Ubuntu 21.04 CPU throttle because of enabled (not used) nVidia gpu

[StoykovK]

I used Prime95 to stress the CPU (using AVX2). It stabilized at 3.0GHz@45W after few minutes. It was expected result so nothing new here.

After that I installed nVidia driver (470.x). In nVidia control panel I selected to use performance mode (only nVidia gpu). I thought that this will free the cpu even more and the clock speed should be at least equal or even higher than initially measured. I run the stress test again and for my great surprise the CPU stabilized at 2.7GHz@45W after few minutes.

I confirmed the results by running several tests. That’s why I decided to switch back to integrated GPU . After the restart everything was as before. Then I tried balanced mode (nVidia gpu only on demand). In this case the results were identical to performance mode .

I run several more tests by switching GPUs/X11<->Wayland, but the results were always the same. With integrated GPU the CPU was running at 3.0GHz/~87 degrees@45W. With nVidia - 2.7GHz/~82 degrees@45W.

After few more research I found that on idle the package cpu power consumption was as follows:
Using iGPU (power saver mode in prime selector): ~2.5W
nVidia (performance/balances mode in prime selector): 5-12W

During prime95 stress the package cpu power consumption was as follows:
Using iGPU (power saver mode in prime selector): 45W (~42.5W for Core itself)
nVidia (performance/balances mode in prime selector): 45W (~33W for core itself)

So somehow enabling the nVidia GPU causes CPU to use ~10W more power. This is only while using Linux. On Windows - everything is fine.

Does anyone have а suggestion or am I missing some kernel/nVidia setting? Is there any way for detailed cpu power consumption inspection?

I’m willing to test/debug anything. I would like really to resolve this issue.

Thanks

[DeividA_Intel]

Hello StoykovK,  

Thank you for posting on the Intel® communities.   

In order to better assist you, please provide the following:  

1. The brand and model number of your system. You can also provide the Intel® System Support Utility for the Linux* Operating System for more details: https://downloadcenter.intel.com/download/26735/Intel-System-Support-Utility-for-the-Linux-Operating-System

2. What is the name of the CPU model used?

3. Is this issue recent or from the first time you used Linux?

4. Does the issue persist if you disable the iGPU from the device manager?

5. Have you checked with Nvidia to confirm if there is a problem with their drivers on Linux?

6. Have you updated your operating system to the latest?

Regards,  

Deivid A. 

Intel Customer Support Technician 

[StoykovK]

Thanks for your reply. Here are the answers to your questions:

1. Asus FX753VD.

2. Core i7-7700HQ

3. I didn't make any such tests before so I don't where now recent this "issue" could be. I don't know even if it is issue or expected behavior. 

4. There is no device manager in linux or at least I'm not familiar with it.

5. I post in their forum but still no answer.

6. Yes, it is updated

Update:

The case turns out to be even more interesting. All said above is true, expect one thing - the OS. After a lot of tests I concluded that the situation is the same both on Windows and Linux.

Let's focus on idle package power consumption (IPPC) of the CPU.

1. Using "Power Saver" mode in Prime results in using only iGPU and completely turns nVidia off. In this case the IPPC is <2.5W.
2. Using "nVidia on demand" mode in Prime results in using iGPU unless explicitly selected nVidia. In this case both GPUs are always active. No matter how many processes are actually using the nVidia - it is always active! In this case IPPC is ~5-12W.
3. Using "performance" mode in Prime results in using ONLY nVidia gpu. It is always active. In this case IPPC is ~5-12W.

Later on I tried to find why the IPPC is in so broad range - from 5 to 12W and how it is connected to the GPU. Luckily I found it. The answer was in power states of the GPU (P-states). When the GPU is using P0 (max performance) then the IPPC is 5W. In any other case the IPPC is more (up to 12W). It sounds quite counterintuitive but.. it turns out to be true. The faster the GPU runs the less affect it has on the CPU thus the CPU package power consumption is reduced and vise-versa - the slower the GPU runs, the more power CPU drains.

Now let's go to Windows. It has only one GPU mode - "on demand". The difference with Linux is that on Windows when no app is using the nVidia GPU then it is turned off thus resulting in IPPC <2.5W. If any app starts using the nVidia GPU then the IPPC go up to 12W. So on Windows the logic is the same as Linux - the faster the GPU runs, the lower CPU's package power is.

Note: I checked power consumption on Windows on idle. Power of the Core/IA/GX were almost 0, while package is 5-12W when nVidia enabled.

According to all findings the issue is not related to the OS/GPU. Somehow turning dGPU on (no matter of the usage) causes high CPU package power consumption (thus leaving to lower frequencies when CPU is loaded). What could be the reason? Is there something wrong with nVidia drivers for both Windows/Linux? Is there some specific CPU component that is activated when dGPU is running?

[DeividA_Intel]

Hello StoykovK, 

Thanks for the information provided.

In order to continue further, can you provide me with the report from the Intel® System Support Utility?

- https://downloadcenter.intel.com/download/26735/Intel-System-Support-Utility-for-the-Linux-Operating-System

Once I receive the report, I will proceed with the research.

Best regards,  

Deivid A.  

Intel Customer Support Technician 

[StoykovK]

Here is the report. Thank you very much for the effort.

Best regards

[DeividA_Intel]

Hello StoykovK, 

Thank you for the information provided 

I will proceed to check the issue internally and post back soon with more details. 

Best regards, 

Deivid A.  

Intel Customer Support Technician 

[DeividA_Intel]

Hello StoykovK, 

In order to continue further, please provide the following information:

1. The issue was happening with Linux and then it happened with Windows, was Windows re-installed as a standalone OS?

2. Can you provide the report of the Intel® System Support Utility from Windows?

- https://downloadcenter.intel.com/download/25293/Intel-System-Support-Utility-for-Windows-

3. Take a screenshot from the task manager using Windows and also a screenshot of the 3rd party tool that you are using.

4. Is there any app from the laptop manufacturer that he can use to check these features?

5. Please, run the Intel® Processor Diagnostic Tool using Windows.

- https://downloadcenter.intel.com/download/19792/Intel-Processor-Diagnostic-Tool

6. Is there any error on the computer? Shutdowns, BSODs,...?

Regards,  

Deivid A. 

Intel Customer Support Technician 

[StoykovK]

Hello. Here is the information you requested.

1. It happens on both systems and I think that it is "by design". The issue (if it is an issue) seems to be normal behavior independent from OS, installation order of OSes or whatsoever. I've posted this questions in other forums as well and other people also confirms it. You can here here and here.

2. The report is attached as SUU-report.txt.

3. I attached three screenshots:

power-no-gpu.jpg: This screenshot indicates the power consumption while nVidia gpu is inactive. As you can see the Rest-of-the-chip power is quite low hence the CPU power package is also low.

power-lowest-freq-gpu: This screenshot indicates the power consumption while nVidia gpu is active but running at lowest frequency as possible. In this case you can see that the Rest-of-the-chip power is quite high hence the CPU power package is also high, which is expected, because Rest-of-the-chip power counts in total CPU power budget.

power-highest-freq-gpu: This screenshot indicates the power consumption while nVidia gpu is active but running at highest frequency as possible. GPU itself is almost running at idle. In this case you can see that the Rest-of-the-chip power is quite lower than the previous case but still a 6 times higher compared to the first case (~3W, compared to ~0.5W).

4. There is no specific software from Asus.

5. The report is attached as diagnostics.txt

6. No, there is nothing wrong with the computer.

[DeividA_Intel]

Hello StoykovK, 

Thanks for the information provided, 

I will proceed to continue with the research and I will post back with any details for you.

Best regards, 

Deivid A.  

Intel Customer Support Technician 

[DeividA_Intel]

Hello StoykovK, 

Based on the information provided, look like the computer is not having a bad performance, however, check if the Enhanced Intel SpeedStep® Technology is active in the BIOS, if so, then to deactivated it, that should increase the base frequency to the base or more than that; therefore, the power consumption too.

If you do not find the option, try contacting Asus for instructions.

Regards,  

Deivid A. 

Intel Customer Support Technician 

[StoykovK]

Your answer is not related to the problem/questions/research. It doesn't answer what effect dGPU has on the CPU's power budget?

It is not having "bad" performance. Just in some cases it should be able to do more. Anyway, that's NOT main part of the questions.

The questions was "Why CPU package power when dGPU is enabled is so high?", which leads to the question "Why enabling the dGPU have any effect on the CPU power consumption?". Also why CPU + iGPU results in less CPU power (both idle) than CPU + dGPU (both idle). 

[DeividA_Intel]

Hello StoykovK, 

Allow me to investigate further in order to provide you with a complete answer, as soon as I have any updates I will let you know.

Thanks for your comprehension.

Regards,  

Deivid A.  

Intel Customer Support Technician  

[DeividA_Intel]

Hello StoykovK, 

I would like to let you know that processor directly communicates with the memory RAM and the PCIe, so when something is connected to the PCIe the processor will need more power consumption for the device to work, in this case, the dGPU. Also, is important to know that this issue could be related to hardware and software installed in the computer, not only to the processor and dGPU.

In addition, I recommended checking the Enhanced Intel SpeedStep® Technology is because this technology will dynamically adjust the voltage and core frequency, decreasing average power consumption and heat production and that could explain that the processor works at really low levels of power and frequency while the dGPU is not active.

I recommend you to check with Asus for the Intel SpeedStep® Technology settings and Linux for drivers and firmware.

Best regards, 

Deivid A.  

Intel Customer Support Technician