Dear all, I just got my first GPU which I’ll be using for deep learning. It’s an ASUS 2070 super turbo evo.
I noticed that while training simple models the card will be pretty silent and low temperature, but when I train more complex models the temperature hits 82C, while the fan stays around 29%.
Later I was able to manually set the fan speed to 50%, reducing the heat significantly to around 62C.
A couple of questions come to my mind… I can’t tell if the fan speeds that the NVIDIA X Server settings lets me use are appropriate for my board. For example, at 50% usage the fan will reach 2500 RPM. The card is now noisy but the temperatures are nice at 62C.
Do you think this can somehow harm the board or the fan in some way or it is totally ok to use this settings during training sessions? (and setting it back to “auto” when training ends)
It’s only dangerous in a way that you’re setting a fixed fan speed (e.g. 50%), so if the gpu heats up further it won’t be able to spin the fan up and may overheat.
Sure thing, but I’m setting a fixed (40~50%) speed while monitored, I’m not leaving the computer alone (I’m testing things, as I told, I just bough the card).
My question is if the card is prepared to run the fan at 50% (2500 RPM) for prolonged time, say 60 minutes for example. Or if, instead, setting a fixed speed from the Nvidia X server settings I’m “forcing the hardware”. I have no idea, it’s my first GPU in decades -last time I had a GPU was a Voodoo 2 when I played Quake II with a modem connection-.
I also wonder if it is possible to create some kind of profile for the fan speed over temperatures…