Lenovo P43s throttling like crazy

[time]

CPU is Intel i7-8565U (4-core 8-thread 1.8-4.6GHz 8MB cache), which seems to be one of Intel's 'fake' 15W CPUs. That is, they *claim* 15W TDP but in practise these things easily pull 30W (according to CPU ID HWMonitor) - and that's with massive thermal throttling.

The icing on the cake is that apparently the heatsink is shared between the CPU and the Nvidia P520 baby Quadro (really a MX150). I should point out that this is supposed to be a "mobile workstation" but only weighs 3.2 lbs.

When your CPU cores nudge 90 degrees C at boot, you know you have a problem. In use, any activity at all can instantly send the temperature of one of the cores into the high 80's C. This is with an ambient temperature of 24 C (75 F).

So yes, the CPU can reach 4.6GHz on a single core, but only for a few seconds. 3.6GHz is more realistic, but even that collapses when the heat rises. Multiple cores are limited to about 2.3-2.6GHz after the first few seconds.

In a single-threaded app with everything turned on in an editable orthographic view, this laptop was about 4 times slower than an Asus ROG i9 topping out at about 4.8GHz. Not a function of core count, RAM or GPU - just unbelievable throttling.

Running the 7-zip benchmark with 4 threads, it quickly sinks to about 2/3 of my elderly i5-4690 (4-core 3.5-3.9GHz) desktop - CPUID says about 2.3GHz, so that makes sense.

Beware, although I certainly wouldn't be confident that switching brands would change things for the better with this fraudulent CPU.

 

[Stereodude]

Are you sure it's working as expected? That sound really bad even for the sort of shoddy corner cutting found in laptops.

I'd be tempted to pull the heatsink and check for good contact between it and the CPU.

 

[Handruin]

I'd also be tempted to pull the back off and reseat the heatsink with some new thermal paste if you're able to. A 90C temp at boot does not seem right to me or at any time for that matter.

Have you looked into monitoring with hwmonitor? They have a free version you can play with, that's usually my go-to on new systems when monitoring temps and power pulls. I combine that with prime95 for stress testing to see where the thermals end up going over time.

 

[time]

Have you looked into monitoring with hwmonitor?

"these things easily pull 30W (according to CPU ID HWMonitor)"

We have an identical one coming back at the end of the month, so I will compare with it.

In the meantime, I need to order another one, but for some reason I have some doubts ...

 

[Chewy509]

Sounds like a crap thermal design for the laptop... Unfortunately, Intel define Base Clock and TDP as:

Processor Base Frequency describes the rate at which the processor's transistors open and close. The processor base frequency is the operating point where TDP is defined.

Which means as long, as all fours cores at 100% CPU utilisation at Base Clock (1.8GHz for this CPU) emit no more than 15W, AND the CPU doesn't fall below base clock under 100% load, then all is good in the world. And it sounds like Lenovo designed their thermal solution based on this, rather than over engineering for some head room (eg handle a 30W+ TDP when turbo boost is active on all four cores), which I expect you assumed when looking at a workstation class laptop, (and to be honest, I would expect better from a workstation class laptop as well).

* Source: https://ark.intel.com/content/www/u...-8565u-processor-8m-cache-up-to-4-60-ghz.html and click on the ? for the Base Processor Frequency.

 

[time]

What's even worse is the Configurable TDP-up on that same spec page. That's an alternative TDP of 25W at a frequency of just 2.0GHz! Intel must be doing some serious undervolting (-18%) at the base clock of 1.8GHz to get it down to 15W.

Configurable TDP-up is better with the 10th generation i7-10510U, they claim it gets to 2.3GHz while averaging 25W.

This graph, courtesy TechSpot, shows what can happen. Based on this, I'd say Lenovo has gone with a 15W thermal envelope (as has probably every manufacturer making 3.2 pound laptops).

 

[time]

I think I understand some of this new world now.

Whether a CPU is an i5 or an i7 is just a function of binning, eg. an i5 requires more voltage to be stable at the same frequency, so Intel reduces the i5 clocks at each step to compensate.

Eg. i7-8565U ranges from 1.8GHz @15W to 4.6GHz max, with TDP-up of 2.0GHz @25W
whereas i5-8265U ranges from 1.6GHz @15W to 3.9GHz max, with TDP-up of 1.8GHz @25W.

And I'll bet the cache reduction from 8 to 6MB is to cover bad cells in the L3 cache, another binning choice.

The CPU itself controls its limits with Speed Shift, so the laptop manufacturer just has to configure it correctly and supply an approved cooling solution to handle an average of 15W, or 25W if they want to add weight and noise and shorten battery runtime. In the case of the P43s, Lenovo has restricted it to as little as 12W, possibly because the heatsink is shared with the Nvidia Quadro graphics chip. The larger P53s must have a larger cooling solution, because it appears to be less restricted.

A CPU like the i7-9750H, on the other hand, will run at much higher average speeds. It ranges from 2.6GHz @45W to 4.5GHz max and will comprehensively out-perform an i7-8565U under heavy loads. According to NotebookCheck, that can easily be by a factor of three with Cinebench R15 looping, eg. scores of 1200 vs 400.

So don't look at the i5 or i7 prefix of a CPU model - it's just marketing bull. Look at the suffix if you want to see what matters.

 

[time]

I had some success in avoiding the worst of the throttling by undervolting the CPU. Intel XTU no longer works with many CPUs, so I used ThrottleStop instead. Undervolting by 0.1V was enough to re-align the steps. You can also fiddle with the clock steps, although Speed Shift uses that more as just a guide.

Previously, the P43s scored just over 400 in the Cinebench R15 multi-core test, and about 24fps in the OpenGL test.

With the CPU undervolted, that increased to over 500 in multi-core and about 100fps in OpenGL.

 

[time]

Under Windows 10 especially, I notice that even single threaded tasks get swapped from core to core, presumably for thermal reasons?

Combined with the sheer amount of crud that is running in the background these days, I wonder how effective a single or even dual core 'turbo boost' actually is? I don't seem to see maximum clock very often.

 

[Stereodude]

With the CPU undervolted, that increased to over 500 in multi-core and about 100fps in OpenGL.

that you were able to make some improvement.

Under Windows 10 especially, I notice that even single threaded tasks get swapped from core to core, presumably for thermal reasons?

That depends on the platform. The newest AMD systems will steer threads to the highest performing available core. I think Intel does something similar. How they determine highest performing available core is somewhat murky. It's not necessarily the core with the highest turbo clock capability.

Combined with the sheer amount of crud that is running in the background these days, I wonder how effective a single or even dual core 'turbo boost' actually is? I don't seem to see maximum clock very often.

How are you monitoring the clocks?

 

[time]

HWMonitor, ThrottleStop and TaskManager. I don't have all that much faith in any of them - after all, it's an internal CPU function and they're just trying to capture evidence of what's happening inside the CPU.

I also don't have much faith in Cinebench R15, particularly the OpenGL test. When I tried the VRay benchmark instead, the improvement was only about 10% or so. And when I tried to compare responsiveness with the problematic program that started it all, I couldn't reach a consistent conclusion - it seemed to help, or maybe it didn't.

Due to Chinese New Year looming, I ended up ordering another Lenovo P43s anyway. The screen and keyboard are good (although I had to settle for the 250 nits variant this time), the touchpad is good as far as touchpads go, the general design is superior (eg. fold-flat screen, very low fan noise, etc) and I really like the Lenovo mechanical docking stations - just pull the slide back and your laptop is fully disconnected and ready to pack in your bag.

From the research I did, neither Dell or HP had a better offering around that price point, for example with significantly better performance. And no-one outside of these three is offering a true docking station (Asus's effort appears to be particularly toxic), or premium support. With the paid Lenovo option, I get to talk to a native English language speaker in Sydney. I also considered an X1 Extreme, but it does not fit the standard Lenovo docks, and also the P53s and its sibling, the T590 with MX250. In the end, availability was too unreliable and/or price was too high.

If the user needs more performance, they can get a desktop.