Topic: Control cards clock speed for lower temperature (Read 1905 times)

If i search around on google, there are plenty of different opinions. ranging from 80 to 95 centigrades   At the moment i'm settling them between 83 and 85 C. I guess the more interesting part is the warranty in case the card fails. in my script I enable overdrive, set the peak range and disable overdrive. Thus fgrlx should IMHO have the possibility to slow down the GPU if it's temperature is getting critical. Are there any experiences if this voids the warranty?

To make this clear. I'm not overclocking the card in terms of firmware editing i'm just using the overdrive command from aticonfig (and furthermore I set the VRAM clock speed to minimum peak range when i start the miner).

Depends on the card, so cards are fine up to 90C, even in the long run. On average though, I'd say you wanna stay below 83C possibly even below 80C, so stabilizzing bvetween 755~82C should be fine

Alright. Here is what i have so far. Not much but it works for the moment.

http://www.bitcoiners.org/files/mining-supervisor

It definitely needs some error checking but at the moment I'm tinkering with an USB temperature sensor in order to control room temperature. As soon as i'm ready with this task i'll have some more time for this script.

Capacotprs are the 1st components to be deteriorated by heat. I woudnt to that high for the long run.

AFAIK this is too much for a card. In the long run you'll decrease it's lifespan significatly.


Of course, I am .

I use two 5870 for mining. With the last days being quite warm in Germany they are getting quite hot. I found, that ATI is quite clever by lowering the clock rate if the GPU is getting too hot. It seems this happens if the GPU is going over 92°C. However the clock rate decreases to the peak ranges minimum (600 MHz on this cards) and increases to the max peak range (900 MHz) a few seconds later. Giving some kind of yoyo-effect. After some time it wont recover from 600 MHz even if the temperature falls significantly.

For this reason I hacked a "Quick and Dirty"[tm] shell script which I can use as a cron job. Each run it iterates through the cards, checking their temperature and in- or decreases the clock rate for a given amount (25 MHz in my case) until the temperature stabilizes between 85° and 90°C. It does this for each card seperately as my cards have different temperatures.

Now I have two questions:

1. Is the temperature (85-90°C) okay for this GPUs in the long run? Or should i consider lower temperatures?

2. Is anybody interested in this code? I could put it somehwere. My hope is that this script gets better over time if more people use it and contribute. But i think it wont make any sense to put more work into this piece of code if nobody is interested. Perhaps it makes sense to flesh it out becoming a real daemon which also handles starting and stopping the miners. Furthermore it could make sense to shut down mining if the room temperature goes above a defined threshold.

I guess this is of no interest for heavy overclockers or people with air conditioning. But i hope for people like me using their desktop PC in a normal home office it might be of interest. It'll give maximum stable performance which is great if you just want to "fire and forget" your miners without the need for checking the temperature hour for hour as the temperatures raise. Just enjoy the summer with your family instead of worrying about the miner temperature 

This is a sample output from syslog:


May  9 01:04:02 newton mining-supervisor[3504]: card 0 at 84.50 C - Below threshold. Dethrottling from 850 to 875 MHz
[...]
May  9 09:05:02 newton mining-supervisor[30469]: card 0 at 90.50 C - Above threshold. Throttling from 875 to 850 MHz
[...]
May  9 11:11:01 newton mining-supervisor[14198]: card 0 at 89.50 C - Within safe range (85-90 C) running at 850 MHz
May  9 11:11:01 newton mining-supervisor[14198]: card 1 at 83.50 C - Below threshold but card is already at full speed (900 MHz)