I7 9750H Power/Current throttling

[Arrival]

Hey,

I’ve recently purchased a pcspecialist 16.1” Defiance V RTX, and overall I’m very pleased with the laptop. Temperatures are very good overall, with the 2080mq hardly breaking 70c whilst gaming. Likewise, the cpu generally hovers in the mid 70s under heavy load.

When using the XTU cpu stress test ( with the power profile in command centre set to performance), max core frequency eventually hovers around 3.3 GHz, with current and power throttling seemingly dictating this. Setting a core/cache voltage offset of -0.130v increases the max core frequency by ~250mhz.

With this in mind, and the decent cooling performance available, I’m wondering if using throttle-stop to limit the power/current throttling would be possible. I’ve avoided this so far as I’m unsure whether this may damage the laptop or the power adapter ( as per the throttlestop warning). Notably this laptop comes with a 180w power supply, which is slightly lower than a lot of similarly specced laptops.
Thanks, and I look forward to any advice you guys can give !

 

[Brask]

Hey,

I’ve recently purchased a pcspecialist 16.1” Defiance V RTX, and overall I’m very pleased with the laptop. Temperatures are very good overall, with the 2080mq hardly breaking 70c whilst gaming. Likewise, the cpu generally hovers in the mid 70s under heavy load.

When using the XTU cpu stress test ( with the power profile in command centre set to performance), max core frequency eventually hovers around 3.3 GHz, with current and power throttling seemingly dictating this. Setting a core/cache voltage offset of -0.130v increases the max core frequency by ~250mhz.

With this in mind, and the decent cooling performance available, I’m wondering if using throttle-stop to limit the power/current throttling would be possible. I’ve avoided this so far as I’m unsure whether this may damage the laptop or the power adapter ( as per the throttlestop warning). Notably this laptop comes with a 180w power supply, which is slightly lower than a lot of similarly specced laptops.
Thanks, and I look forward to any advice you guys can give !

Have you repasted the laptop? I have the 17.3" version and the temperatures are NOT good. Based on my little experience with this laptop (3 days) I have seen that when you undervolt the CPU you get lower temperatures but also it wont be able to be at 3.9GHz on all cores, when gaming the GPU will be able to boost higher but the CPU will be around 3-3.4GHz, what about your model?

 

[SpyderTracks]

Have you repasted the laptop? I have the 17.3" version and the temperatures are NOT good. Based on my little experience with this laptop (3 days) I have seen that when you undervolt the CPU you get lower temperatures but also it wont be able to be at 3.9GHz on all cores, when gaming the GPU will be able to boost higher but the CPU will be around 3-3.4GHz, what about your model?

If the cpu speeds are affected it’s because the undervolt is too severe. An undervolt, when applied properly will not affect cpu performance in any way.

 

[phitol]

Have you repasted the laptop? I have the 17.3" version and the temperatures are NOT good. Based on my little experience with this laptop (3 days) I have seen that when you undervolt the CPU you get lower temperatures but also it wont be able to be at 3.9GHz on all cores, when gaming the GPU will be able to boost higher but the CPU will be around 3-3.4GHz, what about your model?

As mentioned, A sensible undervolt will increase performance, especially for sustained stress testing.

The problem these days is the Intel power management being confusing. (Anyone interested in this, read articles like this that explain the differeing power levels and how they work: https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo)

In laptops, the PL2 (short term turbo) is 78W (Performance mode, 9750h Definance V) for 28 seconds. 78W is enough for the CPU to almost immediately hit thermal throttling, but will allow the processor to get close to it's all core turbo (4.0Ghz), maybe dropping marginally to 3.9Ghz due to the throttling.
After that, you hit PL1, TDP drops to 55W (again for the 9750h Defiance V in performance mode) and the all core turbo will drop to a much lower value. The actual value largely depends on the how stressed the stress test is. For example, the XTU stress test is quite mild and you can get 3.3Ghz quite easily sustained, rising to 3.5Ghz+ if undervolted). If you use AIDA64, expect .2 to .3 GHz less.


Then you have mixed loads, a good example is gaming. Whilst gaming, each core is not 100% max'd out like a stress test, so you find the average all core rates are actually higher still as each core mixes heavy/light workloads and core speeds look nice and high. This is also backed up by the fact people note when background processess periodically kick in, the CPU clocks can hit 4.6Ghz and you get short periods when the CPU goes back to PL2 short term turbo where the TDP increases above 55W and thermal throttling is allowed to occur.

And this is the issue I feel with the modern Intel CPU's.. the cooling solutions clearly can sustain PL1 @ 55W for long periods with people seeing 70-80 degrees in sustained stress testing. the problem comes that for heavy but mixed loads the CPU can jump back to PL2 if it's not been that stretched for a while, so you get these short excursions where cores ramped right up and thermal limits are reached for a few seconds.

This is why its important when analysing your system to be careful about which stress test you use, and to ensure you are monitoring CPU TDP Package power use to know if you are in PL2 / PL1 limits (78/55W) then you can see how good your cooling solution is. I'd say that in reality you want to concentrate on PL1 use, that is easy to sustain for long periods and then use a good benchmark or at least the same one others are using as the exact load of the CPU will determine the clock speed for it's current TDP limit. Even switching AVX on/off in AIDA64 has a big effect on temperatures and sustained clock speeds.

 

[Brask]

If the cpu speeds are affected it’s because the undervolt is too severe. An undervolt, when applied properly will not affect cpu performance in any way.

-.125 on core and cache I dont see that will be very big... Problem is when using gpu and cpu at the same time, the CPU wont use more than 20W

 

[Brask]

As mentioned, A sensible undervolt will increase performance, especially for sustained stress testing.

The problem these days is the Intel power management being confusing. (Anyone interested in this, read articles like this that explain the differeing power levels and how they work: https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo)

In laptops, the PL2 (short term turbo) is 78W (Performance mode, 9750h Definance V) for 28 seconds. 78W is enough for the CPU to almost immediately hit thermal throttling, but will allow the processor to get close to it's all core turbo (4.0Ghz), maybe dropping marginally to 3.9Ghz due to the throttling.
After that, you hit PL1, TDP drops to 55W (again for the 9750h Defiance V in performance mode) and the all core turbo will drop to a much lower value. The actual value largely depends on the how stressed the stress test is. For example, the XTU stress test is quite mild and you can get 3.3Ghz quite easily sustained, rising to 3.5Ghz+ if undervolted). If you use AIDA64, expect .2 to .3 GHz less.


Then you have mixed loads, a good example is gaming. Whilst gaming, each core is not 100% max'd out like a stress test, so you find the average all core rates are actually higher still as each core mixes heavy/light workloads and core speeds look nice and high. This is also backed up by the fact people note when background processess periodically kick in, the CPU clocks can hit 4.6Ghz and you get short periods when the CPU goes back to PL2 short term turbo where the TDP increases above 55W and thermal throttling is allowed to occur.

And this is the issue I feel with the modern Intel CPU's.. the cooling solutions clearly can sustain PL1 @ 55W for long periods with people seeing 70-80 degrees in sustained stress testing. the problem comes that for heavy but mixed loads the CPU can jump back to PL2 if it's not been that stretched for a while, so you get these short excursions where cores ramped right up and thermal limits are reached for a few seconds.

This is why its important when analysing your system to be careful about which stress test you use, and to ensure you are monitoring CPU TDP Package power use to know if you are in PL2 / PL1 limits (78/55W) then you can see how good your cooling solution is. I'd say that in reality you want to concentrate on PL1 use, that is easy to sustain for long periods and then use a good benchmark or at least the same one others are using as the exact load of the CPU will determine the clock speed for it's current TDP limit. Even switching AVX on/off in AIDA64 has a big effect on temperatures and sustained clock speeds.

Why under load (gaming BFV) the CPU doesn´t pass 20W ? My undervolt with throttlestop is not big, -.125

 

[phitol]

Why under load (gaming BFV) the CPU doesn´t pass 20W ? My undervolt with throttlestop is not big, -.125

When gaming, the cpu isn’t under much load. I’ve just been playing vr games and the average load is 22W, however the clock averages for all cores is 4ghz, this tells me that the vr games are not using all available hardware resources in the cpu and it only needs 22w to maintain the all core turbo speeds (which is another limit btw). Remember 4ghz is the CPU’s set clock speed for all core Mac. You could increase this in throttlestop, but whilst it might improve gaming, it’ll kill thermals if stress testing.

In this case if you removed the undervolt your clocks speeds would still be 4ghz but the TDP will be higher and more TDP = higher temps

125mv of UV is reasonable for stability, mine will run stress tests at much higher values, but whilst gaming, occasionally it crashes if I go too high, so I stick to 125 for stability in all situations.