Need guidance undervolting i7 9750H with ThrottleStop

[Ic3m4n86]

Dear TechPowerUp denizens,

This is my first time attempting to undervolt. So please bear with me and my lack of knowledge. I bought a Lenovo Legion Y540 running a Core i7 9750H with a GTX 1660 Ti about 6 months ago. Initially I used XTU (but I didn't change or tinker any settings) just to monitor the temps as I wasn't confident yet with my limited knowledge. Whenever i played Dota 2, I noticed every now and then temps would spike to 92-94C. This was at stock factory settings. So I decided to repaste both the CPU and GPU with Arctic MX4. After the repasting I noticed I didn't get any spikes on temp to 94C. At most it would be around 88C. It did spike to 91C maybe once every 2-3 days.

After watching countless Youtube vids (especially Bob of all Trades), reddit threads and TechPowerUp threads (especially Unclewebb's comments), I decide to give undervolting a go. With the combined advice from all those sources, I used a Core undervolt = -150mV, Cache undervolt = -125mV, Speed Shift = 0, Disabled and Locked Turbo Power Limits, Disabled BD PROCHOT, set my Turbo Long = 70, Turbo Short = 90 and PP0 Current Limit = 140. I only managed to use TS Bench stress test on 12 threads set to Normal, size 768MB and MHz Fixed and it's already Thermal and EDP throttling And the temps are spiking to crazy 94-95C on most cores!! I got freaked out and decided not to tinker any further or use any other stress tests before consulting the experts here.

I'm kinda disappointed actually especially since I've read so much advice and resources of undervolting particularly the Y540 yet my first effort threw up these errors. I've attached my settings printscreen and log file as well. I would very much appreciate any advice or guidance from all. @unclewebb I would very very much appreciate your guidance as well.

Thanks.

 

[unclewebb]

I'm kinda disappointed

Nothing to be disappointed about. Your laptop is running well and your settings look great. Temperatures up to 90°C are normal when gaming. Running a stress test that fully loads your CPU will push the temperature even higher. The TS Bench is nothing compared to the Small FFTs test in Prime95. That would cause massive throttling and a real inferno!

Your disappointment should be reserved for all of the laptop manufacturers that install the 9750H in laptops with barely adequate cooling. They all should have spent a few more bucks and included a better heatsink. Your Lenovo is better than most of its competition. Too much heat is the only thing holding your laptop back from performing even better. The 9750H has a 45 Watt TDP rating. They all seemed to design heatsinks around that spec. When you use ThrottleStop or when a manufacturer sets the turbo power limits much higher, the heatsink cannot dissipate all of the extra heat.

You can become a thermal paste scientist and start experimenting with different thermal pastes and different techniques when applying it. Stay up late at night and try to find the magic one that works best in your laptop. Or, you could say it is good enough as is, sit back and play some games and enjoy your laptop. Intel sets the thermal throttling temperature to 100°C and Lenovo is using an offset value of 6 so your laptop will start to thermal throttle and slow down at 94°C instead of 100°C. This will keep your CPU on the safe side of the fence.

If it was my laptop, I would go into the Options window and on the right hand side, have a look for the PROCHOT Offset variable.

http://imgur.com/aD9WsML

If you do not see the Lock icon above this setting, you can adjust this so your CPU can run a little hotter before it starts to thermal throttle and slow down. Some manufacturers are too conservative with this setting. Intel default for PROCHOT Offset is 0. If you see the Lock icon, this setting cannot be adjusted.

Try running Cinebench R20.

https://www.maxon.net/en-us/products/cinebench-r20-overview/

This is a good test when adjusting voltages. Keep the cache at -125 mV and see if you can get the core up to -175 mV or -200 mV without any errors. This can help shave a couple of degrees off your gaming temps.

Happy gaming.

 

[Ic3m4n86]

Nothing to be disappointed about.

The reason I was disappointed was because I saw so many others with the same hardware as myself without any errors in their Limit Reasons and got a wee bit envious. But I'm glad to hear that my settings actually look great especially from THE guru.

Your disappointment should be reserved for all of the laptop manufacturers that install the 9750H in laptops with barely adequate cooling. They all should have spent a few more bucks and included a better heatsink. Your Lenovo is better than most of its competition.

I truly wasn't aware of this. Perhaps my Arctic MX4 is doing a terrific job. I'm glad to see I didn't botch my first ever repaste. Though I don't think I'll further experiment with new thermal pastes. Judging from your expert opinion it's more than doing well. Haha.

If it was my laptop, I would go into the Options window and on the right hand side, have a look for the PROCHOT Offset variable.

If you do not see the Lock icon above this setting, you can adjust this so your CPU can run a little hotter before it starts to thermal throttle and slow down. Some manufacturers are too conservative with this setting. Intel default for PROCHOT Offset is 0. If you see the Lock icon, this setting cannot be adjusted.

Try running Cinebench R20.

This is a good test when adjusting voltages. Keep the cache at -125 mV and see if you can get the core up to -175 mV or -200 mV without any errors. This can help shave a couple of degrees off your gaming temps.

I'll certainly give these advises a try tonight and see how far i can push the undervolt. I'll post up the results when I'm done to see if you have any further golden nuggets for me. =))

Thank you so much @unclewebb for taking the time with the detailed explanations and advise. Appreciate it.

Dear @unclewebb ,

First set of are at stock settings for CB20. Did it just for reference.

The next set of results are based on a -200mV undervolt and reducing PROCHOT offset to 2. What do you think? Also when you mentioned without any errors did you mean any errors like freezes, lockups or BSODs? Bcoz I didnt experience any. Though I only did the CB20 test once at these settings.

Also I'm guessing this -200mV settings wouldn't work so well on battery power? Any advice on how I can tweak these settings for a battery profile. Just FYI I don't game on battery power. Just regular browsing and watching videos.

 

[unclewebb]

What do you think?

It looks good. By allowing your CPU to run hotter, it can run faster before it starts to thermal throttle and slow down. That is why Intel sets the default PROCHOT Offset to 0. They know their CPUs can run reliably at up to 100°C.

Your PROCHOT Offset setting of 2 tells the CPU to start thermal throttling at 98°C instead of the Intel specified 100°C. That is a reasonable compromise. If you are happy with that setting, I would also check the Lock PROCHOT option. This locks this setting down until the next reboot so ThrottleStop does not have to constantly check and maintain this setting.

I'm glad to see I didn't botch my first ever repaste.

Your maximum CPU core temps show an uneven pattern of high, low, high, low. This is a sign that there is room for improvement when it comes to applying thermal paste. It takes some people a few cracks at it before they get it right.

The heatsink might not be making even contact with the CPU. I have seen some heatsinks that are only attached with 3 screws. This makes it more difficult to get even contact pressure. The heatsink might not be perfectly flat. Different situations might benefit from using different thermal pastes. The peak core temperature is what triggers thermal throttling. If you can balance out the peaks a little better, the CPU can continue running at full speed longer before it starts to thermal throttle. Go buy a handful of tubes of thermal paste and get to work!

Also I'm guessing this -200mV settings wouldn't work so well on battery power?

Why not? Any undervolt that is 100% stable will be stable whether you are plugged in or running on battery power. Less voltage when running on battery power is a good thing. Less voltage equals less power consumption and that will give you longer battery run time.

Your CPU should be stable whether it is running fast or slow. That is why the TS Bench includes a Random MHz option. This test automatically varies the CPU speeds randomly from high to low and every point in between to make sure your computer can handle any situation. If your computer is not stable, it is telling you that you have gone too far and it needs more voltage.

 

[Ic3m4n86]

Dear @unclewebb ,

I decide to make my second attempt at repasting according to your advice. Judging from the photo I attached, the CPU repaste seems decent but looks like I did botch the GPU repaste. A whole corner that's lacking any TIM. Lol. As you said, takes a few tries.

So I decide to maintain the -200mV undervolt, lock the PROCHOT offset at 2, -125mV cache undervolt, TPL settings at 70/90/140 and EPP at 0. After reading some of your comments on other threads about testing 1 or 2 threads for system stability is more important than testing at full loads, I decided to stress test based on those criteria as well. I tested 12T at random and fixed, 1T at random and fixed as well as 2T at random and fixed. And from the printscreens I sent, you can see at I didn't get any limits for 12T at random. But for 12T at fixed for 768M and 6144M it started to constantly throttle PL1(core) and EDP OTHER (ring). Both were blinking yellow to red concurrently. And based on your comments on another thread, this means my Turbo Boost Long is lacking power. I've already set it to 75W while the TDP for this CPU is only 45W. Would you or would you not recommend I push it further?

But as I was about to complete all the TS Bench stress tests and move on to CB20 and P95 to further test my stability, I noticed on HWMonitor that my Core offset had changed from -200 to -125mV on it's own. You can see from my printscreen that the offset was set properly to -200mV. At this point I just got a wee bit annoyed I didn't notice this earlier as I'm not sure how many of these tests were done on the lower undervolt. So I decided not to do the last stress test at 2T fixed and check with you first.

How did the undervolt readjust itself? Because I didn't experience any freezes or BSODs. Did I set something wrongly? Also in the log file, there's something that says POWER STATUS CHANGE. What does that mean? I didn't notice this in my earlier log files. Is this proof that the undervolt change? Your kind advice will be very much appreciated.

On a slightly positive note though, you'll notice that I managed to get my single core Turbo Boost to it's advertised 4.5 GHz (4492 is as close as I can get I guess.)

 

[unclewebb]

I noticed on HWMonitor that my Core offset had changed

The CPU has one register where software reads and writes all of the different voltage information to. When you have more than one monitoring program accessing this one register, you might get some inconsistent results. The monitoring table in the FIVR window shows that your CPU Core offset is set to -0.1992 V (-199.2 mV). ThrottleStop is reading your CPU correctly. You can trust it.

Your log file shows constant POWER STATUS CHANGE notices. That is not right. Are you running some software on your computer from the manufacturer that is trying to keep your battery at a 70% charge level or something like that?

Would you or would you not recommend I push it further?

I always recommend more!!

Your log file shows your CPU PL1 throttling at 70W and your temps now are only in the low 80°C range. Good work on the paste job. That means you can increase your power limit a little more. Maybe try increasing the long turbo limit from 70W to 80W. Try running Cinebench R20 again and see if you have made any improvements.

 

[Ic3m4n86]

The CPU has one register where software reads and writes all of the different voltage information to. When you have more than one monitoring program accessing this one register, you might get some inconsistent results. The monitoring table in the FIVR window shows that your CPU Core offset is set to -0.1992 V (-199.2 mV). ThrottleStop is reading your CPU correctly. You can trust it.

Actually @unclewebb that's my mistake for not making it clear. The printscreen I sent of the FIVR panel was at the point just before I started all the tests. I did have a look at the FIVR panel again when I noticed the drop in undervolt on HWMonitor and it did show an undervolt of -125mV for both core and cache in the FIVR panel. Or have I lost my mind? Do you mind if I run these tests over again and post the results here for your advice?

Your log file shows constant POWER STATUS CHANGE notices. That is not right. Are you running some software on your computer from the manufacturer that is trying to keep your battery at a 70% charge level or something like that?

As a matter of fact I am sir. I'm using the Lenovo Vantage software option that prevents the battery from being charged to more than 55-60% in order to lengthen the lifespan of the battery. Is that a bad thing to have during undervolting stats monitoring?

I always recommend more!!

Your log file shows your CPU PL1 throttling at 70W and your temps now are only in the low 80°C range. Good work on the paste job. That means you can increase your power limit a little more. Maybe try increasing the long turbo limit from 70W to 80W. Try running Cinebench R20 again and see if you have made any improvements.

Haha. I love to see an expert like you being so excited to further motivate us newbies. It feels damn good. But just a tiny doubt to clear though, please do correct me if I'm wrong. Although Intel's TDP for this chip is only 45W, there is technically no fixed limit on how much more power we can pump in as long as the temps stay below the TJmax of 100C. So we can keep pushing till the temps max out just under its safe threshold? Am I right?

 

[unclewebb]

Although Intel's TDP for this chip is only 45W

Most people start getting concerned when they are running WAY beyond the 45W rated TDP. No worries. Intel has been flogging this exact same 14 nm technology in their desktop CPUs for the last 6 years. When the same 6 core CPU is shoved in a box and Intel writes desktop CPU on the box instead of mobile CPU, then this technology comes with a 95W TDP rating. No one thinks twice about running a 6 core desktop CPU up to 150W or beyond while giving it way more voltage compared to when it is running in a laptop. As long as your temps are within spec then I think your CPU will be safe and will live a long life. It is your laptop. If you are worried, set it back to 45W but that does sound kind of boring. Kind of like making your thoroughbred horse pull an apple cart on the weekend.

Or have I lost my mind?

Maybe. You have come this far so you might as well keep going.

I have had my laptop battery at 100% for the last 6 years and it still holds a charge. I discharge it maybe once every 6 months if that. I am not a big fan of software that constantly monitors the battery and constantly tries to keep the battery at a fixed percentage of charge. That just means your laptop is never ready to go mobile because the battery is not fully charged.

Constantly sampling the battery state every 5 seconds and then micromanaging its state seems overkill to me. Batteries are not that expensive. If a battery craps out in 3 or 4 years instead of 4 or 5 years, is that going to be a major expense? Probably not. You are quite capable of changing a battery. If this battery saving feature ever interferes with performance just remember, I told you so.

 

[Ic3m4n86]

Dear @unclewebb ,

So I decided to redo all the testing again after the last "mind lost" session. So I started with the -200mV core offset, -125mV cache offset, PROCHOT offset at 2, TPL at 70/90/140 and EPP at 0.

But this time I was a bit taken aback though, my first effort in the earlier part of this thread never threw up any errors during TS Bench with a -200mV undervolt. It only started to power throttle at the 12T 768M and 6144 fixed MHz stress test. Otherwise no errors. Testing at -200mV undervolt this time started throwing up errors during the 12T 768M random MHz stress test (pic 7). So I had to experiment by reducing the undervolt by -5mV decrements until I arrived at the -130mV undervolt where it didn't throw up any errors anymore. I completed the TS Bench tests for 12T, 2T and 1T for all sizes and speed. No errors at all. No throttling at all as well (pic 6, 10, 12, 14, 16, 19). In pic 19 while I was in the midst of the final TS Bench test there was a power outage at my place. So I think at the exact moment I lost AC power the CPU detected a drop in power the split second before the battery kicked in, hence the EDP OTHER message for all 3.

After the TS Bench tests I did some DOTA 2 gaming at ultra settings (I know DOTA 2 isn't as punishing as other games but I'll test on RDR2 later). This sessions can be seen on the log file at 19/10/2020 around 1915 onward till about 2045. Also no throttling detected (pic 17). I used to get power and thermal throttling even on stock settings I remember.

During the outage I did attempt CB20, all the settings were as I stated above but I think somehow under battery power some settings were overridden as my score was only 2.1k+ (pic 20). I reattempted CB20 when my power was back and got a score of 2.9k+ which is the highest I've managed to get so far ever (pic 22). This session of CB20 did throw up some thermal and power throttling but I think that's to be expected with CB20 am I right? (pic 23)

I attempted Unigine Heaven at maxed out settings and got a score of about 1.7k+ and 70+ avg fps. No throttling detected. (pic 24, 25)

Then finally I ran Prime95 at blend settings because I remember reading in another thread that you mentioned it's better to test for many different situations to reflect real world conditions rather than just at full loads. Ran it for 9 hours from 1am till 10am this morning. You can refer to the log file at these time stamps for further info. No errors, no BSODs, no freezes (pic 26). Though there are plenty of limiting reasons detected as I can see (pic 27). I noticed for the first 2 hours PL1 and EDP OTHER was basically dovetailing together blinking red and yellow constantly. I believe this is the 70W power being limited that we spoke of before. Do you mind explaining to me the rest of the limit reasons that appeared for the first time such as VR Current?

So as a conclusion I believe I'm unable to undervolt to even the -150mV settings that many claim can be done for this CPU. My stable settings are follows :

Core offset : -130mV
Cache offset : -125mV
PROCHOT offset : 2
TPL : 70/90/140
EPP : 0

To be honest, I'm actually satisfied with these settings since I've eliminated most of the throttling issues unless the CPU is pushed to its limit which I believe won't happen so often. Temps have almost maxed out at 97C (pic 28) which I'm actually quite happy with and pushing more wattage into it further will just increase the temps and cause it to throttle when it reaches the 98C limit I set. But I did read your comments elsewhere that said that running a CPU at its limit of stability risks having it becoming unstable once in awhile so I was thinking I will settle on a -125mV offset for both core and cache. What do you think?

What are your thoughts overall? I would really appreciate your feedback. Thanks.

 

[unclewebb]

So I decided to redo all the testing again

You hereby win the award for the most excessive post. 28 pictures and 1 log file. Incredible. I love it.
Maybe you are the reason the TPU image server crashed recently!

while I was in the midst of the final TS Bench test there was a power outage at my place

Did you ever think that the load you are putting on your CPU is putting a hurting on the power grid? The police might drop by for a visit. They probably think you got a grow op going somewhere in the basement.

VR Current means the voltage regulator is complaining because it is being asked to supply too much current to your CPU. Prime95 is an excessive test for most laptops and desktop computers. I am not sure if any tweaks in ThrottleStop will solve this problem. For once, I do not recommend that you try to go further to find out.

Now the bad news. When the TS Bench starts reporting errors, I would start by changing the cache offset voltage. Lots of 9750H are not stable with the cache at -125 mV. I would set that at -100 mV and then see if you can go back up on the core towards -200 mV without the TS Bench reporting any errors. It is a balancing act with these two voltages. It is typically the cache that is the limiting factor.

You seem to like testing. Some more testing should keep you busy for a day or two. I think I need to keep a link to this thread so all this info does not go to waste.

Now I have a 4.6 MB TS log file that I need to study before bedtime.

 

[HenryCase]

This is the type of content I came here for.

 

[Ic3m4n86]

You hereby win the award for the most excessive post. 28 pictures and 1 log file. Incredible. I love it.
Maybe you are the reason the TPU image server crashed recently!

Hahahahaha. Maybe I was. Well you mentioned in another post that more data is always best. So I tried to provide as much as I could.

Did you ever think that the load you are putting on your CPU is putting a hurting on the power grid? The police might drop by for a visit. They probably think you got a grow op going somewhere in the basement.

I wouldn't be surprised if I was. At points I saw it was pushing past 80 watts! It was probably pulling power away from my alleged grow op!!

VR Current means the voltage regulator is complaining because it is being asked to supply too much current to your CPU. Prime95 is an excessive test for most laptops and desktop computers. I am not sure if any tweaks in ThrottleStop will solve this problem. For once, I do not recommend that you try to go further to find out.

I suppose this is where my limit ends. Besides, VR Current only ever pops up during P95 stress testing so I highly doubt I'll get it again while merely doing my usual mundane day to day tasks.

You seem to like testing. Some more testing should keep you busy for a day or two. I think I need to keep a link to this thread so all this info does not go to waste.

Oh I hate it. I really hate testing. Takes up too much of my time. But unfortunately I'm a wee bit OCD about these things so I'm forced to go all in to find out the best undervolt settings for my laptop.

Now the bad news. When the TS Bench starts reporting errors, I would start by changing the cache offset voltage. Lots of 9750H are not stable with the cache at -125 mV. I would set that at -100 mV and then see if you can go back up on the core towards -200 mV without the TS Bench reporting any errors. It is a balancing act with these two voltages. It is typically the cache that is the limiting factor.

Ok so I'll spare the TPU image servers a wee bit by just listing down the main adjusted settings here :

Core offset : -200mV
Cache offset : -100mV
EPP : 0
BD PROCHOT : unchecked
Disable Turbo : unchecked
PROCHOT offset : 2
TPL Long : 70
TPL Short : 90
PP0 limit : 140

So @unclewebb you were right. From the test results, seems that the cache offset was the limiting factor!! With the updated settings, I stress tested with TS Bench at 12T, 2T and 1T at all sizes and speeds. No errors at all. Though I was surprised to see some thermal throttling at 768M and 6144M at fixed MHz (pic 2, 3). But otherwise no other throttling detected for any other of the tests (pic 5, 7, 9, 11, 14).

I did some DOTA 2 gaming for about 2 hours (log file time 1815 to 2045), no throttling detected as well (pic 12). I know I took a risk here playing an online game without properly tested settings but hey, no risk no reward right?

Next up was CB20. I got my highest score so far which was almost 3.1k (pic 15). But this time I'm happy to report no thermal throttling during the CB20 benchmarking (pic 16). Only PL1 which is to be expected from the 70W power limit I set.

I benchmarked with Unigine Heaven next. No issues. 1.8k score and 70+ fps on maxed out settings (pic 17). No throttling detected here as well (pic 18).

And finally I ran Prime95 (blend settings) for 9 hours with no errors or warnings detected (pic 19). Though within 15 minutes of running P95 I noticed that PL1 and EDP were not just blinking red but was constant, which meant throttling was in progress (pic 20). Since the max temps at that point was only 94C, I decided to bump the TPL Long to 80W and see what happens. Lo and behold it wasn't power throttling anymore. But sadly temps spiked to 99C fairly quickly even with my offset set at 98. I bumped it down to 75W and temps maxed out at around 97-98C. So i finally decided to stick to 70W as I rather it power throttles ( which will only happen sometimes as my laptop won't be running at full loads constantly) than constantly thermal throttle which I'm not at all comfortable with. But this morning when I stopped P95 I noticed there was thermal throttling up to 99C even at 70W (pic 21), so yea I guess I won't be bumping up the TPL Long any further. And yes I'm aware that I can experiment with other TIMs like conductive pastes or even liquid metal but no thanks. I'm not that much of a risk taker and I'm just too lazy for that. LOLOL.

So I'm happy to say I'm very very satisfied with my first foray into undervolting thanks in large parts to @unclewebb and many others who've contributed from Youtube and reddit. Oh and yea I'm ecstatic that I can push my undervolt to -200mV while keeping the system stable!! But as I know systems at the very edge of stability always run the risk of randomly becoming unstable even with the slightest ambient changes, I will be bumping down my undervolt to perhaps -185mV so I'm safely away from the edge.

These will be my final undervolt settings for performance :

Core offset : -185mV
Cache offset : -100mV
EPP : 0
BD PROCHOT : unchecked
Disable Turbo : unchecked
PROCHOT offset : 2
TPL Long : 70
TPL Short : 90
PP0 limit : 140

@unclewebb any feedback from you on these results will be very much appreciated.

Now I have a 4.6 MB TS log file that I need to study before bedtime.

Dear guru @unclewebb when you can spare the time, do you mind studying my last and current log to see if there's anything out of place? Thanks a million!!

 

[unclewebb]

I saw a couple of times in your previous log file, VRTDC. I have never seen that one before so I had to go look it up.

VR Thermal Design Current Log When set, indicates that the VR TDC Status bit has asserted since the log bit was last cleared.

That is when I concluded that I think you have pushed the envelope just about as far as one should push it. Many desktop users avoid stress testing with Prime95. A 7 hour Prime95 torture test while you are having a good night's sleep is kind of like asking a person to run a back to back triathlon.

Your last log file shows lots of PL1 power limit throttling right at the 70W value that you set so your CPU is running exactly as you asked it to run. With your new and improved cooling maybe you could get away with bumping that up to 75W but nothing wrong with 70W. There is definitely no shame in running a CPU with a 45W TDP rating at 70W. If you increase this to 75W, you might start seeing VRTDC more frequently in the log file. At 70W, you seem to be staying on the safe side of the fence.

I can push my undervolt to -200mV while keeping the system stable!!

Core offset : -200mV
Cache offset : -100mV

You came up with these magic settings based on lots and lots of user testing and then,

I will be bumping down my undervolt to perhaps -185mV

WHY????
Your computer was running great at -200 mV. You did way more testing compared to most users. I was going to suggest giving -220 mV a try.

It is the cache offset voltage that causes stability problems. Once the cache is set correctly, the CPU core seems to ignore any excessive voltage request. Going beyond approximately -220 mV for the core will be ignored. Cinebench results usually level out at this point. That means you can set the core to -300 mV or -500 mV or even -1000 mV. ThrottleStop and HWiNFO will report this ridiculous value but that does not mean that the CPU is using the full -1000 mV. Some have come up with the theory that a 2:1 ratio is the magic number but you might be able to go a little more than that.

Go ahead. One more test. How about Cinebench R20 at -220 mV on the core? Is it good for a couple of more Cinebench points? I think -200 mV or -220 mV are probably fine for long term use.

Intel XTU forces the core and cache voltages to be adjusted equally. There are 101 YouTube guides that all recommend doing this. ThrottleStop lets you adjust these two voltages individually. Users with 8th, 9th and 10th Gen mobile CPUs all seem to get a boost in performance or better temperatures by setting the core to a bigger number compared to the cache. Your results confirm that it is impossible to get maximum performance when using XTU. Maybe Intel's engineers should try doing some testing while using ThrottleStop.

 

[Ic3m4n86]

I saw a couple of times in your previous log file, VRTDC. I have never seen that one before so I had to go look it up.


That is when I concluded that I think you have pushed the envelope just about as far as one should push it. Many desktop users avoid stress testing with Prime95. A 7 hour Prime95 torture test while you are having a good night's sleep is kind of like asking a person to run a back to back triathlon.

Your last log file shows lots of PL1 power limit throttling right at the 70W value that you set so your CPU is running exactly as you asked it to run. With your new and improved cooling maybe you could get away with bumping that up to 75W but nothing wrong with 70W. There is definitely no shame in running a CPU with a 45W TDP rating at 70W. If you increase this to 75W, you might start seeing VRTDC more frequently in the log file. At 70W, you seem to be staying on the safe side of the fence.

Yes I agree. 70W is certainly the way to go. I'll explain further below. Also I noticed my stock TPL settings from Lenovo was already set at 70 Long but 107 Short.

You came up with these magic settings based on lots and lots of user testing and then,
WHY????
Your computer was running great at -200 mV. You did way more testing compared to most users. I was going to suggest giving -220 mV a try.

Well I was just trying to heed your advice from other threads that CPUs running at the edge of stability will risk failing every now and then somewhere in the future. But turns out I'm actually not at the edge as the results below will show. LOL.

Intel XTU forces the core and cache voltages to be adjusted equally. There are 101 YouTube guides that all recommend doing this. ThrottleStop lets you adjust these two voltages individually. Users with 8th, 9th and 10th Gen mobile CPUs all seem to get a boost in performance or better temperatures by setting the core to a bigger number compared to the cache. Your results confirm that it is impossible to get maximum performance when using XTU. Maybe Intel's engineers should try doing some testing while using ThrottleStop.

Maybe Intel should hire you as their Chief Test Engineer. What say you?

It is the cache offset voltage that causes stability problems. Once the cache is set correctly, the CPU core seems to ignore any excessive voltage request. Going beyond approximately -220 mV for the core will be ignored. Cinebench results usually level out at this point. That means you can set the core to -300 mV or -500 mV or even -1000 mV. ThrottleStop and HWiNFO will report this ridiculous value but that does not mean that the CPU is using the full -1000 mV. Some have come up with the theory that a 2:1 ratio is the magic number but you might be able to go a little more than that.

Go ahead. One more test. How about Cinebench R20 at -220 mV on the core? Is it good for a couple of more Cinebench points? I think -200 mV or -220 mV are probably fine for long term use.

Ok so today's results seem to have thrown a curve ball at me. Yesterday while running CB20 at -200mV, I only experienced some power throttling. But today at -220mV, I almost immediately started experiencing thermal throttling instead (pic 2). Which is strange considering there's less voltage being fed to the CPU. Also the CB20 score is lower than my -200mV score yesterday (pic 1). @_@ Then I figured maybe it's my ambient temps affecting the results because I had my AC turned off. With the ambient temps back to a breezy 23C, I ran CB20 at -220mV again. Same results. Almost identical CB20 score (pic 4). And thermal throttling yet again (pic 5) (please take note of this as it'll be the theme for the rest of the test results).

Then I decided to attempt my tried and tested -200mV and run CB20. This time the score was lower than yesterday's by about 50 points even at identical settings (pic 9). Also thermal throttling yet again when yesterday's test at similar settings only triggered power throttling (pic 10). @_@ Unsatisfied with these lesser results, I ran CB20 again at -200mV and got within 25 points of yesterday's result (pic 12). Thermal throttling yet again (pic 13).

I was still not happy with the results so I ran CB20 again at -220, -200, -180, -160, -150 and -125mV. This time, -180mV came closest to my best score of 3087 with 3070. Just 17 points shy. And as with earlier tests, all undervolt settings experienced thermal throttling (pic 16, 19, 22, 25, 28, 31). Temps were generally higher across the cores at the lower undervolts (pic 17, 20, 23, 26, 29, 32), which is to be expected. And CB20 scores generally started to dip with undervolts lesser than -180mV (pic 15, 18, 21, 24, 27, 30). So basically the 'sweet spot' was between -180mV to -200mV of undervolt which returned a score of between 3030 and 3090 for an average of about 3060 (which is actually higher than a 12 core Xeon CPU!!).

After the CB20 runs, I did a couple of hours of Dota 2 on Ultra settings. No throttling detected (pic 33) and temps maxed out at a chilly 82C (pic 34). I clearly notice a vast difference in my Dota 2 sessions prior to and after undervolting. At stock factory settings even just after boot up I notice a plethora of throttling in the Limit Reasons window from BDPROCHOT, Thermal, PL1, to EDP Other. And while gaming my fans would be at full speed and basically sounding like a freaking jet engine getting ready to take off. But with the -200mV undervolt, it's all silent, cool and doesn't throttle at all.

Though just one thing bugs me about the results today which is the thermal throttling during CB20. Still no idea why...

But otherwise the following settings is going to be my permanent settings for the rest of this laptop's lifetime :

Core offset : -200mV
Cache offset : -100mV
EPP : 0
BD PROCHOT : unchecked
Disable Turbo : unchecked
PROCHOT offset : 2
TPL Long : 70 (any higher and I just start to thermal throttle)
TPL Short : 90
PP0 limit : 140

Please do take note that this settings are what work for me and my laptop after copious amounts of testing. Even if you're running the same spec and laptop (Lenovo Legion Y540) as mine, these settings won't necessarily work for you. It would be best to do your own testing to determine the best settings for your laptop.

Guru @unclewebb , your thoughts on tonight's results?

 

[unclewebb]

just one thing bugs me about the results today which is the thermal throttling during CB20

Guess what this means? Time to redo the thermal paste, again!

Look at your max temp screenshots. There is an almost 25°C temperature difference between 2 cores sitting beside each other. That is never a good sign.

Arctic MX4

Some users on this forum with similar mobile CPUs have not had great results with this paste. Sure it works great at first but for long term use on a laptop, it does not. Keep an eye on your temperatures to see if they get worse during the next week. Noctua NT-H2 has been recommended for better longevity in a laptop. I have not done any hands on testing in recent memory so start doing some Google searching for advice. Mobile CPUs and desktop CPUs are different. Mobile CPUs do not have a heatspreader on top of the cores so a thermal paste that works well on a desktop CPU might not work great long term on a mobile CPU.

This is the perfect hobby for anyone with OCD. Always something that needs to be tweaked. Play some games for a week and get some enjoyment out of your laptop before it is time to paste again. As always, thanks for sharing your results.

 

[Ic3m4n86]

Guess what this means? Time to redo the thermal paste, again!

Look at your max temp screenshots. There is an almost 25°C temperature difference between 2 cores sitting beside each other. That is never a good sign.

I thought I did a proper job the second time around but yea the gap in temps keep persisting... Ugh...

Some users on this forum with similar mobile CPUs have not had great results with this paste. Sure it works great at first but for long term use on a laptop, it does not. Keep an eye on your temperatures to see if they get worse during the next week. Noctua NT-H2 has been recommended for better longevity in a laptop. I have not done any hands on testing in recent memory so start doing some Google searching for advice. Mobile CPUs and desktop CPUs are different. Mobile CPUs do not have a heatspreader on top of the cores so a thermal paste that works well on a desktop CPU might not work great long term on a mobile CPU.

This is the perfect hobby for anyone with OCD. Always something that needs to be tweaked. Play some games for a week and get some enjoyment out of your laptop before it is time to paste again. As always, thanks for sharing your results.

@unclewebb you are a bad bad influence on me. I was so determined to not further experiment on TIMs anymore but after reading this reply I immediately went online and ordered a tube of Noctua NT-H2.

Though the Noctua is double the price of the MX4 at 12 dollars for a tube of 4 grams. I really hope it's worth it. LOL.

And again, thanks for always being so supportive.

 

[nguyen]

Well what you are experiencing is the pump out effect of TIM. At high load temp the TIM will liquidfy and squeeze out, creating gap between the core and the heatsink.
So for laptop the higher viscosity of the TIM the better, right now Kingpin KPx and Thermalright TF-X are probably the best suited for laptop.
NT-H2 might behave the same as MX4 because it's very liquidy (low viscosity).

One way to mitigate is after you repaste and mount the heatsink, use hair dryer to heat up the heatsink (thus heat up the TIM for a few minutes) and then tighten the screws a little more, this way you can reduce the gap between core and heatsink after the pump out.

 

[Ic3m4n86]

Well what you are experiencing is the pump out effect of TIM. At high load temp the TIM will liquidfy and squeeze out, creating gap between the core and the heatsink.
So for laptop the higher viscosity of the TIM the better, right now Kingpin KPx and Thermalright TF-X are probably the best suited for laptop.
NT-H2 might behave the same as MX4 because it's very liquidy (low viscosity).

Hmm...First time I'm hearing of the pump out effect. Does make sense though. Since I've already ordered the Noctua perhaps I'll try it out first for a few weeks then maybe I might switch to the Kingpin or Thermalright to see which works best.

One way to mitigate is after you repaste and mount the heatsink, use hair dryer to heat up the heatsink (thus heat up the TIM for a few minutes) and then tighten the screws a little more, this way you can reduce the gap between core and heatsink after the pump out.

I'll try this with the Noctua and see how it works. Thanks brother @nguyen !!

 

[nguyen]

Yeah Der8auer talk about it here, many users in this forum also experienced this pump out effect first hand with Kryonaut, myself included

 

[Ic3m4n86]

Dear @unclewebb , is there any way to set different Turbo Power Limits for different profiles? Right now I have 3 profiles, AC, Battery and Low Battery. I can set different voltages for different profiles, but I can't seem to do that for Turbo Power Limits.

 

[nguyen]

Probably not, but I would tick the "Disable Turbo" box with the Battery Profile, laptop barely feel slower at all, combine with higher EPP value and you can extend the battery life a fair bit.

 

[Ic3m4n86]

Probably not, but I would tick the "Disable Turbo" box with the Battery Profile, laptop barely feel slower at all, combine with higher EPP value and you can extend the battery life a fair bit.

Yea I thought so too @nguyen . I tried to tinker with the settings but couldn't find a way to do that. But yea I did set a higher EPP value and Disabled Turbo for my battery profile. Thanks.

 

[Ic3m4n86]

Dear @unclewebb & @nguyen ,

Just repasted for a 3rd time this afternoon with Noctua NT-H2 this time. And I'm seeing some crazy drop in temps. Yesterday while still on MX4, I had max temps of about 80-82C while gaming on Dota 2 (refer to log 25/10 at 23:19:43). Today after the repaste, the max temp I reached was 71C (refer to log 26/10 at 20:56:18 as well as pic 1)!! Am I crazy or is Noctua really that much better??

*Update

I decided to run CB20 test after I posted the above. Got my second highest score of 3078, just 9 points short of my best (pic 2). And this time, no throttling detected. Not even power or thermal (pic 3). And max temp reached was only 83C (pic 4)!!. And this time the difference in temps between cores were at most 5C from each other. Unlike previously where there were 20C difference at times.

Maybe I've yet to experience the pump out effect since the paste is still fresh? I did as nguyen adviced and heated the heatsink with a hairdryer before further tightening the screws.

@nguyen any thoughts?

Any and all feedback is highly appreciated.

 

[nguyen]

Jup, cores temp are so close to each other indicate that the NT-H2 is working wonderfully. Max temp of only 83C with PL of 80W (I supposed PL2 is 80W ?) is really good.

Just save your TS core temp somewhere so after a while you can come back and compare the result.

Well when you have lower load temperature, there would be less pump out effect. So yeah this would last you much longer than MX4 did

 

[Ic3m4n86]

Jup, cores temp are so close to each other indicate that the NT-H2 is working wonderfully. Max temp of only 83C with PL of 80W (I supposed PL2 is 80W ?) is really good.

Just save your TS core temp somewhere so after a while you can come back and compare the result.

Well when you have lower load temperature, there would be less pump out effect. So yeah this would last you much longer than MX4 did

Yea I'm really pleased to see the core temps so close to each other. Used to really irk me to see a 20C difference. And I'm honestly surprised to see the max power draw at 81W. Used to only see a max of 71-72. Though my PL1 is set to 70 and PL2 set to 90.

Yea I've saved the log file so I can use it for later comparisons. I really am very happy with the current core temps.

And yea that makes sense. Lower temps, less pump out. I'm hoping these figures stay as they are so I don't have to repaste again.

On a side note, I am quite disappointed by the MX4 though. Many reviews I read had only good things to say about it. Feels like I got conned. Lol.