Alienware m16 r1 - Undervolt & PTM7950

[Alienware_Melting]

Hi everyone, how you doing?

TL;DR: are wide differences in temperature between different Cores (up to 20° Celsius) indicative of bad application of PTM 7950?

Now, some context, I started this thread here months ago to learn how to unlock and undervolt my laptop in the title.

Loading…

www.techpowerup.com

I managed to achieve what seemed to be stable -144 touching only CPu and cache, not the other stuff.

Months later I decide to get tools and open the inverted MB to repaste the CPU and GPU. I dind't change any thermal pad and I applied PTM7950 as instructed by the maker: apply it to the heatsink. I dind't make a blob, I dind't use just a little bit, I covered the entire area which is including also plenty of leftovers. I've one need many heat cycles to get the better out of PTM7950. I already did that, for aa week, plenty of Cinebench stress testing.

Now, the GPU I cannot complain, it stays cool enough never going much above 70C, but the CPU goes fast to close to 100, but half the cores are low 80s, almost the other half mid 80s, and 4 cores always around 10C above the second highest temps. Sometimes 79 vs 99 in total difference. I'm drawing close to 165 Watts but not quite, because the undervolt is now -160 (but I'm not sure it is stable); I believe the full wattage is never reached, it tops at ~157.8 Watts, but it triggers the POWER limit anyways (which is 165 Watts).

Anyways, I also notices that the FID for the lower temps goes 47x during Cinebench multicore, but stays are 37x for the higher temps, but even for those not in the 4 core group that heats like f*ck, it also stays 37x.

Anyway, not sure how @unclewebb made the core sorting algorithm.

Finally, yes my Cinebench score is better for the CPU, from 1705 with -75 undervolt and original paste to ~1950s. When I stop the test, temps fall almost instantly some 30C, so low 60s, but take a couple of minutes of top fan speed to go back to low 50s. GPU score in Cinebench is not that greatly affected by PTM7950 application alone.

So, here's the dilemma, to open this F$#@*% machine again will be such a PAIN since the wireless antennas are absolutely SoBs to connect both, in the 16 inches model, and close the entire thing with them not unplugging. This literally took me 90% of the entire service time, no joke.

Ok so here's some screenshots:

ps: Oh I've also changed IccMax from 230 to 240 amps but I'm not sure that does anything special and yeah now I noticed there are 2 wattage limits as in the last pic, MMIO is 157, MSR is the one 165. I don't know what those mean

 

[ChemicalDruid]

Regarding the some-cores-run-hotter-than-others issue, it seems to be either bad paste/pad application or just silicon lottery. I cannot judge your application so if you think it's good enough, then it's just silicon lottery left to explain things.

I might be wrong here - if anyone else wants to jump in, feel free.

Some more definitive advice follows:

Now, the GPU I cannot complain, it stays cool enough never going much above 70C, but the CPU goes fast to close to 100, but half the cores are low 80s, almost the other half mid 80s, and 4 cores always around 10C above the second highest temps. Sometimes 79 vs 99 in total difference. I'm drawing close to 165 Watts but not quite, because the undervolt is now -160 (but I'm not sure it is stable); I believe the full wattage is never reached, it tops at ~157.8 Watts, but it triggers the POWER limit anyways (which is 165 Watts).

No, the true PL is the lower of MSR and MMIO, as @unclewebb has stated several times in other posts. The proof: you stated it tops at 157.8 and your MMIO is set to 157.
Lesson: lock MMIO in Turbo Power Limits so that the (higher) MSR values (of 165 in your case) are used.

ps: Oh I've also changed IccMax from 230 to 240 amps but I'm not sure that does anything special and yeah now I noticed there are 2 wattage limits as in the last pic, MMIO is 157, MSR is the one 165. I don't know what those mean

Use the highest IccMax value possible, which should be 511.75 for your CPU (also endorsed by @unclewebb)
Besides what MSR and MMIO mean (which you can easily find yourself with a web search) what matters the most is that the CPU will use the lower of those figures in the pairs of PL1 and PL2. As stated above, if you Lock MMIO, 165 will be the limit for both, otherwise it will be 157 as is already happening for you.

 

[FreedomEclipse]

Regarding the some-cores-run-hotter-than-others issue, it seems to be either bad paste/pad application or just silicon lottery

Silicone lottery. Just like on some chips you'll have one or two cores that wont OC as well as all the others. If PTM is keeping the chip cool. There is nothing else the OP can do short of trying to replace his current chip with another which isnt going to be the easiest thing to do since its a laptop.

 

[Alienware_Melting]

Regarding the some-cores-run-hotter-than-others issue, it seems to be either bad paste/pad application or just silicon lottery. I cannot judge your application so if you think it's good enough, then it's just silicon lottery left to explain things.

I might be wrong here - if anyone else wants to jump in, feel free.

Some more definitive advice follows:

No, the true PL is the lower of MSR and MMIO, as @unclewebb has stated several times in other posts. The proof: you stated it tops at 157.8 and your MMIO is set to 157.
Lesson: lock MMIO in Turbo Power Limits so that the (higher) MSR values (of 165 in your case) are used.


Use the highest IccMax value possible, which should be 511.75 for your CPU (also endorsed by @unclewebb)
Besides what MSR and MMIO mean (which you can easily find yourself with a web search) what matters the most is that the CPU will use the lower of those figures in the pairs of PL1 and PL2. As stated above, if you Lock MMIO, 165 will be the limit for both, otherwise it will be 157 as is already happening for you.

Thnak you for the reply. Not sure I'll meddle with such a much higher IccMax right now, this is the default value for the P Cache, but the default for Core was 230, so you know the vormula V = R*I right? So this would raise voltages almost double at some curve point or whatever, but I'm undervolting?! Let's see what uncle says.

Concerning the differences, I think a vid is better to visualize the differences in temp and multiplier I'm talking about. I had to zip the mp4 to upload here, it is attached.

EDIT:
Ok so I've changed IccMAX all the way up to 511.75 Amps for Core, enabled Power Limit Controls to reach 165 wattage in both PL1 and PL2 and checked "sync MMIO" which I assume makes MMIO == MRS.

Nothing changed, benchmark even worse and higher temps touching 100C pretty quickly.

I'm only touching Core and P Cache, all the rest equal.

 

[unclewebb]

you know the formula

The settings in ThrottleStop are limits. If a CPU needs 180 Amps to run at full speed, whether you set the IccMax limits to 200, 300, 400 or 500 will make zero difference. There will be no EDP throttling caused by the IccMax power limit. I like to compare these limits to a limbo bar. If you set the bar 10 feet high or 100 feet high, does it make any difference? No it does not.

To avoid throttling, I always recommend setting any of the limits sky high. My only goal is maximum performance. If you think increasing any of these limits could damage your CPU then do not do it. Recent Intel CPUs have proven to be somewhat fragile so proceed with caution. If you choose to adjust IccMax, set the core and the P cache to the same value.

sync MMIO

I prefer checking the MMIO Lock box and not checking Sync MMIO. Your video shows that your laptop is still throttling at 157W. Most Dell laptops use an embedded controller to enforce a maximum power limit. If Dell has set this maximum to 157W, which is very likely, adjusting the MSR power limits higher in ThrottleStop or using MMIO Lock or Sync MMIO will not accomplish anything. If this is the case, that means Dell is in charge of the maximum power of your CPU, not ThrottleStop. There is no easy solution for that. Some like to mod the BIOS by changing the IMON Slope variable which can trick the CPU into reporting less power consumption. This can help keep the CPU under any EC power limit that might have been set. I have never done this so do a Google search if you want to learn more.

To be honest, I would not waste time trying to increase this power limit or messing with IMON Slope. Your CPU is already running at 94°C and there are both P and E cores occasionally reaching the full 100°C thermal throttling temperature. The heatsink is unable to dissipate any more heat. You have reached the thermal wall. Your undervolt is already within a hair of the max possible. Your voltage settings might already be over the edge. If you get a crash / BSOD during normal use or while playing a game then you need to reduce your undervolt or try increasing the mV Boost value by 50. A BSOD when undervolting means the CPU needs more voltage.

Nothing changed, benchmark even worse

When a CPU is running on the edge and is sometimes thermal throttling, your benchmark scores and overall day to day performance might not be consistent. Slight changes to the room temperature or which way the wind is blowing might change how much throttling your CPU needs to do to stay at a safe temperature. Just another sign that you are at the thermal wall. Raising the various limits does not accomplish anything in this situation. More power equals more heat equals more thermal throttling. It is like a dog chasing its tail.

PAIN

If changing the thermal paste is a painful experience then I would not bother. If the heatsink Dell chose to use can only dissipate 160W of heat and if they set an EC power limit of 157W then you are already getting 99% of the possible maximum performance out of your laptop. Spending days searching for 99.5% does not sound like a good use of one's time.

Most people seem to use PTM 7950 thermal pads. I think this is easier than trying to decide how much thermal paste should be used and how it should be spread. Search Amazon for PTM7950 thermal pad. Maybe next time.

 

[PHVM_BR]

Thnak you for the reply. Not sure I'll meddle with such a much higher IccMax right now, this is the default value for the P Cache, but the default for Core was 230, so you know the vormula V = R*I right? So this would raise voltages almost double at some curve point or whatever, but I'm undervolting?! Let's see what uncle says.

Concerning the differences, I think a vid is better to visualize the differences in temp and multiplier I'm talking about. I had to zip the mp4 to upload here, it is attached.

EDIT:
Ok so I've changed IccMAX all the way up to 511.75 Amps for Core, enabled Power Limit Controls to reach 165 wattage in both PL1 and PL2 and checked "sync MMIO" which I assume makes MMIO == MRS.

Nothing changed, benchmark even worse and higher temps touching 100C pretty quickly.

I'm only touching Core and P Cache, all the rest equal.

Which performance mode was selected at AWCC?

Are you using a laptop stand to move air vents away from the table top?

I have been using PTM 7950 in the phase change pad format for a few years and in different models and I have always applied it directly to the die and always with good results.

The temperature difference between the processor cores is normal, but with a good application of PTM 7950 (phase change pad) the difference should be smaller.

Up to about 8 to 10°C between P-cores and up to about 3 to 4°C between E-cores.

Generally 2 P-cores work hotter than others and generally 8 E-cores work a little less hot than other 8 in the case of i9.

On my Precision 7540, KryoSheet managed to deliver more than PTM 7950, around 10-12% more thermal headroom.

I tested KryoSheet on a new laptop, Inspiron 16 Plus 7640, and KryoSheet delivered less than PTM 7950.
I believe it is due to lack of adequate contact between the heatsink and the CPU die.

 

[Alienware_Melting]

The settings in ThrottleStop are limits. If a CPU needs 180 Amps to run at full speed, whether you set the IccMax limits to 200, 300, 400 or 500 will make zero difference. There will be no EDP throttling caused by the IccMax power limit. I like to compare these limits to a limbo bar. If you set the bar 10 feet high or 100 feet high, does it make any difference? No it does not.

To avoid throttling, I always recommend setting any of the limits sky high. My only goal is maximum performance. If you think increasing any of these limits could damage your CPU then do not do it. Recent Intel CPUs have proven to be somewhat fragile so proceed with caution. If you choose to adjust IccMax, set the core and the P cache to the same value.


I prefer checking the MMIO Lock box and not checking Sync MMIO. Your video shows that your laptop is still throttling at 157W. Most Dell laptops use an embedded controller to enforce a maximum power limit. If Dell has set this maximum to 157W, which is very likely, adjusting the MSR power limits higher in ThrottleStop or using MMIO Lock or Sync MMIO will not accomplish anything. If this is the case, that means Dell is in charge of the maximum power of your CPU, not ThrottleStop. There is no easy solution for that. Some like to mod the BIOS by changing the IMON Slope variable which can trick the CPU into reporting less power consumption. This can help keep the CPU under any EC power limit that might have been set. I have never done this so do a Google search if you want to learn more.

To be honest, I would not waste time trying to increase this power limit or messing with IMON Slope. Your CPU is already running at 94°C and there are both P and E cores occasionally reaching the full 100°C thermal throttling temperature. The heatsink is unable to dissipate any more heat. You have reached the thermal wall. Your undervolt is already within a hair of the max possible. Your voltage settings might already be over the edge. If you get a crash / BSOD during normal use or while playing a game then you need to reduce your undervolt or try increasing the mV Boost value by 50. A BSOD when undervolting means the CPU needs more voltage.


When a CPU is running on the edge and is sometimes thermal throttling, your benchmark scores and overall day to day performance might not be consistent. Slight changes to the room temperature or which way the wind is blowing might change how much throttling your CPU needs to do to stay at a safe temperature. Just another sign that you are at the thermal wall. Raising the various limits does not accomplish anything in this situation. More power equals more heat equals more thermal throttling. It is like a dog chasing its tail.


If changing the thermal paste is a painful experience then I would not bother. If the heatsink Dell chose to use can only dissipate 160W of heat and if they set an EC power limit of 157W then you are already getting 99% of the possible maximum performance out of your laptop. Spending days searching for 99.5% does not sound like a good use of one's time.

Most people seem to use PTM 7950 thermal pads. I think this is easier than trying to decide how much thermal paste should be used and how it should be spread. Search Amazon for PTM7950 thermal pad. Maybe next time.

Hi Uncle, thanks for the answer.

Yes, as per OP, I'm using PTM 7950 already. My main question is whether the temp difference in cores is due to possible bad application. I think I'm just gonna open it up again and repaste since PTM 7950 can be reused.

And yes, it did settle down better after some more time. The temps subside faster and average low usage temps are also quite ok, in the 40s, low 50s sometimes. The thing is really temp diff is too high, which makes it throttle faster. I'll be probably allowing it to reach only 160 watts and be done with it. I'm stable at -158 volts and 800 boost ==159 volts seems to be ok too. If I notch it beyond -160 it fails.

Which performance mode was selected at AWCC?

Are you using a laptop stand to move air vents away from the table top?

I have been using PTM 7950 in the phase change pad format for a few years and in different models and I have always applied it directly to the die and always with good results.

The temperature difference between the processor cores is normal, but with a good application of PTM 7950 (phase change pad) the difference should be smaller.

Up to about 8 to 10°C between P-cores and up to about 3 to 4°C between E-cores.

Generally 2 P-cores work hotter than others and generally 8 E-cores work a little less hot than other 8 in the case of i9.

On my Precision 7540, KryoSheet managed to deliver more than PTM 7950, around 10-12% more thermal headroom.

I tested KryoSheet on a new laptop, Inspiron 16 Plus 7640, and KryoSheet delivered less than PTM 7950.
I believe it is due to lack of adequate contact between the heatsink and the CPU die.

Hi PHVM.

Yes, laptop up, etc.

PTM 7950 is working very well for almost all cores, but the last 4. I'd not think about changing it if not for these, temps are sometimes almost at 20C gap. I applied to the heatsink becuase the maker said to do this, this time I'll remodel to cover the processor.

Since they say we can remodel PTM 7950, I'll do this.

I don't care much about maximum performance, I just didn't want to ever need to repaste it again, so it is PTM 7950 for me in this machine.

 

[PHVM_BR]

Hi Uncle, thanks for the answer.

Yes, as per OP, I'm using PTM 7950 already. My main question is whether the temp difference in cores is due to possible bad application. I think I'm just gonna open it up again and repaste since PTM 7950 can be reused.

And yes, it did settle down better after some more time. The temps subside faster and average low usage temps are also quite ok, in the 40s, low 50s sometimes. The thing is really temp diff is too high, which makes it throttle faster. I'll be probably allowing it to reach only 160 watts and be done with it. I'm stable at -158 volts and 800 boost ==159 volts seems to be ok too. If I notch it beyond -160 it fails.


Hi PHVM.

Yes, laptop up, etc.

PTM 7950 is working very well for almost all cores, but the last 4. I'd not think about changing it if not for these, temps are sometimes almost at 20C gap. I applied to the heatsink becuase the maker said to do this, this time I'll remodel to cover the processor.

Since they say we can remodel PTM 7950, I'll do this.

I don't care much about maximum performance, I just didn't want to ever need to repaste it again, so it is PTM 7950 for me in this machine.

Do not reuse the PTM 7950 unless you want to become an expert in disassembling your Alienware...

If you don't have it anymore, it's best to leave it as is.

Your laptop is already very close to its maximum potential performance using PTM 7950.

I've seen some that can maintain 165W indefinitely...more than that only with another better thermal compound.

If you have more PTM 7950 available, cut the correct size of the CPU and GPU and apply it completely covering both dies.

When replacing the heat sink, place it in line with the holes and press on the largest area possible with your fingers so that it sits correctly while you tighten the screws, one by one, half a turn at a time.

That's how I do it.

The settings in ThrottleStop are limits. If a CPU needs 180 Amps to run at full speed, whether you set the IccMax limits to 200, 300, 400 or 500 will make zero difference. There will be no EDP throttling caused by the IccMax power limit. I like to compare these limits to a limbo bar. If you set the bar 10 feet high or 100 feet high, does it make any difference? No it does not.

To avoid throttling, I always recommend setting any of the limits sky high. My only goal is maximum performance. If you think increasing any of these limits could damage your CPU then do not do it. Recent Intel CPUs have proven to be somewhat fragile so proceed with caution. If you choose to adjust IccMax, set the core and the P cache to the same value.

In my tests on Cinebench R23 with the i9-9980HK, which has an ICCMAX standard of 140A, it is only necessary to increase it in overclocking from 4.5GHz on all cores (4.2GHz is the maximum original multicore clock of this i9).

With ICCMAX at 140A it is possible to sustain 4.4GHz on all cores in Cinebench R23, but for the CPU to be able to maintain 4.5GHz or more it is necessary to increase it.

I believe this works for any Intel CPU and for those who are not going to overclock it is not necessary to change the ICCMAX.

 

[Alienware_Melting]

Do not reuse the PTM 7950 unless you want to become an expert in disassembling your Alienware...

If you don't have it anymore, it's best to leave it as is.

Your laptop is already very close to its maximum potential performance using PTM 7950.

I've seen some that can maintain 165W indefinitely...more than that only with another better thermal compound.

If you have more PTM 7950 available, cut the correct size of the CPU and GPU and apply it completely covering both dies.

When replacing the heat sink, place it in line with the holes and press on the largest area possible with your fingers so that it sits correctly while you tighten the screws, one by one, half a turn at a time.

That's how I do it.


In my tests on Cinebench R23 with the i9-9980HK, which has an ICCMAX standard of 140A, it is only necessary to increase it in overclocking from 4.5GHz on all cores (4.2GHz is the maximum original multicore clock of this i9).

With ICCMAX at 140A it is possible to sustain 4.4GHz on all cores in Cinebench R23, but for the CPU to be able to maintain 4.5GHz or more it is necessary to increase it.

I believe this works for any Intel CPU and for those who are not going to overclock it is not necessary to change the ICCMAX.

So I had some more PTM 7950 and decided to take the hard path. Opened it up and did like that. Same temp diffs but now it feels like it is gonna settle down better, because right now it is already lowering temps and reaching the same benchmarks as before after over a week of heat cycles.

I believe the problem was also that I've put too much PTM the other time.

I applied the new in smaller squares over the CPU and GPU surfaces this time, after freezing the PTM a bit to take the plastic out easily.

I did recover the old PTM and melted the shards in an aluminum foil and stored it. It does look like it can indeed be reused if you are patient enough to melt and apply it, just like regular thermal paste although harder.