Does anyone know what's up with AVX2 heating up Haswell to seemingly unsafe temperatures? Check out this anandtech thread where the OP claims that the stock heatsink at stock settings will heat up to 100C with AVX2 instruction stress tests.
I believe him. I build quite a few PCs at work, and I'm teaching technicians how to do it so I can focus on more appropriate software development tasks.
With Haswell CPUs I have found that a really good closed loop cooler at a mere 1.25V fixed VCore will heat up to nearly 90C at 4.3 GHz with AVX2. I do this with Prime95 small FFTs. That's just fucking nuts. Without AVX2 these things barely hit 70C. I could easily jack up VCore to 1.3V and get most CPUs to 4.5 GHz, so this is lowering potential OCs.
Should we not worry about this considering Intel's own stock cooler/settings hits 100C?
Is Broadwell going to get rid of this absolute garbage thermal solder between the die/shield? Because with the same settings on Sandy Bridge at 4.6 GHz and 1.3 VCore, I barely breach 55C.
AVX2 primarily heats up Haswell by, outside of manual mode, allowing the CPU to receive up to 0.1v more than what you've got it set to, presumably for stability reasons. But on top of that, AVX even with the older chips did heat them up more than without and AVX2 is no different in that regard, even with a fixed voltage. Intel's thermal solution is inadequate at this point, certainly.
Devil's Canyon (the Haswell Refresh) is supposedly fixing the thermal issues (though Intel haven't disclosed how, aside from ambiguous "improved thermal interface" and "better packaging" type statements), so hopefully that carries over to all future CPUs. I'm curious how Intel will continue managing voltage usage under heavy AVX load, though, as that's my primary complaint with the current chips and Skylake is supposed to feature some monstrous 512-bit AVX3.2.