Nvidia throttling - Thermal slowdown
Thermal throttling is a state that happens when your GPU is taking too much load and starts overheating. For Nvidia 3000 series such events are quite common, so it is extremely important to have good cooling system and smart position of your GPUs. It can happen that GPU drivers sometimes doesn't report this state. Thermal throttling can cause: unstable GPUs, hashrate drops, drivers crashes, rig crashes, and any other instabilities.
For some Nvidia GPUs, you will see this error displayed as:
Nvidia throttling thermal slowdown can happen when temperature is too high.
When this event is triggered the core clocks will be reduced.
Thermal throttling can be solved by:
- Increasing the speed of your fans to cool down the GPUs even more;
- Reducing your overclocking settings and making them less intense;
- Improving your cooling and ventilation system of the rig;
- Replacing thermal pads.
Keep in mind that not all GPUs are affected by this.
Where can I read memory temperatures?
Currently, memory temperatures can't be read from the drivers for Linux. We are looking at ways to add it and Nvidia developers also said they will add it to the drivers reporting soon. As soon as we will be able to read and report this data, we will add it. On Windows workers, memory temperature is reported.