Info Zero RPM GPU fan settings are the enemy!

[alceryes]

I just discovered something VERY interesting. Zero RPM fans have been a thing for over a decade, but I didn't realize how much of a detriment they can be to GPU temps.
Zero RPM fan settings are the enemy!

BY DEFAULT, the Radeon software had my GPU fans completely off until around 62ºC (guessing standard GPU temp sensor). Now I believe that, to go from zero RPM to say 60% fan RPM probably takes around 2 seconds. Next, for the airflow to start having a positive effect on the heatsink/radiator (cooling it down) probably takes another 2-3 seconds. So we have a span of about 5 seconds where the cooling potential of my reference cooler is not realized and the GPU is allowed to get hotter than the cooling is supposed to allow.

Previously, my GPU hotspot was ALWAYS peaking at 99ºC in a particular game. It was a 100% definite. After playing this game for any length of time HWiNFO64 would report a hot spot temp of 99ºC. I am now fairly certain that the zero RPM setting is the culprit for this burst temp reading! I just set a very lazy/low fan curve (barely audible even with headphones off) and now, with the same game, the hot spot temp peaks at 78ºC. A full 21ºC lower!!! I am completely floored by this result! I want this card to last 4+ years but Radeon's own stock driver settings has the card getting WAAAY hotter than it should (even if it's just a 5 second burst temp), possibly lowering the lifespan of the card.

Unfortunately, I have to run into the city and will be there for a few days so won't be able to test further but holy S$%#!!! Does any one else here run with a zero RPM fan setting?

 

[Karadjgne]

Nope. I turned that thing off years ago and don't bother with it. My Asus kicks in @ 65°C. Gaming temps would hit that in a decent amount of time, game wasn't that demanding so it'd constantly kick in, ramp up high, shut off. Very annoying. Turned off 0rpm, game gets to a comfy 55° with 35% fans continuous.

You assume it's a burst. It could very well not be. Consider the VRM's and vram is also relying on that heatsink and without active cooling they will roast until the fans kick in.

 

[alceryes]

You assume it's a burst. It could very well not be. Consider the VRM's and vram is also relying on that heatsink and without active cooling they will roast until the fans kick in.

Possibly, but it could still be a burst temp reading.
I'm assuming it's a burst temp reading because of the relatively short amount of time it took to get that 99ºC reading before. Less than a few mins, previously.
Nothing else has changed but more testing does need to be done. As far I i know, 'hot spot temp' is the hottest part of the GPU die.

 

[helper800]

I personally have always had the fans spin at their minimum rpm until 60 C then ramp up 15% per 8C. With those fan settings games that are naturally very demanding usually immediately go to around 65-75c, so no problems there. Games that are much lighter usually never go past 58 or so with minimum fan settings so the card is super quiet.

 

[helper800]

As far I i know, 'hot spot temp' is the hottest part of the GPU die.

I think the hotspot reading is the hottest sensor reading on the card period.

 

[alceryes]

I think the hotspot reading is the hottest sensor reading on the card period.

Source?
Most posts/informational articles talk about it being just the die's hottest part.

 

[hotaru.hino]

I've played around with this on my RTX 2070 Super. I have a case fan basically pointing right at the video card and that one's always blowing, while the video card's fan curve is 0 until the GPU hits 50C with a hysteresis setting on so it actually waits a while before it commits to an RPM change. Unless I'm a reasonably heavy game, the GPU stays under 50C. I can even throw a 4K video on and the GPU doesn't ramp up that much. I'm not concerned with the other parts of the video card since, unless I'm throwing a large enough load on it, HWiNFO reports the video card runs anywhere from 23-35W.

Otherwise in the past I've left it at basically 30% until it gets to 50C.

 

[helper800]

Source?
Most posts/informational articles talk about it being just the die's hottest part.

I said I think, but obviously am wrong after doing a bit of looking into it. I would think the hotspot would include all components and not just the GPU die. I personally have an EVGA 3080 that comes with a bunch of extra thermistors all over the card. I think I may have conflated the two somewhere.

 

[Phaaze88]

The feature might be helped along if the gpu is mounted vertically, just like a psu with a 0rpm mode should be turned fan up and not down.

 

[alceryes]

The feature might be helped along if the gpu is mounted vertically, just like a psu with a 0rpm mode should be turned fan up and not down.

It would help with heat dissapation through convection, definitely.

The biggest shocker was the actual difference in hot spot max temp between fans spinning but practically silent and a zero RPM setting. I would've never expected there to be an over 20ºC difference.
I definitely need to do some more testing. Unfortunately, I was about to get ready to go into the city when I stumbled upon the huge temp difference.

Additional reason why I think it was a burst temperature reading before the fan cooling kicked in is that my current fan curve is actually lower, across all points, than the default setting. Only difference is that I just have it spinning slower and slower instead of stopping all together at a certain threshold.

 

[alceryes]

I said I think, but obviously am wrong after doing a bit of looking into it. I would think the hotspot would include all components and not just the GPU die. I personally have an EVGA 3080 that comes with a bunch of extra thermistors all over the card. I think I may have conflated the two somewhere.

Ahhh, makes sense.

 

[Karadjgne]

As its name suggests, the hotspot is the hottest spot on the GPU, measured from a network of thermal sensors across the GPU die, unlike conventional "GPU Temperature" sensors, which reads off a single physical location of the GPU die. AMD refers to this static sensor as "Edge temperature." In some cases, the reported temperature of this sensor could differ from the hotspot by as much as 20°C, which underscores the importance of hotspot. The sensor with the highest temperature measurement becomes the hotspot.
https://www.techpowerup.com/278606/nvidia-gpus-have-hotspot-temperature-sensors-like-amd?cp=2

So, not hard to figure the 99°C Hotspot is just that, the hottest actual point in all the gpu's sensors and could likely stay around that temp easily, if the 'gpu temp' is what the fan curve set by 0rpm or default is reading, and it's closer to 79°C ±. Explains why nvidia starts throttling at 83°, because it knows all to well the Hotspot is getting overly close to 100°+ possibly or periodically.

 

[alceryes]

So, not hard to figure the 99°C Hotspot is just that, the hottest actual point in all the gpu's sensors and could likely stay around that temp easily, if the 'gpu temp' is what the fan curve set by 0rpm or default is reading, and it's closer to 79°C ±. Explains why nvidia starts throttling at 83°, because it knows all to well the Hotspot is getting overly close to 100°+ possibly or periodically.

However, when I set a low but constant (no zero RPM) fan curve, the hot spot didn't get above 78ºC. This was the shocker for me.
Again, I think it's due to the estimated ~5 second delay in the fans turning on, spinning up, and airflow/static air pressure cooling down the heatsink/rad. If the GPU is really getting hammered during those 5 seconds this zero RPM setting allows the GPU to get way hotter than it does in identical scenarios but with a very low, practically silent fan speed.

 

[helper800]

However, when I set a low but constant (no zero RPM) fan curve, the hot spot didn't get above 78ºC. This was the shocker for me.
Again, I think it's due to the estimated ~5 second delay in the fans turning on, spinning up, and airflow/static air pressure cooling down the heatsink/rad. If the GPU is really getting hammered during those 5 seconds this zero RPM setting allows the GPU to get way hotter than it does in identical scenarios but with a very low, practically silent fan speed.

You should check clock speeds in a high polling graph next to power draw and fan rpms during this instant as an experiment. If the hotspot is getting too high and the fans take about 5 seconds to respond then you would see some pretty severe downclocking or power draw dips in that 5 second time frame, right? There are not many ways a GPU can regulate temperature when the fans are not moving. Some of them is downclocking, undervolting, and power dipping.

 

[alceryes]

You should check clock speeds in a high polling graph next to power draw and fan rpms during this instant as an experiment. If the hotspot is getting too high and the fans take about 5 seconds to respond then you would see some pretty severe downclocking or power draw dips in that 5 second time frame, right? There are not many ways a GPU can regulate temperature when the fans are not moving. Some of them is downclocking, undervolting, and power dipping.

I will definitely test further when I can.
The temp spike as the fans spin up and start to cool the heatsink/rad is still within spec for hot spot temp for this GPU so I don't think any downclocking is occurring. Although it is within spec, it still spikes much higher than when I have the fans on all the time.

It's possible these spikes are shortening the life/causing unnecessary degradation. That's my take away and why I will always have these fans spinning, going forward.

 

[Karadjgne]

However, when I set a low but constant (no zero RPM) fan curve, the hot spot didn't get above 78ºC. This was the shocker for me.

Exactly, Hotspot hitting when no fans. Having fans running is constantly dissipating heat from the heatsink at a much higher rate than no fans. Having the 5 second delay in 0rpm mode is making the heatsink (which is already getting, if not already saturated) try to absorb that heat in a passive mode during the spike.

 

[Krotow]

Probably depends from how particular graphics card is built. I can't say about latest Radeon-s, have no one of them. With more or less latest Nvidia cards which I and friends had/have, complete fan stop while temps are in acceptable preset limits, caused no issues neither overheating happened. I know that Radeon-s generally run hotter. Probably it is better to have fans always spinning in them.