CPU Performance => 100%/100% and set the cores multipliers so that the frequency holds during load, idle, and in between, like I said above we can step through a few settings and tests for those settings to get it sorted out outside this thread then come back when it's sorted.

Tuning and testing is an ongoing thing, first to get sorted on all the details, then time to test things, then over time as things occur to you to test or you see behavior in apps / games that make you go hmmm. You are closer but you are still going to need some ramp up time. Perfectly normal, and you've done well starting from scratch so quickly.

That's why for most people I try to only do the -100mV undervolt and Balanced power plan, with fan tuning or setting to 100% to get them cooled down enough to stop thermal throttling so they can get back to gaming or work they got the laptop for in the first place. Most people don't have the time, patience, or technical interest to spend more than the absolute minimum, which is aok by me.

Thanks for hanging in there.

 

the 39 was from desktop idle, jumping up for a split second. I get throttling on core 2 only at 35, and power doesn’t exceed 45 watts, on 36 throttle, power exceeds 45 watts. Performance marks in firestrike change by about 50 per MHz, so why is it better to stay under 45 watts? I’m not quite sure I understand. And although I reset heinfo between tests, I apparently wasn’t smart enough to reset it as the test starts, so I had low initial clocks screwing everything up

 

Put Windows High Performance power plan and clocks shouldn’t drop below max 6 core Boost.

 

Of course a high multiplier will get higher score on benchmarks that run short internal computations not longer than the Power Timeout Window, they add up lots of small calculations that will run at high performance during their short runs and add up to a higher score.

If the calculation pushes through longer than the timeout Window the multiplier drops to the that is stable at the long term 45w power usage. It only runs at your higher set multiplier for the short Power Timeout Window, and then after that times out the multiplier is dropped to sustain no more than 45w.

This is just supposed to be a long CPU load sustained test, you set the power plan to High Performance so the CPU won't drop the multiplier normally, only when the power timeout window is exceeded will it drop multiplier - which is what you are looking for - to see what multiplier / frequency matches all core load at 45w.

Normally you would then set the power plan to balanced, OC to highest all core multiplier that doesn't thermal throttle - short CPU process loads that run at full multiplier within power timeout window. That would give you best performance for most applications / games.

I use the 35x all core settings profile only for long running batch jobs, like rendering, data file processing, etc that run calculations way past the power timeout window, this gives me a profile that I can undervolt further than high OC profile and doesn't spend time adjusting the CPU multipliers to match power load - the CPU is usually more conservative and will give less performance than locking it to 35x.

 

Also, is the effect of ambient temps on thermals linear or not? I notice running at 73F temps are around 6F higher than running at 69F

 

You'd have to nail down lots of variables in the laptop to be able to reliably get exact correlation between ambient temperature changes to be visible over those normal variances.

Windows itself runs lots of things in the back ground on a schedule, apps can install services doing the same, games vary their terrain / levels, and other settings you forgot you changed can affect the results - until you remember them.

Try to keep the ambient / room temperature stable, run the tests in the same ambient environment - make sure the laptop if brought into a cooler / hotter room stabilizes it's internal temperature to the ambient, it can take a couple of hours. If the ambient temperature is higher or lower enough to affect results, then start multiple lists - daytime hot ambient condition results, night-time cool ambient condition results, and maybe after a while - if you stick with it over time - you'll get enough data to correlate ambient temperature to results.

You could go by the motherboard temperature as the "ambient" baseline, but cool air coming in at a lower temperature will also lower temps quicker and give better results, so again it's hard to nail down what is the "ambient" instantaneous temperature, unless everything stabilizes to the same temperature.

Theoretically you can add / subtract ambient changes, but not directly, it's going to be a ratio not a direct 1:1 difference. 5c cooler air might result in a 1c-2c lower measurement, given everything is controlled, but that varies for different laptop cooling systems.

I try to use an air-conditioned stable room to test in, making sure I don't move the device around under or away from a cooling source vent, and for real certification level testing the room is much more instrumented and controlled.

Spoiler

I remember the first time I was doing laser imaging using multiple optical paths with many temperature sensitive components and I got a great first run late at night, then later that next day the verification group couldn't reproduce my settings or results.

In between the time I did the setup and ran my tests, and when the next group tried to verify the setup and reruns the tests, the maintenance group had "fixed / added" a cooling vent at the top corner of the room and it was every so slightly moving air that was unevenly cooling components. Noone could find the problem over a few days, so I was called back in to debug it.

I've found the best way to find those problems is to light a candle and put it in the middle of the test bench - elevated so the soot from the "soot free" candle doesn't contaminate optical elements - and watch for flame flicker - once I saw the direction of the flame flicker I could see the direction the air was moving.

Then I used a high intensity light aimed at that area of the ceiling and saw fresh metal work under the perforations in the ceiling. I duct-taped a right sized sheet of cardboard over the vent, and reran the tests successfully. All ambient conditions can be very important to the success of testing. Not just temperature.

If your location is having hot days and cool nights, and you have no air conditioning, use the windows actiely varying their open state at night to get the temps down for testing, and make your runs at the same ambient range each time.

 

@Kevin@GenTechPC The CPU fan on my month old GS65 is still not working and would really love to fix that. I've tried calling and emailing, granted a did get a few replies via email but nothing helpful. I've had this problem since Thursday and I contacted you the day after, I made this account solely to reach out. I would love to be able to use the laptop I purchased to its full capacity, please assist me.