Pushing my i7-9750H to it's limits

[MutatedT7]

Hey everyone! (and especially @unclewebb )

I have been reading and following this forum silently for quite a while now, and
I want to take you on a journey to really squeeze my intel i7-9750H to it's limits!

Here is some background information on my system:
I have a Lenovo Y740, single heatpipe shared between the i7-9750h and RTX 2070max-Q
With two fans, one for each.
Default Thermal Paste seems badly applied, since the temperatures between cores fluctuate quite a bit during idle and load
During Idle here is about a 3 degree difference, between the hottest and coldest core, during Cinebench there is an 8 degree difference.
The Lenovo Legion series has a Debug-BIOS, which allowes me to change Power Limits and other features
I'm currently sticking to ThrotteStop since I prefer it's clear visual interface, but let me know if I should move to the bios for anything.

I ordered an aliminium stand off of Amazon, to provide the laptop with better airflow for cooling
(I didn't get one with all the fans, since apparently they do not provide much better cooling *Credits to Jarrod's Tech)

On top of the stand, I also purchased some Noctua NT-H2 Paste, I figured this one was the best as I've seen a lot of
users report that it is one of the best, if not the best laptop pastes. (Feel free to recommended others).

Let's start with default settings on ThrottleStop, I will be using the laptop stand during the entire process.
You can see the screenshots during idle and load, the CPU is idle at around 53 degrees everything else looks normal.
During load we start seeing problems, we immediately Thermal Throttle, Partially because the default PROCHOT setting is so bad
We suck 70 Watts of power, and because of the throtteling the CPU struggles to even hit 3.8Ghz on all cores.
We end up with a Cinebench R20 score of a whopping: 2791! (Log Attached)

We Thermal Throttled the entire time, there was some Power Limit throtteling on top of that,
and without maximum performance this is about as bad as it can get.

This is the baseline, let's start by undervolting and adjusting some settings!

 

[AOne]

I would suggest - first repaste and then start tweaking.

 

[MutatedT7]

I started by changing the PROCHOT offset, I have seen 98 degrees being the recommended setting,
so I'm going to use that. I also set a Performance and Battery profile.

I took notice of the "Powersaver C0%" setting, I assume this turns on Windows Power Saver when the Utilisation rises above set number?
I'm not sure so I would appreciate some clarification on that

Next up are the profile settings, I enabled SpeedShift with a value of 64, I also disabled BD PROCHOT since I absolutely dispise this
feature. I read that C1E does nothing on modern processors, so I'l leave that enabled for compatibility reasons.

Moving on to the FIVR Window, I changed the Turbo Ratio limits to 4.5Ghz on all cores.
The max equal undervolt for CPU Core and Cache seems to be -135.7mV, I have also tested
the CPU Core with a bigger offset, but I was able to go up to -1000mV without seeing real benefits,
So I will need some help with that later.
I'm going to leave the IGPU and other settings alone for now, I'm not sure if it can benefit me
But let me know if there is any way to gain performance in that area.
Lastly, I enabled the "Disable and Lock Turbo Power Limits" feature, for use later in the TPL window,
and I set the voltages to immediately save.

Lastly there is the TPL Window, here I set both options on Clamp, to make sure that the CPU listens to them (If I'm right)
I changed the PL1 (Long Power limit) to 80 Watts, as I saw Power Throtteling back to 70 Watts.
I also changed the PL2 (Short Power Limit) to 90 Watts, as this seems like a safe value.
My CPU is rated for a TDP of 45W, my numbers are almost double, should this concern me?
I think the answer is no, however what would be a good number for my PL2?

Regardless of that, I ticked "Enable Speed shift When ThrottleStop Starts" to ensure that speedshift works.
I'm not sure what Turbo Time Limit does, nor the PP0 Limit. I recall some people putting this slider all the way up,
but I will need some advice on this.

This should wrap up my settings for now, let's check out the results.

======
Results
======
First thing I notice, my CPU is immediately running cooler at idle.
Running at 49 Degrees, instead of 53 previously.

Running Cinebench, everything seems fine!
Though we hit surprisingly high temperatures, I think this might be might be because of the increased power limits.

Nevermind, everything went wrong I regret everything
Having adjusted the PL1 and PL2, we still Power Throttle at the end of the test, and we are very close to the 98 Degree limit during the test.
I Think I might have made a rookie mistake by not restarting my computer, or something else is going on.
I'll try restarting but I would appreciate some advice and help in the mean time

Despite all that, a score of 3079 is a big improvement already!

As always, there is a data from everything I said

I would suggest - first repaste and then start tweaking.

@AOne Thank you for the advice!
I just wanted to have a before and after result, to see the difference
that it makes

Any advice on the Power throtteling?
And should I increase the core undervolt, decreasing the Cache?

 

[unclewebb]

If you changed the PROCHOT Offset value to 2, you should also use the Lock PROCHOT Offset option above that. This makes sure that this setting will not get changed. Some Lenovo laptops will change this value.

Powersaver C0%

This setting is not used when Speed Shift (SST) is enabled. No need to adjust this.

Use Cinebench R20 when testing different voltages. Start with the core and cache offset set to -125 mV for a baseline. Leave the cache at this setting and start increasing only the core in steps of -25 mV. Most users see improvements in their Cinebench scores with the core up to -200 mV or -225 mV.

http://imgur.com/qcqYRmN

http://imgur.com/xjWGPv2

http://imgur.com/WL3UI4K

Not a huge difference but you can definitely see the progression.

If you have a Nvidia GPU, I see little reason to undervolt the Intel GPU. When the iGPU is used, it is not significantly loaded so power consumption is not important.

My CPU is rated for a TDP of 45W, my numbers are almost double, should this concern me?

This exact same technology is used for Intel's desktop CPUs where users run stress tests for hours at 200W. I agree that 90W is a big number but I think this technology can handle it.

The PP0 Power Limit is rarely used so do not check this option. Leave it set to 0 and set the PP0 Turbo Time Limit to the minimum, 0.0010.

high temperatures, I think this might be might be because of the increased power limits.

Science 101. Increasing the power limits will result in higher temperatures. A 90W light bulb runs hotter than a 45W light bulb. Your laptop cooling system is way beyond average but still, there is only so much heat that it can dissipate. When you reach that magic point, thermal throttling will result. These are powerful CPUs when allowed to run wild. If you want less heat and lower temperatures, it is easy enough to reduce the turbo power limits. You can infinitely adjust the power limits and the turbo time limit to fine tune your laptop. What works best for you is up to you. For my desktop computer, I set the turbo power limits sky high, 4095W. If the CPU gets too hot, it thermal throttles. Intel coming up with layer upon layer of different throttling schemes seems excessive to me. Too many engineers with nothing better to do.

a rookie mistake

I do not see that. Your Cinebench score increased and is already up to where a 9750H should be. I have seen some 6 core CPUs throttling so bad that they score less than 2000 points in R20. You are definitely doing something right. A few more minor tweaks to the voltage and power limits might get you over 3100 in R20. After you are finished hunting for big benchmark numbers, then you can think about reducing your power limits so your laptop runs at a temperature that you are comfortable with. If you use your laptop on your lap, holding on to a toaster all day is not for everyone.

Edit - Almost forgot. The 9750H is a locked CPU so there is no reason to adjust the turbo ratio limits. You cannot overclock these CPUs. Set these back to their default values; 45, 44, 43, 42, 41, 40. Is this important? Not really.

 

[MutatedT7]

@unclewebb Thank you for the advice!!
I saw some youtuber recommend to adjust the turbo ratio limits, quite hilarious to find out it does nothing on locked CPU's
I will start testing the extra core undervolting tomorrow, and I had a question regarding that:
Should you keep the extra undevolt if the Cinebench Score stays the same? Is there random variation in the benchmarks?

I also had a question about the power throtteling, since restarting did not seem to apply the power limits
Can I do anything else to prevent that? Change it in the bios perhaps?

Excited for tomorrow and thanks for the time

 

[unclewebb]

I saw some youtuber recommend

Don't believe everything you see on YouTube. I grit my teeth every time I see some bad ThrottleStop advice. Anyone with an internet connection can become an expert on YouTube. Glad you signed up at TechPowerUp for some legit advice.

Should you keep the extra undervolt if the Cinebench Score stays the same?

Only increase the undervolt if you can prove that it has increased performance or decreased temperatures. If you make a change and you see no benefit, why keep that change? You might have to do a few runs to get consistent results. It is impossible to keep all of the background tasks consistent between runs. That means there is always going to be some random variation from one run to the next.

a question about the power throttling

If you have the long power limit set to 80W in ThrottleStop and your CPU still throttles at 70W, there is probably nothing you can do about that. Your last log file showed PL1 throttling right at 70W. You did not post a screenshot of the TPL window so I assumed that you had this set to 70W.

Some laptops use a separate set of power limits to enforce a long term power limit lower than what you can set in ThrottleStop. If your laptop is forcing your 45W CPU to run at a maximum of 70W, that is not quite perfect but it is really not that bad. Most Dell laptops, etc., force a 45W CPU to run long term at a maximum of 45W and there is nothing you can do about it.

 

[MutatedT7]

Hey everyone!

I apologize for the long response time, luckily it was 100% worth it
Here are the results of my repasting!

So I repasted both my CPU and GPU, I started by opening up my laptop
and man, the screws from both the chassis and the fans were a real pain.
They were very soft screws, so they almost immediately self destructed when I barely touched them with my screw driver.
Had to use a lot of pressure in order to take them out.

After a bit of frustration, and yelling out of anger to an engineer who thought these dumb screws were a good idea,
I finally arrvied at the actual pasting job. Let me tell you this, I have never repasted a CPU before, but man was I shocked.
When I took off the fans, together with the heatpipe, I looked at the CPU and GPU, and I can NOT believe what I saw.
I have seen some photo's of bad pasting jobs, but this was the worst thing I have ever seen in my life.
Both the CPU and GPU looked like someone ate overdue TacoBell and threw up a gallon of vomit all over it.

I took my rubbing alcohol, using paper towls, cotton swabs and about a half hour of my time,
completely cleaned off the old thermal paste. I also cleaned out a lot of dust from the fans, and
intake vent grille to hopefuly improve thermals and airflow.

I then arrived at the actual pasting!
I bought a bigger tube so I could do some test runs on a piece of paper towel,
after a couple of tries and a gameplan for the pasting method, me and my Noctua's NT-H2
Went to work. Here are the methods I used for the repaste of both my CPU and GPU:
a small line/dot for the CPU, and a Penta-dot for the GPU. (Unfortunately I forgot to take a photo of the pasting job)

The penta dot for the GPU might seem a little bit overkill, but I ensured the corner dots where seperated from the corners
to ensure that the paste could spread properly. I then slowly put the heatpipe and fans back on the chips, I ensured that I
used an even amount of pressure and would go as far to say that I did a good job on that last part.

Putting everything back together, and booting up windows
I can immediately see results in my idle CPU temperature.
It went down from 49 degrees, to 45 degrees!

Secondly, I tested my pasted CPU in a couple of Cinebenches.
This resulted in something weird, even though my CPU is clearly cooler, shown
by it's temperature during games and idle, my Cinebench Score did not improve, or even went down.
I am fairly confident in the pasting job I did for the CPU, so no issues there, but I have no idea why the score went down.

My theory is that because of the lower temperatures, the CPU now stays at higher wattages for longer, which results in overall
higher temperatures and the CPU hitting the Thermal Throttling limit of 98 degrees. Because of the the throtteling, the clockspeeds
go down, and the score is lower.

This is my best guess, but I hope someone (@unclewebb) can help me improve the scores
As always, I have attached some data about the scores, I should note
that the CPU is power throtteling during the entirety of the Cinebench, and sometimes
PL1 Throttles, just like before. (I have not adjusted my settings from the previous post yet).

-First Photo (CPU During Idle)
-Second Photo (First Cinebench)
-Third Photo (Second Cinebench)
-Fourth Photo (Third Cinebench)
-Fifth Photo (CPU During Start of Cinebench Load)
-Sixth Photo (CPU During Middle of Cinebench Load)
-Seventh Photo (CPU During end of Cinebench Load)
-Eight Attachment (ThrottleStop Log during same Cinebench)
-Last Photo (Internals after cleaning with Alcohol)

@unclewebb Just laughed because of the YouTuber "Expert" thing, I'll be careful to take any advice from there, starting now
Thanks for that advice

 

[unclewebb]

@MutatedT7 - I see a problem. In your mid and end screenshots, there is a 20°C difference between the first core and the second core and between the second core and the third core. I hate to tell you this but you might have to try doing the thermal paste again.

It takes most users a couple of tries before moving from apprentice to journeyman. Maybe @AOne has some advice about Noctua NT-H2.

Some of this difference might be sensor error or perhaps the CPU and heatsink are not perfectly flat. Leave it for a week and see if this difference improves.

Your laptop is running great. Even if you get the paste and temperatures 100% perfect, there is very little room for any performance improvement.
Keep that in mind before deciding anything.

 

[AOne]

These scores are very good, but there's definitely a problem with the repasting. These huge differences from core to core are the 100% proof for this. Don't mind what's written on Noctua's application manual. I believe it's not the best way as if you use a dots method it would eventually lead to layer thicker than needed and air trapped in the spots where the paste from different dots touches after pressuring. Too thick paste is not a good conductor and air is an insulator, which above all expands when heated, acting like a jack. My way of repasting is to apply a very thin and even layer over the whole chip and tighten the screws very carefully, in order according to the numbers beside them to achieve the correct and properly distributed pressure between heatsink and chips. If it doesn't happen from the first attempt don't hesitate to try again and again. Sometimes more patience is needed and more then once or twice trying but at the end it's worthed. It's been more than a year since I repasted the last time and all is better than good. I'm with 9750H/RTX2070 set and the temps in games and high demanding tasks (CAD/CAM, rendering etc.) do not exceed 87 degrees. After CB20 they're mostly under 83 C and in TSBENCH never go beyond 75 C. In all cases Limit box remains black, indicating no limits were applied. So, brace yourself with patience and don't hesitate to redo the pasting. Don't close the lid until you see a difference in cores under 7-8 degrees C.

http://imgur.com/zu5NJyV

My results right after two CB20 tests in a row. (Airplane mode On and CB20 set to Realtime in Task manager's Details graph).

 

[MutatedT7]

@unclewebb Rough to hear, but I'm going to eat it up and repaste it again later!

I'll game on it for a little bit to see if things improve, regardless of that I'll still redo it though, since this isn't the job well done that I wanted

@AOne Should I order a thermal paste spreader for this? Or is this not necessary?

Oh and, perhaps I let the screws too loose because they were frustrating, is that something that could factor in on this?

And I was wondering if you work the screws back in reverse (6,5,4...) or the way you took them out (1,2,3...)

Thank you for the advice, I'm definitely going to do a repaste soon and with some advice it will hopefully be an ace

 

[AOne]

You don't need a spreader. I use whatever comes handy - in my case a plastic flat "thing" for opening phones. Use whatever you've got. The numbers are followed in straight order - 1->2->3. Tighten the screws, so a pressure is applied (it's very necessary and it'sprobably the cause of your issues). You could just try to tighten them and make a test, before starting a new repasting. Who knows - might be enough and spare some paste and effort. Don't overdo the screws anyway, as I've had to resolder two of the threaded bushings to the motherboard of my older laptop (an easy job, but why the hassle if you could avoid it).

 

[MutatedT7]

@AOne I have a few of those prytools laying around as well, I'll use that to spread the paste and then follow up with the heatsink, tightening the screws in order

I'll make sure to get some more photo's this time as well

By the way one last question, some people say that doing a dot or a line is better than spreading the paste, because of air that could get trapped in the paste, is this true? Why is spreading the best method?
Curious to know, thank you for the instructions!

 

[AOne]

Different people, different thoughts, different arguments. I found that for me spreading works better and the way I'm explaining this to myself is, when placing dots you need more pressure to squeeze the paste. Besides I think when the paste coming from different dots touches, in these places are forming air pockets. Spreading it thin and even works better for minimizing the amount of paste. Too much is bad idea as it should fill only the tiny gaps and pits.

 

[MutatedT7]

Hey everyone! (@AOne and @unclewebb)

Took me some time, but I ordered new screws online and decided to repaste my system
I first did another repaste for a different laptop, to get some more practice. I started using a thermal paste spreader,
and the results were good, lower temps and no big difference in the different cores.

I then repasted my own laptop!
I firstly replaced the screws, since I figured if I was ever going to open
it again, I would be unable to, since the screwheads were badly damaged.

I was very curious to see why my first repaste did not work out and the photo's
will show you that the repaste was indeed not very succesful,
there was a lot of air in between and it was just a big mess in general

I thought I did a good job, but I think I might have let the screws too loose or the
paste did not spread properly, I wonder why the paste did not spread.

I then thoroughly cleaned both the CPU and GPU, after that I applied two lines on the chips,
one for each, and spread the paste using one of my pry-tools. You can see the results from the photo's,
one for each, with and without using flash.
Please let me know how I did!

I then screwed back my laptop, and immediately noticed something
When my right fan is on Low RPM, you can hear a little high pitched "brr" sound.
It sounds like someone is very cold, and their teeth are chattering. (But very soft and pitched up slightly)
*Update: After one-two days, the sound is now unnoticeable and I think it's resolved??

Regardless of the little fan noise, I checked my throttlestop to see some results.
As you can see from the idle screenshot, my CPU is now idle at around 45-46 degrees
With a difference of 3 degrees in between cores.

The idle temperature seems to be cooler than before, however I am worried since there previously
was a 2 degree difference, now there is three.

I then did some benchmarking, there was some weird variation during my testing.
Good news first: The temps definately went down!
Compared to my previous repaste, the CPU would hit temps of around 96 degrees,
and sometimes even the maximum of 98. (End cinebench previous message)

During the end of a CB20, the temperatures are now only 91 degrees, which is a clear improvement!
I also achieved my highest ever score, a 3123!

Now for the bad news:
During the test, there is a 13 degree difference between cores
I did some more tests, to get an average, however my scores sometimes
went as low as 3000, with no clear indication of what the cause was.

I attached data for both my best run, and what was the worst run I could pull data from
(the scores have not been as low as 3000 again)

Another thing, during some runs, my CPU PL1 throttles just at the end
It seems to happen at random since it does not happen consistently at every single run.
Another things that frustrates me.

I have also attached all of my ThrottleStop settings, perhaps those could be of some use!
There are a lot of attachments in general, so please have a look.

Here are some questions I have:

1. Did I do a good/decent job at the Thermal Pasting? Did I use too much thermal paste? And did I spread it properly?
2. What was up with the weird fan noise?
3. Do the results indicate a good repaste?
4. What is left for me to do, how can I push my Chip to reach AOne's scores of 3200, is it even possible for my system?
5. Why are my results so inconsistent sometimes?
6. Can I prevent the PL1 Throtteling?

Thank you for all the great advice so far,
I'd say we make a good team

 

[AOne]

1. Did I do a good/decent job at the Thermal Pasting? Did I use too much thermal paste? And did I spread it properly?
- I would repaste again. In my opinion, the paste is more than needed. Anyway it could be lots of things, even some bad luck in tightening. Just don't give up and try again and again until you're satisfied.
2. What was up with the weird fan noise?
- some dirt, paste dropped there or whatever. If it's gone, don't think about it.
3. Do the results indicate a good repaste?
- Your results are strange. Vary good scores, but the difference in temps is too much. Don't mind the difference when idle. There are always some processes running single core, affecting only one of the cores, so just ignore it. What matters is the difference, when all the cores are loaded equally- 100%.
4. What is left for me to do, how can I push my Chip to reach AOne's scores of 3200, is it even possible for my system?
- I would remove the clamping for PL1 and PL2. Your scores are already pretty good.
5. Why are my results so inconsistent sometimes?
- #3
6. Can I prevent the PL1 Throtteling?
- #4

 

[unclewebb]

I am worried since there previously
was a 2 degree difference, now there is three.

Congrats. OCD has fully set in.

my highest ever score, a 3123!

More congrats. I sleep well at night whenever I achieve a new high score.

The Bench2 End screenshot shows PL1 power limit throttling at 70.1W. The TPL screenshot you posted shows that the PL1 power limit is set to 75W. Did you make a change and forget to mention it? If this was set to 75W and you are getting throttling at 70W, that would indicate that the laptop has its own hard limit set at 70W. I am betting that you just forgot to mention the change.

I think your thermal paste is good enough so you can run Cinebench at full speed with zero throttling.

Higher memory speeds are useful when going for the gold. Reducing background tasks to the bare minimum also helps. There are a few more tricks like running Cinebench at a higher Windows priority.

 

[AOne]

By the way, what's you ambient room temperature? Try lifting you laptop a little bit and check if there's some improvement. I 3D printed myself 2 simple legs long time ago and I think they help a little bit.

http://imgur.com/4oQNLvh

http://imgur.com/d5YYnOW

unclewebb is right and I think I've also mentioned it somewhere above - stop all unnecessary task running in background; turn on Airplane mode; set priority in Taskbar's Details menu to Realtime (the screen will freeze during CB test). This adds more to the scores in testing and that's how I got my scores. unclewebb achieves 0.5% CPU load on idle but I can't never go under fluctuating constantly between 0.7-1.5 as I've got a lot of background tasks, most of which restart if I kill them.

 

[MutatedT7]

@AOne What does clamping do exactly? Would you care to explain a bit about the TPL menu?
(Or could you link me to the manual perhaps)

I will Repaste my system again later, hopefully then I'll be able to get a low
difference in between cores

As for the ambient room temperature, there is a heater underneath my desk that I have turned on sometimes,
it heats up the room to a maximum of 21 Degrees Celsius, however the warmth going up could heat up the area to around 24 degrees I'd say.
When the heater is turned off, some cold air slips through my big windows surrounding the corner of the room, the ambient room temperature
will then drop to around 18 degrees. I run the benchmark during the same conditions (heater off) and I use a stand, so that should be good for airflow.

About the tips, thank you! I will be sure to try all of those and run a few more benchmarks. When I freshly installed windows on my machine,
I followed Xander Baatz's guide on a light system, I also tweaked all of windows' settings myself to ensure performance there.
I like to game in my spare time, so some things like Steam or MSI afterburner are running in the background, even though I have disabled them for startup.
I'll try to kill a bunch of tasks and see if it makes a difference! Here goes

@unclewebb Forgot to mention it, after I saw the PL1 throtteling, I noticed that the PL1 was set to 70W.
Do you remember my situation from a little while ago? I was also PL1 Throtteling and the settings in TS would not affect the Power Throtteling.
I will run some more benchmarks and see if the PL1 persists.

Oh and one more thing, should I have my options on clamp or not?
Thank you, and I can most certainly sleep well with these scores

Even with the clamp option disabled, there is still
PL1 Throtteling at 70W

Good news though! After following the advice,
I killed a bunch of processes, set RealTime and Opened my physical window

All those measures got me a score of 3199 first try!!
All the tries after that did not go above 3187, I wonder why

@AOne @unclewebb
Few things left to ask,
By how much should I decrease my cache undervolt, so I can start increasing the core undervolt? And what increments do you recommend for this process?

About the Power Throttling,
I am able to acces the power limits within the bios, should I try to adjust it there? Or would some sort of bios flashing be able to help me? Is it hardwired? Lot of questions about that part so hoping someone can clarify on that

 

[AOne]

"The Clamp option will force the CPU to throttle as much as necessary to prevent the CPU from exceeding the turbo power limits that you have set. If Clamp is not set, the CPU will only throttle the turbo boost and will not go below the default speed." Copy-pasted from and old unclewebb's post.
So your room temps are pretty good for running a hot laptop (Mine are always between 21-24 in the winter).
Finding the Core uv limit is achieved with tests (I use only TSBench for that). Find where it completes a test without any errors and just in case back it up with 10 mV. Steps of 10 mV are good enough I think. My uv settings are -180/-95 as I don't have the patience for playing too much, finding the sweet spot. Maybe I could push it further, but these values suit me perfectly. I believe your only problem left is the differences between core's temps. All else seems to be perfect, looking at the scores you achieve.

 

[unclewebb]

Do you remember my situation from a little while ago?

Sounds familiar. A laptop with a turbo power limit that has been hard locked by the manufacturer. Not much you can do if that is the case.

I wonder why

It always drives one crazy when you knock off a great Cinebench score and then you are never able to back it up. There is always a random number of tasks doing a random number of things running in the background on any Windows 10 computer. Any full load test is always going to show some random variation from one run to the next. It is impossible to control all of the background tasks.

At a score of 3199, you are in rare territory on the HWBot Cinebench R20 leaderboard. You might not ever get a medal but you are right there near the podium.

Venky`s Cinebench - R20 score: 3387 cb with a Core i7 9750H

The Core i7 9750H @ 4490MHzscores getScoreFormatted in the Cinebench - R20 benchmark. Venkyranks #910 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

When testing voltages, cache at -125 mV has always been a good place to start. I usually suggest bumping the core up in steps of -25 mV. Many get the core to -200 mV without any issues. After that, you can try adjusting either one in small steps of 5mV or 10 mV. I have never owned one of these CPUs so you might already know more about them than I do. See what your CPU likes. Anything you read on the internet might be a good place to start but you need to do your own testing to see if you can improve upon this.

I am able to access the power limits within the bios, should I try to adjust it there?

I would. Set PL1 to 75W or 80W in the BIOS and in ThrottleStop and see if it makes any difference. There might still be a hard coded 70W limit that overrides your BIOS request.