X1E X1 Extreme throttles at 80C with BIOS Virtualization On

[JoeCanada]

My X1E is throttling at 80C when Virtualization is turned on in the BIOS.  It throttles at 100C when it is turned off (assuming Intelligent Cooling is off and power/battery settings are at highest performance.  This drastically affects performance negatively (using Benchmark in XTU and Cinebench CPU drops can be 20 to 30%).  Other people have been having 80C throttle issues and generally they are solved by turning off Intelligent Cooling and setting power/battery settings to highest performance, but this is a different issue that most people won't see unless they have Virtualization on in the BIOS .

Any way to throttle at 100c with Virtualization ON in the BIOS?  I and other people need this for Docker for Windows and we need the performance!  Does this have anything to do with Meltdown and Speculative Execution?   Can it be overridden?

[mnylen]

I've resigned trying to solve this issue. There's absolutely nothing that helps besides turning off virtualization, which is not an option for me.

I was able to read the MSR_TEMPERATURE_TARGET in Windows and it seems to be set to 100 C, so there's something else that causes the thermal throttling to kick in at 80 C. Nobody seems to know what that something is.

I reported this issue using the managed technical support access and got offered a replacement system board, which I refused, as I had the exact same problem before my system board was replaced for separate issue. I'm at a point where I'm trying to seek either a refund or a swap to P52, which shouldn't have such huge thermal problems, but I'm skeptical if I'll get either.

[JoeCanada]

TL;DR;  Repaste could make a big difference whether or not you're throttling at 80C or 100C.  Results below.

Disclaimer:  Don't try this at home unless you're comfortable with computer assembly/disassembly and able to delicately handle very thin fan wires with very dainty connectors.

Excellent repaste results with Arctic MX-4 (with an 8850H):
- Instructions here https://imgur.com/a/Blvpjd0

- WAY too much paste was on there from factory, far beyond the edges of the chips and piled up on the chips and the heatpipes and onto the plastic film.  Ridiculous.  Maybe they had too many units that were short on paste and they started globbing it on there at the factory, I don't know.  If it's not automated metering/dispensing at the factory (I'm a Computer Engineer and I've been to an electronics factory in China twice, and not everything is automated) then there's a definitely possibility of a variation between factory workers.  If there is too little or too much paste either situation will hurt performance and spike your temps.
- Spread a very thin film of MX-4 on both the CPU and GPU first using a plastic spreader stick.  Then placed a drop of MX-4 in the middle of each chip and screwed down according to the linked instructions.  Then waited a couple minutes, undid it all to check there was the right amount (complete coverage to the edge of the chip with very little extra outside of that).  Then I did it again from scratch (completely cleaned it) and used the same technique using the correct amount.

- The following results are with XTU settings of:
  - Core Voltage Offset: -0.125V

  - Cache Voltage Offset: -0.125V
  - Graphics Voltage Offset: -0.040V
  - Multipliers per Active Cores maxed out at 47x/46x/46x/45x/45x/44x
  - Turbo Boost: 45W (this was the default)
  - Turbo Boost Short: 75W for 28 seconds (this was the default)

Also BIOS was updated to 1.17, and mesh was removed from vents.

w/ Virtualization in BIOS Off:
BEFORE -> AFTER

Max temp: 100C -> 96C
XTU: 1638 -> 1684
Cinebench R15 CPU: 1003 -> 1222
Cinebench OpenGL: 127.21 fps -> 133.95 fps
(Note: After repaste it is now Power Throttling instead of Thermal Throttling.  Great!  That is a good thing.)

w/Virtualization in BIOS On:
BEFORE -> AFTER
Max temp: 80C -> 80C
XTU: 1458 -> 1526
Cinebench R15 CPU: 841 -> 964
Cinebench OpenGL: 113.85 fps -> 123.69 fps
(Note: Still Thermal Throttling at 80C of course but MUCH better benchmarks!)

I believe the results speak for themselves, the repaste was worth it in this situation.  Also the idle temp is as low as 49C now.

As a side note, after repaste I think I'm okay with the system throttling at 80C now with Virtualization ON in the BIOS... the temps are lower which is probably better for the processor, the fan comes on less, and the performance is now pretty good.  As another side note, this 8850H processor is really meant for a bigger case and better cooling than the X1E to squeeze the most performance out of it, but hey, I'm stuck with it now and I'm finally okay with everything (after going through the BIOS 1.15 Discrete Graphics brick affair) knowing that I've done just about as much as I can do (apart from getting a laptop cooling pad which is arriving tomorrow which might help even more).

Honestly this system is engineered pretty well for a powerful processor given the space constraints, I just wish more care was taken with the thermal paste at the factory so there wouldn't be so many people having "Your Mileage May Vary" issues on these forums.

[JoeCanada]

UPDATE!

TL;DR;  Performance is even better with Discrete Graphics in BIOS (DO NOT, I repeat DO NOT! do this unless you have 1.17 BIOS already installed and working with everything, unless you want a brick).

Well lo and behold, I DARED (I can't believe I even dared to try this, but I had everything backed up and a spare laptop handy just in case) to switch my graphics from Hybrid to Discrete in the BIOS (and of course after going through the fiasco with prevous changing from Hybrid to Discrete Graphics bricked unit and 6 weeks of waiting for it to be returned to me from Lenovo).

Well suffice it to say it booted and ran fine.  I changed from Hybrid to Discrete graphics with nothing plugged in but the power cable just to be safe.  It booted and then I plugged everything else in, two big monitors (USB-C and HDMI), a bunch of USB stuff, all worked fine and I re-ran all my tests from last post, and with Virtualization still ON in the BIOS (which still throttles at 80C) I get the following:

Changing to Discrete Graphics:
BEFORE -> AFTER
Max temp: 80C -> 82C
XTU: 1526 -> 1579
Cinebench R15 CPU: 964 -> 1112
Cinebench OpenGL: 123.69 fps -> 129.56 fps

Now we're talking turkey!  Now I feel even better about this machine than I did last post, which was less than 3 hours ago!

So with a repaste and Discrete Graphics ON it's possible to get a lot out of this machine, even if you need Virtualization and therefore throttling at 80C.

[mnylen]

Interesting. I guess when the graphical desktop environment runs exclusively on the dGPU, the CPU has more capacity to perform  better in other tasks. But I would imagine the overall temperatures might be worse as now the dGPU is heating more too and it's connected to the same heatsink as the CPU.

And of course, the battery life is likely worse too - it was bad already at around ~3-4 hours on 4K display, but with discrete graphics enabled I've seen people report only 1-2 hours. Not to mention on battery you seem to get only half of the performance anyway (I scored around 500 in Cinebench R15 on battery power and with virtualization on).

[JoeCanada]

A lot of the heat from the dGPU would go to the 2nd fan based on the physical design with 2 heat pipes and the 2nd fan being closer to the dGPU. It doesn't seem to affect the CPU temp too much if at all, so it likely just ends up in increased power usage which gets sent out as heat through the 2nd fan. In the BIOS I set my "on battery" settings to max performance, and then I can control it from Windows from there using Cool & Quiet or Balanced or whatever I want. Haven't tested it though so I can't speak to battery as I use my machine on power 99% of the time.

I also can't speak to the dGPU too much because it doesn't seem to factor into anything in XTU since I went to Discrete. I can't even control Graphics Voltage Offset anymore, so I guess that setting in XTU must have been for the iGPU.

[freakwave10]

I opened a service request regarding this issue and from what I can see in the ticket they are shipping some hardware, but the ticket does not mention anything that is in a human readable form.

1/23 1344 I have exactly the same problem as described here in the forum:
https://forums.lenovo.com/t5/ThinkPad-X-Series-Laptops/X1E-X1-Extreme-throttles-at-80C-with-BIOS-Vir...
Due to this issue the machine can not perform as expected and is throttled. * *SEE TIB* *
01/28/2019 05:25pm: Order#407NHRG-01,P#01AY981, WB#: 1Z17704F0470061959
01/28/2019 01:10pm: ETA: 01/29/19 07:00:00
01/28/2019 01:10pm: Order#407NHRG-01,P#01YU923
01/23/2019 01:45pm: FSR Applied

Wolfgang

[mnylen]

Looks to be "thermal fan". So not likely to help You can use the parts lookup here https://support.lenovo.com/fi/en/partslookup

[freakwave10]

Oh well, this is really usefull ;-).

I tried it again and it is exactly as described here.

The CPU-Z benchmark when virtualization in BIOS is enabled results in about 20% performance loss when the system is throttled at 80 degrees celcius. This makes the machine quite redundent since you could buy a machine with 4 cores that is cheaper and does not have the issue.

@Lenovo, I updated the ticket and please escalate this to the BIOS developer. All my details are in the ticket. I can provide any details for reproducing the issue. Request number: 407NHRG

Wolfgang

[JoeCanada]

Agreed, it was a pretty big price upgrade to the 8850H and it's a waste of money if the system is thermally-limited and not able to take advantage of the processor.

I don't see how giving you a new fan assembly is going to change anything, my fans are working fine and it still throttles at 80C with virtualization ON in the BIOS.

However, the paste job was horrible and as i said before, i got much better benchmarks after proper repaste.