Trying to increase the power limit of my i5-8265U.

[dominat0r]

Hello everyone,

since almost four years, I'm a quite happy owner of a HP Pavilion 15 laptop with the i5-8265U CPU and a 1050 Ti Max-Q GPU. I chose this device as a compromise between a thin and light device and the opportunity of light gaming, without the need to play newer AAA titles.

Early on I moderately undervolted the CPU to have it run cooler and more efficient. I since used several ThrottleStop profiles for max performance, balanced and power-saving scenarios, essentially by using different Speed Shift values of 0, 60, 128 and 196 as well as changing between Windows power plans.

Recently I was curious to see how ThrottleStop developed since then, installed the newest version and read some threads on the forum. Especially this one was interesting to me, as this ULV i7 should be quite similar to my i5, and it's running inside of a thin and light device as well. So I tried to replicate some values and adjust them for my use case. So here's my problem: Whatever values I set for the PL1 and PL2 power limits, they seem to be "overwritten". I tried to moderately increase them to 20W and 25W respectively, as my CPU runs quite cool even under load, never reaching temperatures over 85°C. It seems to work for some time in test benchmarks, but after maybe 40 to 60 seconds, a PL1 of 15W kicks in and throttles the CPU, which then hangs around at 55 to 60°C, with much lower frequencies. Switching between my max performance and balanced profiles seems to make no difference.

So I would like to find out with your help if this indeed is a case of the manufacturer's embedded controller taking over (thanks HP!), essentially rendering my ThrottleStop attempts hopeless, or if there maybe is some hope for me within the options of ThrottleStop that I may haven't found yet. I attached some screens of my current settings as a starting point.

Thanks in advance and best regards
Dominik

 

[unclewebb]

embedded controller taking over

That is the problem. The CPU constantly compares the MSR, MMIO and EC requested power limits. The power limits that are set to the lowest values will win control of the CPU. If the EC is programmed to throttle your CPU at 15W long term then it will throttle to 15W. There is nothing ThrottleStop can do about that.

EC induced throttling is typical of many HP laptops. Here is an example of a Lenovo C930 with a similar low power 8th Gen U series CPU. This CPU can run fully loaded at full speed at more than twice the 15W rated TDP. Thank you Lenovo for not using an EC to force feed low ball power limits to the CPU.

 

[dominat0r]

That is the problem. The CPU constantly compares the MSR, MMIO and EC requested power limits. The power limits that are set to the lowest values will win control of the CPU. If the EC is programmed to throttle your CPU at 15W long term then it will throttle to 15W. There is nothing ThrottleStop can do about that.

EC induced throttling is typical of many HP laptops. Here is an example of a Lenovo C930 with a similar low power 8th Gen U series CPU. This CPU can run fully loaded at full speed at more than twice the 15W rated TDP. Thank you Lenovo for not using an EC to force feed low ball power limits to the CPU.

That's a sad outcome, but thank you for confirming what I expected! Also, I want to take the chance to thank you for giving us ThrottleStop, it's such a great tool. At least I was able to undervolt the device, even though HP won't allow me to bring it to its limits. Maybe I could at least try to undervolt it even more, so that I get the most out of the 15W I have in the long run. But I shouldn't expect much more than the -125mv I applied so far, right?

 

[unclewebb]

-125mv

For most 8th Gen U series CPUs, -125 mV is already over the limit and might not be 100% stable. Trying to go further will probably get you a crash and a blue screen.

thank you

You are welcome. ThrottleStop has become an extremely useful tool for a lot of different throttling issues. Sad that Intel has already started completely blocking software CPU voltage control on most of their 12th and 13th Gen mobile CPUs. Even some of the unlocked K series CPUs might have undervolting locked out after a microcode update. The end of an era.

 

[nucbeard]

This is my first post so forgive me if I am a little unpolished.

I have been playing with TS on many different devices for the last few months and the latest one I have messed with is an HP ProBook 440 G5 with an i5-8250U which is very similar on paper to dominat0r's i5-8265U. I think the processors are the same physical part but programmed differently: https://www.intel.com/content/www/us/en/products/compare.html?productIds=149088,124967

My experience with the 8250U is completely positive and wholly different from the OP. I bumped up PL1 and PL2 both to 48W on the TPL screen and cranked TAU (TTL) to infinite. For benchmarking I want to run WOT (wide open throttle) without any turbo time limits kicking in. I chose 48W since I have two Intel Nuc 8's (i5-8259U and i7-8559U) that can sustain 48W without bouncing off of the VR Current limit. (Interestingly, the i5 Nuc can go a watt higher to 49W before it hits any hard VR Current limits.) I use y-cruncher 1B as a dyno. I prefer it over CB20 since I can see what the wattages and temps are while it is running in Realtime. In CB20 I cannot do that. TS will not update when running CB20 in Realtime. I can run CB20 in High Priority and not Realtime but that is not the full load the computer can take. y-cruncher is better for me. It is the crucible.

I just went downstairs and ran the 8250U at WOT in y-cruncher and she pulled a max of 42.1W in TS. I have a power meter for RC aircraft wired into my main 20V power feed going to the laptop and it hit 65.5W on an aftermarket 65W adapter. The *adapter* is keeping the 8250 from going higher. My Nucs run a 90W adapter so no bottleneck there. ANYways, the moral of this story is that I think dominat0r's 8265 hasn't been woken up. Maybe unclamping PL1 and PL2 and move them up to 42W. Crank the TTL to infinite though it shouldn't matter if both PL1 and PL2 are the same.

Also, I would install the latest Throttlestop 9.5. Next, turn off Speedstep and turn on Speedshift and set it to "0". Turn on BD Prochot. Your laptop is gonna get hot if this works. To counter this, turn it over, remove the bottom covers and get a fan blowing on the heat sink if you can see it. Run an external monitor, keyboard and mouse.

 

[unclewebb]

@nucbeard

Your 8th Gen U series success story is the same as the 8th Gen Lenovo laptop that I have. Wonderful performance without any power limits being enforced.

Unfortunately, not all laptops have this ability. On his laptop, HP is using an embedded controller (EC) to enforce a 15W long term power limit. The MSR and MMIO power limits that ThrottleStop gives you access to are ignored when this happens. Long term the CPU will throttle right at 15W.

Some HP laptops might have become locked down like this after a BIOS update. Some were locked down from day 1. If checking the MMIO Lock box in ThrottleStop and setting the MSR power limits higher in ThrottleStop does not work, it is usually an EC power limit that is being enforced that is the problem.

 

[dominat0r]

This is my first post so forgive me if I am a little unpolished.

I have been playing with TS on many different devices for the last few months and the latest one I have messed with is an HP ProBook 440 G5 with an i5-8250U which is very similar on paper to dominat0r's i5-8265U. I think the processors are the same physical part but programmed differently: https://www.intel.com/content/www/us/en/products/compare.html?productIds=149088,124967

My experience with the 8250U is completely positive and wholly different from the OP. I bumped up PL1 and PL2 both to 48W on the TPL screen and cranked TAU (TTL) to infinite. For benchmarking I want to run WOT (wide open throttle) without any turbo time limits kicking in. I chose 48W since I have two Intel Nuc 8's (i5-8259U and i7-8559U) that can sustain 48W without bouncing off of the VR Current limit. (Interestingly, the i5 Nuc can go a watt higher to 49W before it hits any hard VR Current limits.) I use y-cruncher 1B as a dyno. I prefer it over CB20 since I can see what the wattages and temps are while it is running in Realtime. In CB20 I cannot do that. TS will not update when running CB20 in Realtime. I can run CB20 in High Priority and not Realtime but that is not the full load the computer can take. y-cruncher is better for me. It is the crucible.

I just went downstairs and ran the 8250U at WOT in y-cruncher and she pulled a max of 42.1W in TS. I have a power meter for RC aircraft wired into my main 20V power feed going to the laptop and it hit 65.5W on an aftermarket 65W adapter. The *adapter* is keeping the 8250 from going higher. My Nucs run a 90W adapter so no bottleneck there. ANYways, the moral of this story is that I think dominat0r's 8265 hasn't been woken up. Maybe unclamping PL1 and PL2 and move them up to 42W. Crank the TTL to infinite though it shouldn't matter if both PL1 and PL2 are the same.

Also, I would install the latest Throttlestop 9.5. Next, turn off Speedstep and turn on Speedshift and set it to "0". Turn on BD Prochot. Your laptop is gonna get hot if this works. To counter this, turn it over, remove the bottom covers and get a fan blowing on the heat sink if you can see it. Run an external monitor, keyboard and mouse.

Hey, thanks a lot for your input! Nice to see that indeed, you can be successful with Intel ULV CPUs of the 8th generation. And you're right, the 8250U is almost the same as my 8265U. But as unclewebb already commented, in my case there seems to be a fixed limit at 15W that is enforced by the embedded controller after some time of turbo power. I tried out the values you mentioned, with the same results, sadly. I also have the feeling that my dedicated GPU may play a role here, as the device seems to limit the package power (CPU + GPU) to 25W in the long run. It looks like I will have to use different profiles for CPU-heavy and GPU-heavy use cases, either limiting the CPU frequencies to get more GPU performance or having the CPU run at "full" power (which means max turbo frequencies and power + the aforementioned 15W for long-term loads).

 

[nucbeard]

Ahh, I am starting to understand. This EC controller business would really tick me off. I guess I got lucky in picking the laptop I did. I guess picking one with only integrated graphics was key.

I wonder if there is any tick box in the 8265's spartan HP bios that would turn off the off-chip external gpu chip and thus allow higher cpu limits. I would also like to know what a Kill a Watt shows when running this machine full tilt. Wonder how close to the maximum for the adapter it is. I cannot imagine that it is close but it would be interesting to know. BTW, I have seen a 10W overhead for all consumers not including the CPU wattage so if the cpu is pulling 45W then the whole machine is probably around 55W. That's with this ProBook 440 G5 laptop though. Like I mentioned, it has onboard graphics only.

I am now not a proponent of updating bioses just because. Not being able to use TS would be a total drag. I sold a Nuc11i7PAH because of that plus the lack of S3 sleep. Modern Standby is ridiculous.

Thanks for the kind explanation from both of you. I know more now.

One other interesting thing I found this morning is that while looking at my inline dc power meter I see that y-cruncher pulled 65.5W max and Cinebench R20 multi core only used ~45._W running Realtime priority in both. I need to get a higher capacity adapter. I would just give it 20V from a 24V battery and dc-dc converter but there's more to the later adapters with a pilot voltage. I haven't figured out to hack supplying voltage to the computer without that pilot voltage.