Originally Posted by madweazl
I've been messing around with 1002 since I shot you those results earlier; tonight while doing some stability testing I saw the temps reading in the 80s but the air coming off the HSF is fairly cool and I'm seriously doubting the temps actually being 80+. I hit the back of the socket with the temp gun and it's reading about 47° at the hottest area. Skew is disabled.
There are two possible interpretations for what happens with ITB AVX:
Three consecutive +10 C offsets are applied one after another, for a total of +30 C on top of the fixed +20 C for X CPUs. The first two happen in direct succession when memory allocation starts, the last one happens when Linpack AVX calculations start.
1. These could be pure offsets and the real temperature would be 50 C lower on a X CPU. This would be lower than the socket sensor, though.
2. Or they could be offsets only for the first few milliseconds to seconds until the real temperature finally catches up to the offset value and then takes over. If this was the case then these offsets could be seen as kind of "jump start" to ramp up fans before the real temps kick in.
Here is a screenshot of what happens when ITB AVX is started. Take a good look at the moment when Linpack calculations start! First it jumps by the +10 C offset and then it decreases
at first, before catching up again.
Sorry for using a different scale on the socket sensor. Watch out when you compare the temps (0-100 socket, 20-100 Tctl/Tdie).Edited by Timur Born - 4/14/17 at 5:01pm