[SOLVED] How do you figure out the throttle temperature?

[Jmusic88]

Hello all,

I am using MSI afterburner to try and squeeze a little more fps for CS GO on my GeForce 635M (yes I am aware I "shouldn't" be OCing laptop GPUs, but it has been done before),

The problem is that when I OC the Core Clock, between 70-85 degree C (85 tends to be the max during gaming), the core clock changes and drops to 662 Mhz (which is the base core of the GPU), but the memory clock does not change. Therefore, I am assuming that I am getting thermal throttling which drops the clock speed.

Question - how do I know exactly the temperature that it is throttling at ? The in-game monitoring through msi afterburner is not exactly precise, so it's hard to tell as there is a bit of a delay between the temperature and when the clock speed changes.

Thank you in advance

 

[Karadjgne]

There's always a delay. It's unavoidable. Intel cpu's poll the temp every 256ms, 4x a second. If you were to try and read those exact temps, all you'd see would be the digital '88', nothing but a blur. So there's a natural pause of a few seconds between a polled temp, the reported temp and the next polled/reported temp. Just to allow you to visibly see it. Only graphing will show everything and even that gets confusing with multiple cores as the reported temp is almost always the hottest core, which can change with loads.

Because of the silicon, the way its made, it's impurities etc, you will inevitably end up with stronger and weaker cores and cores with slightly different characteristics. So while one core might throttle at one temp, another core might throttle at a slightly higher or lower temp. I don't believe there's an exact measurement of exactly what temp the cpu as a whole will start to throttle. Intel says it's guaranteed to throttle at Tjunction afaik, could be wrong, but for sure it will throttle at right around that temp. How exactly nvidia calculates throttle temp is anyone's guess, I'd assume they'd follow something similar to Intel as the chips are supplied from the same manufacturer.

But you be much healthier keeping gpu temps below 80, it's easy to get high fps in CSGO, well beyond refresh, so there's really no advantage to it. You'd only be looking at 5-10fps gains at most, depending on Which 635M you have, there's like 4-5 different versions. That's chump change and not going to affect your game play at all.

 

[Phaaze88]

Download and run Gpu-Z: https://www.techpowerup.com/gpuz/
Click on the Advanced tab, then click the drop-down box and click Nvidia Bios.
That shows you the current, minimum, and maximum temperature limits.

 

[Jmusic88]

There's always a delay. It's unavoidable. Intel cpu's poll the temp every 256ms, 4x a second. If you were to try and read those exact temps, all you'd see would be the digital '88', nothing but a blur. So there's a natural pause of a few seconds between a polled temp, the reported temp and the next polled/reported temp. Just to allow you to visibly see it. Only graphing will show everything and even that gets confusing with multiple cores as the reported temp is almost always the hottest core, which can change with loads.

Because of the silicon, the way its made, it's impurities etc, you will inevitably end up with stronger and weaker cores and cores with slightly different characteristics. So while one core might throttle at one temp, another core might throttle at a slightly higher or lower temp. I don't believe there's an exact measurement of exactly what temp the cpu as a whole will start to throttle. Intel says it's guaranteed to throttle at Tjunction afaik, could be wrong, but for sure it will throttle at right around that temp. How exactly nvidia calculates throttle temp is anyone's guess, I'd assume they'd follow something similar to Intel as the chips are supplied from the same manufacturer.

But you be much healthier keeping gpu temps below 80, it's easy to get high fps in CSGO, well beyond refresh, so there's really no advantage to it. You'd only be looking at 5-10fps gains at most, depending on Which 635M you have, there's like 4-5 different versions. That's chump change and not going to affect your game play at all.

Thanks for your answer. It's a bit hard to keep it under 80 with CS GO. I have a laptop cooling pad and its maybe helping by 1-2 C. Maybe a vacuum will help but otherwise it maxes out at 85 degrees and stays there at the high peak.

I will be changing my thermal paste soon on it so maybe that will help.

That 5-10fps gain will help me greatly for 1080p resolution because I lag like crazy when it gets below 30 fps. I play 720p on it right now and its fine, but resolution is not the greatest.

 

[Jmusic88]

Download and run Gpu-Z: https://www.techpowerup.com/gpuz/
Click on the Advanced tab, then click the drop-down box and click Nvidia Bios.
That shows you the current, minimum, and maximum temperature limits.

Does this make sense?

Current: 90C
Min: 0 C
Default: 90C
Max: 90C

I think I saw somewhere on the specs that the max temp is 105C. If 90 C is the actual max then I am having cooling issues for sure.

Another question - GPU-Z says the GPU Clock & Default Clock are both at 475Mhz. Memory is 1015 while the default is 900Mhz which makes sense because I OC'd it. The boost is 855Mhz and default is 800Mhz which also makse sense because I was playing with the OC. Shouldn't the core clock OC be displayed at the gpu clock/default clock though? Specs say the default clock is 660 Mhz so why is it showing 475 Mhz? I thought it would be 715 Mhz for the GPU clock and default clock is 660 Mhz.

Thanks again,

 

[Karadjgne]

There's different values for temps. Tjunction, TjMax, Tcase etc so those numbers will be different, depending on which one you saw.

Tjunction is the temp where transistors can start loosing stable functionality, so that's the temp (@ 5°C ± below TjMax) where you'll see throttling occurring. TjMax is the point where you are into permanent damage temps and throttling is inaffective, so shutdown after an alloted period. If you push a cpu beyond TjMax, you'll be into thermal runaway and you'll not be able to shut the cpu off fast enough to prevent permanently damaging it. You are basically burning the cpu from the inside.

So if you saw a number of 105, that's most likely TjMax, which would put Tjunction at 100°C ±

Tcase is the maximum allowable temp at the IHS, but that's only relevant to engineering samples as there's no sensors there on production models, so can't be accurately measured. The core heat has to travel out of the core, through any Tim used and be at the IHS itself. Basically irrelevant temp.

Best advice, stay as far away from Tjunction/TjMax as is reasonably possible.

For a much more thorough explanation.

How To - Intel CPU Temperature Guide

Update: March 12th, 2024 Preface The topic of processor temperatures can be very confusing. Conflicting opinions based on misconceptions concerning terminology, specifications and testing leaves users uncertain of how to properly check cooling performance. This Guide provides an understanding...

forums.tomshardware.com

 

[Phaaze88]

Does this make sense?

Current: 90C
Min: 0 C
Default: 90C
Max: 90C

I think I saw somewhere on the specs that the max temp is 105C. If 90 C is the actual max then I am having cooling issues for sure.

Another question - GPU-Z says the GPU Clock & Default Clock are both at 475Mhz. Memory is 1015 while the default is 900Mhz which makes sense because I OC'd it. The boost is 855Mhz and default is 800Mhz which also makse sense because I was playing with the OC. Shouldn't the core clock OC be displayed at the gpu clock/default clock though? Specs say the default clock is 660 Mhz so why is it showing 475 Mhz? I thought it would be 715 Mhz for the GPU clock and default clock is 660 Mhz.

Thanks again,

The 475mhz is the guaranteed frequency in the worst case scenario.
The OC applies to the boost clock. There's no real reason to apply it at base speeds, as that just increases idle power consumption and reduces battery time.


If, and big IF, the actual TJ Max for the gpu direct from Nvidia is 105C, then the manufacturer of your laptop went into the gpu's vbios and lowered it further to 90C.

 

[Jmusic88]

There's different values for temps. Tjunction, TjMax, Tcase etc so those numbers will be different, depending on which one you saw.

Tjunction is the temp where transistors can start loosing stable functionality, so that's the temp (@ 5°C ± below TjMax) where you'll see throttling occurring. TjMax is the point where you are into permanent damage temps and throttling is inaffective, so shutdown after an alloted period. If you push a cpu beyond TjMax, you'll be into thermal runaway and you'll not be able to shut the cpu off fast enough to prevent permanently damaging it. You are basically burning the cpu from the inside.

So if you saw a number of 105, that's most likely TjMax, which would put Tjunction at 100°C ±

Tcase is the maximum allowable temp at the IHS, but that's only relevant to engineering samples as there's no sensors there on production models, so can't be accurately measured. The core heat has to travel out of the core, through any Tim used and be at the IHS itself. Basically irrelevant temp.

Best advice, stay as far away from Tjunction/TjMax as is reasonably possible.

For a much more thorough explanation.

How To - Intel CPU Temperature Guide

Update: March 12th, 2024 Preface The topic of processor temperatures can be very confusing. Conflicting opinions based on misconceptions concerning terminology, specifications and testing leaves users uncertain of how to properly check cooling performance. This Guide provides an understanding...

forums.tomshardware.com

I had a look at the thread, good info there thanks. It looks like I can't really find the temperature specs for the 635M. I most likely confused it with my CPU which is in fact 105 C.

So if gpu-z is correct which most likely it is. I am pushing that laptop gpu to its max pretty much. It doesn't seem to go above 85-86C. It's not on it continuous though, it ranges between 80-85 most of the time during play.

Thanks for the help!

 

[Jmusic88]

The 475mhz is the guaranteed frequency in the worst case scenario.
The OC applies to the boost clock. There's no real reason to apply it at base speeds, as that just increases idle power consumption and reduces battery time.


If, and big IF, the actual TJ Max for the gpu direct from Nvidia is 105C, then the manufacturer of your laptop went into the gpu's vbios and lowered it further to 90C.

Understood on the speeds, thank you!

It looks like I remembered wrong on the temperatures. 105C is for my cpu not my gpu. So I'm pushing it real close to the 90C mark.

I'm waiting on Arctic mx-4 to come in, so I'll reapply the thermal paste to it. Ever since I got my laptop (2012) I haven't changed the thermal paste, so I hope that will help!

Thanks again!

P.S, I am in the process of building an egpu for my laptop with a GTX 950. Just waiting on the exp gdc beast to come in. That will probably boost my fps anyway. Just thought meanwhile I'll OC my dGPU so I can squeeze some more performance out of it. It looks like at about 80C is when it throttles and goes back to stock clock speeds. So unless I can control the temperature somehow, overclocking won't do much anyway.

 

[Karadjgne]

Heh, now you know why ppl suggest not to bother with any OC on a laptop, you'll get bigger benchmark numbers that'll decline rapidly as throttle sets in, and can often lead to worse performance than stock. It's a laptop. Doesn't matter if it says 'Gaming' or not, there's some physical restrictions that simply cannot be ignored or worked around. Heat dissipation being The major hurdle.

 

[Jmusic88]

Agreed on the heat dissipation and why overclocking on laptops is not ideal.

Thank you all for the responses. For some reason I didn't get any updates so sorry for the late follow up!

Thanks again!