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

Yup, stopping thermal throttling is only one thing undervolting does, it can also give you more power budget for performance, it all adds up, each little -mV and tuning tweak, until you get real cooling and performance benefits out of the software tuning.

 

Ok, actually overclocked the chip to 4.1ghz let’s see how it goes

 

Running at 3.9 (actually overclocking the 6 and 5 core count by 100 MHz, temps cap out at 85 in a small batch fpu burn test, and 73 in a regular stress test. It will not allow me to overclock to 4.1, even though it throttlestop says so, and eventually it runs so long at 3.9 the turbo gets tires and it drops to 3.8, staying steady at 82C. I could probably push it 5C lower if I ran fans at 150% instead of 120%. I’m getting almost-LM temps with no LM

 

Yeah, I expected a drop, but not by 24C. It seems laptop is already about 10C under what I expected with no underfoot, and with and underfoot it goes so much lower. Probably not going to run it at 3.9, though, as I prefer the quiet and the low temps of 3.6

 

Combined with rma overclock s (which I’m not going to do), I should be able to get this thing to 14400in firestrike. I wish I could overvolt the gpu, I could give this a monstrous overclock. Even with my 220mhz core and 413mhz mem overclock, gpu never goes above 70 C (unless I run Furmark or msi kombustor, then it hits a “whopping” 73C

 

Undervolting at lower multipliers gives even more headroom, so you can move deeper into the low -200mV range

I have several preset's, for benchmarking, for gaming, for day to day lower demand usage, and a couple of long term batch job running settings - usually right around 45w constant power load as the computations are running longer than Power Time Window, unless you have an unlocked CPU - then it's all up to how loud you can stand to have the fan's running for hours on end. And, I don't use battery power often, but a profile to save power to extend battery time would help too.

Lots of fun stuff still to do.

 

Well, at 3.9ghz, -190 seems stable, I’ll probably use the same underfoot for 3.6, just Incase I switch back to 3.9 on a 210 underfoot or something and panic when it crashes

 

Congratulations, that's what I keep trying to get people to do on their own, tune their laptops, and skip the hardware hacking until you have control of the baseline potential of software tuning your laptop.

Once you know how to do this, after doing it a few times, you can do this on the fly by taking a USB flash drive to a retail store to test your new laptop before coming home with it. You know before walking out the door whether you want to keep it or not.

No more returns, or RMA's, no more hardware hacking. It's right there for everyone to benefit from. And, if you have to drive a couple of hours to the closest retail location, now you have a way to make it worthwhile in one trip - just make sure they have a couple in stock so if your first one is a lemon you can test it out right then and there, so you can reject it for another unit without taking it home and needing to do that round trip drive all over again.

Also good for checking open box units, or floor models, with the added benefit that you usually don't have to pay for it before being allowed to "open the box" and test it out.

Glad it's working out for you on the 2nd laptop go around, thank you for sharing your results!

 

Still don’t think I hold the record in this forum for and undervolt. Damn

 

-190 got a bsod after about an hour of dying ligh, the WHEA_uncorrectle_error one. I dropped ( or raised, I guess) the undervolt 5 more and I’ll see how it does. -185 is still really good.

 

The last test is Idle, let it sit unused, all programs and as many systray / services shutdown so that no processes start up to raise the CPU clock / voltage, and after about 30 minutes if it doesn't crash you should be ok, otherwise drop another +5mV if it crashes at idle. You may there now.

 

Is idling really a better test than gaming?

 

No, Idle testing is the other end of the swing of voltage use from 100% load, and the two different points where your undervolt affects stability.

When at 100% load stress testing, or gaming, or other active use you are using the undervolt to reduce voltage to the minimum needed for the task, and that drops temperature and can stop power throttling.

At the other end of the spectrum, at idle, with nothing running, the CPU will ratchet down the voltage a little more and a little more, until the voltage + undervolt might be not enough voltage to run stably.

If you BSOD at idle with your 100% load undervolt, drop the undervolt +5mV to stop the Idle BSOD... your top load undervolt will still work, just +5mV less.

Does that make sense?

 

Yeah, that makes sense

 

So -185 seems stable, that’s a pretty damn good undervolt. Under combined loads (namely, the massive load of dying light, 95% gpu and 55% cpu, I get average gpu temp of 71C and average cpu temp of 69C

 

And, those values should be stable, given the same ambient temperature (higher / lower ambient raises / lowers laptop temps).

Keep an email draft or txt file with the results now, and over time you can check every 6 months or so to see if anything changes.

Now you can get back to enjoying your laptop instead of tearing it apart multiple times, and spending many hours over weeks (months?) trying to get what you got with software in a few hours.

Could you get the temps lower with re-pasting / re-padding? Maybe, maybe not, but temperatures lower won't get you better performance.

 

When I first encountered this, I thought that it was a typo, but it isn't (although he probably increased the voltage after that, he most likely still has the highest negative value for the offset):

http://forum.notebookreview.com/thr...scussions-lounge.815216/page-79#post-10738516

 

It's all about quieter fans, especially with these thin and light high-pitched hair dryers..

 

Ordered mine (1070 version) with conductonaut, Fuji pads, screen calibration, 500gb 970 EVO, fresh Windows install, and a free t-shirt . Do y'all think this will be enough for some streaming to twitch? I've been using my desktop but it seems like it would be nice to do some streaming with a laptop.

Sent from my SM-G950U using Tapatalk

 

To much to sacrifice in taking them apart and reassembling, over and over and over, a huge waste of time. All for no performance benefit, and no perceptible fan noise reduction at gaming load, it's still gonna whine.

Better and likely cheaper to get a nice set of headphones for you, and whoever else is in the room.

 

It does help if the stock TIM is sub-optimal. They often are.

 

If you can undervolt the CPU enough to stop thermal throttling under load, tune it for gaming and your applications, it's good enough to leave as is and let it do it's job, you've already paid for it, tune it and make it work. That's the best kind of DIY, the one that doesn't cost more $$$.

 

What I'm saying is you shouldn't discount it. They can work together. Obviously undervolt first.

 

This is my first time undervolting, thanks everyone here for the helpful posts. I'm currently at -200 core and -150 cache, stable so far at idle and stress testing, here is a screenshot of my throttlestop settings. It's been stable on idle and prism95 as well as xtu stress testing. I'll keep pushing it and we will see what happens.

Cinebench at about 1070, pretty good with such a big undervolt. Idle temps at 45-48C, and I have had probably a 5-10C drop in max temps while gaming.

 

Is this chip on of the ones where the core and cache have to be on the same setting? On my previous 6700hq, it didn’t work unless they were the same

 

I do discount it - hardware hacking, as I don't consider re-pasting or re-padding part of the "normal" steps of ownership of a laptop.

I expect my laptop to work out of the box without thermal throttling after undervolt + all other available software tuning, otherwise I send it back for another one.

There's nothing wrong with having fun DIY'ing hardware, it's just not something most people want to do to get their new laptop to work right out of the box.

Most people are looking for solutions that don't require hardware hacking - it's just not something anyone thinks of when buying a new laptop.

Tearing apart a laptop isn't normal activity, any more than tearing apart a new TV, or Stereo, or Car would be normal. Whatever we buy should work, and if it doesn't work to our needs or desires we return it for another one. It's not considered normal to disassemble new products in order to make them work as they should, and usually voids the warranty.

And, most importantly, if it is working after software tuning and optimization, you don't fix it with hardware hacking.

Knowing that you can do something that can make a change, like re-pasting and re-padding - or lapping /etc, doesn't automatically follow that you *should* or *must* do it.

If you can drop temperatures further than software tuning, but it doesn't help performance - then hardware hacking is not functionally required to follow through on that path.

Considered thought before an action can help you avoid taking an unnecessary path before you take it. Actions have consequences, and no matter low the chance of a negative outcome an action has, not taking that action may be the best path to choose.

 

I have been seeing people report different undervolt for cache and core on 8th gen, but each CPU sku is different.

What are you using for undervolting? XTU or TS, or BIOS?

If you can make a value change, drop down selection, or entry into the setting, try to save it and see if it works. That's how I do it when working with a new CPU / BIOS / laptop / version of tuning software.

Experimenting with software, trying things out, is the joy of software tuning vs hardware hacking, you can undo the software setting and try something else quickly, it doesn't take long to reboot after a crash.

As a suggestion to try, if you can separate the core and cache multiplier and undervolt, start with reducing the cache ratio lower than stock first. I do a CPU benchmark or two to see when the reduced cache ratio drops performance, and then up the multiplier on the cache +1x to get a good minimum setting.

After that you have more room to undervolt the cache, with the lowest multiplier found and set.

Have fun

 

Cause it seems I’m able to push the core much lower than the cache. Currently at -215. Previously, I was dropping both equally

 

Yes! that's a good observation, and is a thing that can happen. Splitting the cache / core undervolt can allow one or the other to drop lower than both together, awesome.

Reducing the cache multiplier might allow for more undervolt on the cache too...

 

Btw, would you say the most likely program to cause a bsod during undervolting would be firestrike, prime 95 26.6, or looped unsentimental heaven? Obviously I’d want to find the one that crashes first to find out quickly wheither an undervolt is unstable, like starting a sculpture with a jackhammer and ending with a chisel. Oh, and I use TS. The best outcome would be if prime 95 crashed first, cause then I could run prime 95 in the background and keep dropping the undervolt

 

at -220 now on core, -185 cache, haven’t hade any “**** you” crashes, where firestrike immidiatly bsods (or orange screen of death, got a couple of those) after it finishes loading. I guess the worst 5hing I could of would be tell it to immidiatly drop by 70. That would cause quite a few problems, including racing to TS to turn it off before my pc blue screens
Edit: currently at -225 core, much further than when I had core and cache at same undervolt

 

I kinda use P95 as a last resort, most of the time it's the last test to run when I think I have a stable setting. It's an important "heaviest" load test, so do run it before deciding your settings are final. I don't run it for long these days, as I've found 5-15 minutes is enough to see if it's stable. Especially on thin laptops it's more torture than test.

I use XTU only because it's the easiest to show novices how to undervolt, it's a quick install, and an easier undervolt setting for them to visualize that I can share an image from XTU with Yellow numbers, Apply, and eventually Save when they have stable setting.

XTU also has a Stress test and Benchmark they can use to test stability and performance changes / drops from settings, again all easy for the novice to get into easily.

So, for other testing I use a range of whatever I am currently using, a game - or just the benchmark in / from it, Heaven, Futuremark tests, etc.

Even if you are doing CPU tuning, you can use the GPU, as it's all stressing the system. I try to avoid using applications that save state or do work while testing an undervolt.

When I am testing undervolt or some other setting that might make things unstable quite often I'm called away or interrupted, and so I make sure to have "stock" or "safe" settings ready in a profile that I can select to make myself stable on that machine so I can do work and not crash while doing it.

I encourage people to get creative, to find your own mix of stuff for testing, you'll find it helps if you can connect with the process through your own choices, make it yours.

When posting results, we like to have hwinfo64 showing CPU / GPU conditions along with results, and if possible snap an image just before a test ends so all the "gauges" are fully engaged in the test. Snapping an image when it's over to see the Peak settings is ok, but it's more fun to get a snap in process.

You are moving quickly, making multiple changes, maybe a bit too quick since sometimes changes are on the edge and failure won't show up right away, or on every test you run, so you might want to take written notes on the progression of changes - what you've tried - what did and didn't work and why - so when you find yourself painted into a corner crashing, you can walk yourself back to stable settings and work on one setting at a time again before branching out with other turntables.

If you think you are done and a couple of days from now you BSOD, try not to get upset, instead backtrack software installs / driver installs, to the point you were stable, before giving up and going back to completely stock - that is unless you need to get work done, then of course go back to a known stable setting - which you might not have now except for stock.

Good luck.

 

Right now, I know -185 cache is stable, and I know +200 core and +300 mem on gpu are stable. Some game crashes at 400mem, but that might be due to the game in question (dying light). The real thing is that I’m currently at -230 cpu vcore, and showing no problems. Other than the one guy on this thread who had like a -950 vcore (no idea how he did that), I believe that’s the lowest in the thread

 

Would you recommend the real bench stress test for about 15-30 minutes

 

Can someone tell me if the stock 1080P display takes a 30 or 40 pin display connector?

I'm toying with the idea of upgrading the screen myself to a 4K panel. Will something like the Sharp IGZO 4K panel from the XPS 15 be plug and play?

I bought an XPS 15 9570 and the poor thermals had me running back to the GS65.

 

Currently at -250, Jesus christ

 

O

i don’t believe that TS allows me to drop the cache multiplier, and I don’t want to drop the cpu past the current 36 multiplier, already lower than the default 38( I think that’s the default). Anyway, currently at -255, although I’ve only been doing full runs of firestrike (demo included) to test

 

I’m at -340. What the **** is going on

 

Can an undervolt allow the pc to freeze without a bsod? Happened a couple of times at very low underclocks, usually I associate underclock failure with bsod’s. Have to restart computer to get out of freeze

 

ignore this, made this post by mistake

 

The Asus RealBench test? I think it may take that long on it's own anyway.

The RealBench test isn't like P95, RealBench is about real work - a suite of real work tests using a database of files, including IO, memory, etc, not just CPU.

P95 is 100% small FFT are CPU + special instructions all meant to push the PSU, CPU, cooling to the limits. That one is the one I would limit exposure to. In the old days I'd use it to find weak PSU's I suspected of randomly folding and causing crashes.

For quick tests you can use any CPU heavy test, to quickly find new undervolt limits, and then when you back off use longer tests for stability.

Usually the longest stability test is the Idle test... it's the one that can show up a week later when you get a phone call, leave, and forget to turn off the laptop, only to come back to a rebooted laptop - or at a BSOD screen if you disable automatic boot after BSOD - recommended for testing BTW.

You don't really need to spend a lot of time finding 100% load test failures, they usually find you quickly.

Another trick, if you think you are stable, but don't have time to make sure, add +10mV (or +5mV) and save that as a stable setting - take time later to shave off another -1mV to -5mV for grin's, or live with the slightly less aggressive undervolt.

You are already at -250mV and apparently stable on your current testing - you might as well set it to -225mV and live on it for a while - enjoy some gaming, or Netflix, etc.

Yes, BSOD is one symptom, but so are other anomalies, like threads in P95 failing, while others continue, and there are other tools that will check work and find failures without crashing the app or system BSOD. Like RealBench, as I recall when I was tuning memory I would get failures in the Realbench run even though the laptop and the app stayed running.

So in your case, hitting -300mV at reduced CPU settings, you might try P95 small FFT - disable AVX/FMA3 ( http://www.mersenneforum.org/showthread.php?t=21462 ) to avoid overheating - run for 5 minutes and see if any threads die... let it run up to 15 minutes - reduce undervolt when you see threads die - and try again. You can run longer if you watch the temperature while watching in real time for the threads to fail.

It's the walking away leaving P95 running - over night or longer - that I don't recommend. As long as you are there watching and monitoring the test to stop it or shut it down if it starts overheating, threads die, or it hangs or crashes.

It seems like you are having lots of fun.

 

I’m actually at -330 now, running realbench for 30 minutes. Got a bsod on 335 after 11 minutes

 

Interesting. You do know the test is running through a range of different work, so it might be helpful to know what test was running when it crashed. Not sure, but I think there are command line options or settings to run specific tests, so you could repeat that same test again instead of waiting for another 11 minutes for RealBench to get to that test.

 

Well, I know what crashed because I recognize the bsod. It’s the WHEA one, which I usually associate with a cache undervolt too low, not a core undervolt. So the previous cache undervolt I thought was stable (-185) was slightly unstable. I pushed it back to -180, am currently 22 minutes into the realbench stress test on -330 core, -180 cache, going smooth and steady. I’ll update upon completion. Worth noting is thatpm at this point of almost an hour of both max gpu and cpu load, there has been no cpu throttling, max cpu temp was 87 for a split second, but averages around 83, and max gpu temp is 72

 

Stable at -330, -180

 

You can't do -330, -180.
It ignores the 330.
Cache and core are directly linked on skylake era chips.
8700K or Coffee Lake is still skylake era tech (still uses the same architechture, just different chipset).
it will use the lesser of the two.
So your undervolt is 180.
The -330 gets ignored.
You can never do -330 on both either, ever. Instant BSOD.

 

I know that was true for sky lake, and I know -330 on both is a terrible idea, but why can I see my cpu temps drop as I lower the vcore? And if it ignores the vcore, then why do I get freezes when I lower the vcore too much? If it’s ignored, it shouldn’t do anything

 

Also, how much of a gpu undervolt have people been able to get? I’ve flattened my line from the right edge almost all the way to the left, and it still runs. Is that normal? My voltage curve spikes to max quickly, then stays steady at max after 825mv
Edit: also, does anyone know why when I flatten the curve, gpu-z says I overclocked the card by almost 100mhz, which I know I can’t do because it can’t even take a 220 MHz overclock? I was flattening the curve of a 200 MHz overclock

 

Could you try run Aida64 as the guide with your stable -330 Core / Cashe -180mv undervolt. Do exactly the same but this time change to -180mv for both Core and Cashe. Post screenshots from middle/near end of both runs and when finish. Thanks

Have you also tested with let’s say -280mv undervolt for both domain? Or even higher.

 

hey check this out i think it might help you out