Intel GPU detection and throttling

[Mike]

I saw that there was a previous message about this, but when I went to reply there, it recommended I start a new topic since the thread hadn't been posted in for a while.

I have an Intel i5 3210M Ivy Bridge processor that has the HD4000 graphics. After installing new drivers and the latest version of BOINC, BOINC now recognizes my GPU and utilizes it. However, TThrottle isn't detecting the GPU. Therefor, I can't throttle it. Another program I found (CPU-Z) does detect it and tells me temp and speed, but doesn't have anything that would allow me to throttle it. Now that many projects are added tasks that work on Intel GPU, can we get an update of TThrottle that will also recognize and throttle them.

Right now, when my computer starts to warm up TThrottle only throttles the CPU, not the GPU. However, most times, that actually causes the GPU to run up to it's max and much hotter. I've learned that this is because this opens up more CPU resources that many GPU based tasks also need to feed the GPU info, thus allowing the GPU the ability to run full out. Unfortunately, this just causes the system to heat up faster/higher which then leads to TThrottle clamping down further on the CPU causing a cascade effect until it reaches the point that all the computer is actually running are GPU tasks only. So right now, it seems I can have GPU can either run at max (without CPU tasks really being processed) or not at all.

Is there a way to get TThrottle to recognize these Intel GPUs and allow us to throttle them like it does for other GPUs? This would really help save many Intel computers from having these problems. Also, many projects are adding more and more tasks for Intel GPU processing, so it'd be great to have a TThrottle that could handle them. Please! add this.

[fred]

I saw that there was a previous message about this, but when I went to reply there, it recommended I start a new topic since the thread hadn't been posted in for a while.

If you want to install this: http://www.hwinfo.com/ and see if it can monitor the GPU temperature.

I may integrate it in TThrottle itself or use a third party program like the one above.

[Mike]

Thanks for responding Fred. I've been monitoring the GPU temps, load/usage, power, and speed with GPU-Z. It looks like the sensor that GPU-z is using to read the GPU temp is what HWiNFO is calling the CPU package sensor (or CPU IA core sensor, they're almost always the same temp on mine).

So, either way I can keep track of the GPU temp, but neither program has the ability to throttle the GPU. That's where TThrottle comes in to save the day!! Is there anyway to throttle the GPU portion separately on an Intel chip? From what I'm seeing, it seems like the CPU portion is being throttled independently of the GPU, can it be done the other way around as well?

If any of the rest of you folks on the forum are interested in TThrottle adding Intel GPU detection and throttling, please reply or somehow second the motion. As Fred stated in the Work schedule thread "And as alway the more request for a certain feature the higher it will go on the list.
"

[fred]

Thanks for responding Fred. I've been monitoring the GPU temps, load/usage, power, and speed with GPU-Z. It looks like the sensor that GPU-z is using to read the GPU temp is what HWiNFO is

Does HWiNFO read the GPU temperature?

[Mike]

Not that I can see. It shows GPU clock, GPU D3D Usage, GPU D3D Memory Dedicated, and GPU D3D Memory Dynamic, but no specific category that says GPU temp. as far as I can tell.

That's why I mentioned that GPU-z seems to be using the CPU package sensor (or CPU IA cores sensor) temp readings for their reading of GPU temp. HWiNFO also has another temp reading called CPU GT Cores temperature reading, which, from what I've read, could also potentially be the temp of sensor at the GPU portion of the integrate chip. I'm not sure which one of those 3 temp. sensors (CPU package, CPU IA cores, or CPU GT cores) is the correct sensor for reading for GPU temps., if any of them. But no,, nothing in HWiNFO that specifically states GPU temps as far as I can see.

[Mike]

Even so, can TThrottle just use the data from the temp sensor located at the GPU portion of the integrated chip (or the sensor that reads hottest when GPU is under a full load/test) as the GPU temp data (it appears this is what other programs may be doing)? And is there anyway to then throttle the GPU portion separately on the integrated Intel chip? It appears that the CPU portion is being throttled independently of the GPU (since the GPU can run under full load/speed, while CPU is throttled low), so couldn't it be done the other way around as well?

[fred]

Even so, can TThrottle just use the data from the temp sensor located at the GPU portion of the integrated chip (or the sensor that reads hottest when GPU is under a full load/test) as the GPU temp data (it appears this is what other programs may be doing)?

I'm integrating GPU-Z, it works on my machine.
But it will take some time to get it up to production.

It should be ready early January.

[Mike]

That's great Fred! Thank you so much for looking into this issue! 

Along with integrating GPU-Z for the temperature and load monitoring, does that also mean you'll be adding throttling ability for Intel GPUs to TThrottle as well?

Thanks again,

Mike

[fred]

Along with integrating GPU-Z for the temperature and load monitoring, does that also mean you'll be adding throttling ability for Intel GPUs to TThrottle as well?

That should work as well.

[John Neale]

Hi Fred, over two years have passed, and I wondered if you still have plans to incorporate Intel GPU detection and (especially) throttling into TThrottle in the near future?

[fred]

Hi Fred, over two years have passed, and I wondered if you still have plans to incorporate Intel GPU detection and (especially) throttling into TThrottle in the near future?

As far as I know there isn't a temperature sensor for the build in GPU.

On a system without additional GPU's this will work fine, but the GPU tasks isn't throttled, in itself not a problem.
On a system with additional GPU cards this will cause a problem.