Start a Conversation

Unsolved

U

8 Posts

4771

November 16th, 2018 12:00

Dell 5491 8850H throttling, DIY troubleshooting, conclusion

As you (DELL) keeps ignoring obvious issues with throttling on your laptops (hello hexacore lattitude 5491) I thought I write down a little guide for some tech-savvy people.


Disclaimer:
I don't take any responsibility for the damage caused by any of those modifications. You are doing it on your own. If you fail, don't blame me for your poor engineering/manual skills, though they are probably better than any of those Dell's designers...

Specs:
Dell 5491, 8850H, 32GB ram, no discreete GPU

I ran multiple scenarios and these are the results:
Every command is in Linux, so sorry Windows guys! I'm pretty sure there are some tools for this.

Every scenario involved running the prime95 console benchmark for about 30 minutes. I doubt this will be done in real life, but at least to some extent it shows what happens during compiling large code on all 12 cores. All times are approximate, may be different for your laptop. Also I do not have a discreete GPU and if you have one - you entered a big poo and any below suggestions may not be valid for your rig.

To observe frequency on all cores:
watch -n1 "grep -i Mhz /proc/cpuinfo"
To observe temperature:
watch -n1 sensors


1) no modification done, no cooler pad
Basically temperature rises in a few seconds to nearly 100deg, then the fan goes full speed. Every time it hits the limit, clock goes down do 800Mhz after 2 seconds it goes to 3GHz, then after 2 seconds it goes to 800 and so on - thermal throttling, obvious stuff

2) thermal grizzly paste applied, zalman coolerpad with 195mm slow rotating single fan
A bit better, temperature does not cross 90 degrees. Turbo starts with 3.4Ghz on all 12 logical cores then it drops quickly to 3,1 then to 3,0 and it sits there for a bit over a minute. Then TDP throttlings engages and for every 10 seconds with 3.0G clock it goes to 800 for a moment then goes up to 3.0G in like ~8s, then the process repeats. After another 2-3 minutes max freq is ~2,7G only and it goes the same as previously after 10s sitting on max.

3) Same as 2) + Turbo and Speedshift disabled in UEFI
Disabling Speedshift means losing 800Mhz frequency, min clock is 1,3G then - please keep that in mind.
The benchmark runs fine with temperatures about 85 deg with clock sitting spot on 2.6Ghz (this is a nominal max freq for 8850H cpu) for like 5-6 minutes. Then it goes down do 800 every 10s and then goes up back to 2,6G in bigger increments in frequency for about 7-8s. CPU cannot hold it specs through the time - unbelievable! Even when I undervolted CPU with some handy python script by 100mV, this is exactly the same, perhaps the symptoms occur slower.

4) Same as 2) + added two smaller heatpipes alongside over the main one. A bits of black paint were scrapped on both ends and thermal glue was applied for better thermal conductivity. Also two smaller blobs of glue were added to secure modification (it withstands 350 deg Celsius). Heatsink area that sits on top of CPU was polished using 600, 2000 and 3000 sand paper. This is how it looks now

Two additional heatpipes made a difference that when the starting point is like 40 deg, it gets more time to heat the whole heatpipe and thus for the throttling to engage. After 3 minutes on 3,1G, Turbo sits on 3.1G for like 15 seconds then it slows down to 800M then it goes back in like 5s, so there is some difference, but not a big one. Temperature does not cross 90 deg, but the trottling still occurs. Unfortunately after few more minutes (like 3-4min) max clock is set to 2,7G and the story is  the same, only recovery to max frequency is quicker.

5) Same as 4) + lowered down max_per_pct
echo 50 > /sys/devices/system/cpu/intel_pstate/max_perf_pct
It limits max cpu clock to 2,2Ghz without the need of disabling anything in UEFI. Obviously you can also disable turbo clock completely by echo 1 > /sys/devices/system/cpu/intel_pstate/no_turbo. But without limiting max_perf_pct clock will stay on 2,6Ghz though. And I skipped the variant without a coolerpad as it obviosly rise the temperature by ~10 deg up.

But let's look on bright side, temperature is 75deg with a fan spinning 5kRPM  (MAX) and a cooler pad running 2,2GHz on 12 cores without turbo. Imagine that!!! 8750H without a turbo boost sold for the price of 8850H!!! This is how the Dell new business plan looks like now. Give me your money dear customer and go to **bleep**. And our engineers will work on a BIOS issue, this must be a BIOS issue, nooo that is impossible that we **bleep** up our cooling system with a paperthin heatpipe and 1$ cooler. Who would have thought that the power emitted from a cpu must be dissipated and needs some thermal mass to push it outside.... geeez! basic science people. The person who allowed this to happen should be fired immediately. I highly doubt that some reasonable individual from Dell will start some recall service to replace those #!@$%^ cooling systems, because the time they were doing a good products apparently is over. First of all they should have somebody designed it properly.... this is the last time I bought a laptop from Dell. This is a high spec model. I payed almost 2k$ for it. And if some support specialist says he/she is sorry, I will go there and make him/her eat the processor.


And the second hilarious problem. PSU included with my laptop is 90W and I thought I'll buy WD15 dock with 130W - should be enough, no discreete GPU, less worries. And it reports during POST that the PSU is too weak!!! Palmface!!! I've ordered 180W, perhaps this one will work. I have to say that the docking station is very power efficient - it would appear it gets at least 40W - nice!

Thank you very muchheatpipe.jpg










































 

8 Posts

November 17th, 2018 10:00

I'm starting to think that this may be a voltage regulator issue (given amount of people complaining on their latitude and XPS series). I did some more tests with my modified heatpipe and:

- turbo sets at ~3.1Ghz and lasts for about 4 minutes

- after 4 minutes it goes up and down from 3.0Ghz (for 12-15s) to 800Mhz (going up to 3.0 in 6-8s) and it lasts for about 2-3 minutes

- then max freq sets on 2.7Ghz and it goes down to 800 in a similar way as above for like 2-4 minutes

- suddenly clock jumps to 3.0Ghz going up and down as usual for 2-4 minutes, then it goes down to 2.7 and then everything repeats Probably due no radiator as in desktop motherboards VR array get overheated. I wonder if it will be similar to famous IBM's VR issues in most pSeries servers....

- interestingly at idle with performance governor enabled, it sets on 4.3GHz with temp <50deg and 0,1% load :)

10 Posts

February 1st, 2019 07:00

Disappointing with Dell Product , I'm using Dell G7 and more way throttling :) . 
New BiOS make my machine throttling too much and newest Bios too much throttling

April 10th, 2020 00:00

@UglyJoe  What the heatpipes did you use? I mean donor. Where I can found the same?

1 Message

February 21st, 2023 01:00

I have the same system and CPU and the following solved the thermal issues for my on my side. The problem I had:

  • CPU load was about 10% and temperature was around 90+ degrees Celcius.
  • I have used the Dell Power Management to see if the behavior is different when using a different Thermal profile. When using Cool mode I saw the temperature drop, but the performance of the system was terrible, so I've set it back to Optimized. Doing this resulted my temperature to go high again up to 95 degrees when the system was not doing anything.
  • I have upgraded BIOS without any result.
  • I replaced the cooling FAN without any result.
  • A Dell engineer replaced my motherboard incl. CPU and still without any result.
  • Results: Then I went to BIOS, under Perfromance section and disabled the Intel Turbo Boost option. After restarting the system the temperature dropped to about 40 degrees Celcius with about 10% CPU load. As soon as I went doing stuff on the system, CPU went up to 50-60 % and temperature went up to 60-70 degrees, which was ok. Seems that disabling this option in BIOS helped in my case.
  • Conclusion: I think it might have to do something with the power management with the CPU and voltage regulation that increases the temperature a lot. Might be something that can be fixed in a BIOS update. Hopefully my solution helps others here as a workaround for now.

 

No Events found!

Top