Is my R9 280X too hot?

[MrTeal]

What is this nonsense? Of course there are things which can be described as right approaches and wrong approaches. JoePublic thinks a synthetic and irrelevevant power virus benchmark is all the rage.

He even says real world use load doesn't matter to him if it fails said irrelevant benchmark.

The rest of the industry disagrees with him. Joe is certainly entitled to hold a stupid opinion. But it doesn't prevent the rest of us from pointing out that it is very stupid.

He didn't originally say it was poor engineering if it failed a benchmark. He was responding to Despoiler's claim that running Prime95 will break Haswell-E. An overclock causing instability with certain loads that you deem acceptable because those are use cases your computer will not see is one thing. The CPU failing catastrophically is entirely another. This isn't the 90s anymore, and it's generally considered SOP that a CPU will not allow itself to cook to death.

 

[Techhog]

This! General-purpose compute devices should be capable of executing ANY arbitrary code that conforms with the ISA, no matter how much power it draws (or it should throttle the execution such that platform power or thermal limits are not exceeded, for safety reasons).

Imagine if bridges were designed for "average" loads, and not worst-case loads, and one day, nothing but semi-trailers cross the bridge, and it collapses under the load. Would that level of engineering be acceptable? "You shouldn't have driven only semis over the bridge, it was designed for a mixture of semis and cars."

Thank God RS isn't a bridge designer (mechanical engineer). (Or are you?)

I know where you're trying to go with this analogy, but it doesn't work. That scenario for a bridge realistically can happen, so it has to be overcompensated for. A virus, however, is a virus and blaming the hardware maker for damaged caused by a virus is like blaming Microsoft for a virus corrupting Windows.

Uhm, in case you didn't know, Prime95 is a real-world program. It's NOT a "synthetic power virus".

What does it do? Or is this an "it exists and is thus a real-world program" type of arguement? Either way, the CPU and motherboard do have a way to to deal with the voltage spike: they shut off. Throttling isn't an option here due to how it works. Let me break it down:

1. AVX instructions push the CPU harder than any other instructions
2. Prime95 represents a worst case for AVX instructions, and hits the CPU all at once
3. Adaptive voltage adds extra voltage to help boost performance in high load scenarios

Now do you see the problem? All of these result in a conflict where, if all of them are working correctly, something has to go wrong, and the only way to avoid that is for one of these to stop working correctly. That's why you have to stick to fixed voltage if you want to use those instructions, because the only other way to be safe would be for Intel to require an extreme VRM solution for all motherboards. That's not reasonable, and if something can only be safely run at fixed voltage it's not a good stability test because it doesn't tell us if the CPU is stable when using adaptive voltage. The nature of the instruction makes any overvolting past stock risky on any affected CPU without a $300+ motherboard and at least a high-end CLC if you want to use adaptive voltage, making the actual stability of the CPU almost irrelevant and removing any point to overclocking in the first place.

As for Furmark, that's just completely useless at this point. With GPU Boost and the fact that graphics cards are now designed to throttle when it's run, the only thing it does is verify that the card can run at its max voltage in an extreme scenario. So, if Furmark doesn't crash within seconds, you've verified that your VRMs aren't faulty, and running it longer won't get you anywhere. Again, we've run into a situation where the only viable solution is a workaround that doesn't really get us anywhere as far as determining how stable our overclock is, and thus hasn't helped us in a meaningful manner.