Graphics Would you return this 1070?

boiled_elephant · 14 Jan 2020

Posterity edit 2020-01-14: this card ended up being fine. The artifact in Elite Dangerous' menus was simply an anomaly. I have never found an explanation for it, but nothing else ever went wrong. Also, EVGA's warranty/support people are lovely and top notch.

Bought this card yesterday. Works fine in Tomb Raider (2013), Far Cry 3, Elite: Dangerous and Unigine Valley at 3840x2160/60Hz. It artifacts in this way in the menus of Elite: Dangerous, and only there (so far?), when scrolling through menu items.

I've seen plenty of GPU failures but I can't really decide if this really looks like a hardware artifact or just a GUI/graphics glitch within the game engine. Thoughts?

GeorgeK · 19 Mar 2019

I would check firstly that all cables / monitor leads were fully plugged in and then I'd hammer it with something like furmark for a few hours and see what it's up to then. Seems very odd to me.

boiled_elephant · 20 Mar 2019

Furmark is informally suspected to be a hardware killer due to its unrealistically continuous battering of the power circuitry (link lost, I used to have a very technical explanation of it from a forum). Handily EVGA have 3-year transferrable warranty so I'mma see if the seller can provide original receipts, just in case, and then just get as much gaming time in on it as possible to see if it falls over. Such a chore, but it's gotta be done...

edit - found the original posts, from Bobnova on Overclockers:

Power draw kills 570s and 590s.
Furmark kills cards, it's a proven fact.

Unlike a CPU, a GPU is not designed to be 100% utilized. The thing to keep in mind is that a game that shows 100% utilization is not the same as a math problem that actually fits into the couple k of L2 cache in each shader, most of that 100% utilization in games is spent waiting for data to show up.
That's what makes Furmark (and OCCT GPU) so nasty, the cores aren't waiting for data, they're all doing something.

It's much like how Linpack/IBT gives higher temps than Prime does, Linpack is designed to minimize wait states.

In any case, it is very much not FUD, nor is it entirely heat related, except in that a saturated mosfet or trace generates far more heat than that same mosfet 00.5% less loaded. We're talking a difference along the lines of the mosfet keeping 5w and the mosfet keeping 25w here. Very large.
That's what happens to gtx570 mosfets and gtx590 mosfets, the core(s) demand more currant than the mosfets can cough up, and the extra demand turns into massive, massive heat in the mosfet, much more than can be removed via a standard heatsink. That, in turn, equals BOOM!
Click to expand...

On temp vs amps:
First a disclaimer, the specific numbers in this example are made up on the spot.
The effect and situation described is real.

Technically, it's always heat that causes the death. Always.
The thing is, if you exceed the maximum amount of amperage that a mosfet can deal with it suddenly goes from something like 99% efficient (only spewing out 1% of the power going through it as heat) to 90, 80, 50, or less.
So this poor little mosfet that is designed to switch 50 amps at 1.3 volts (65w going through it) is used to having to get rid of maybe .65w of heat. Not really that much, but enough that a tiny component should probably have a heatsink.

Now lets say that the absolute maximum amperage that this mosfet can sustain is 55a, that gives us a 10% safety margin, not bad at all.
Now some user finds the voltage slider and the OC slider and cranks it up to 1.4v, and due to that and the OCing this poor unfortunate mosfet is now switching 60a at 1.4v. There's 84w going through it now, but rather then putting out 1% of that (.84w, still not that much) efficiency has fallen through the floor and now it's putting out 30% as heat. That is 25 watts.
The component can't even get 25w to it's surface without overheating, let alone get it to the surface, through the TIM, and into the heatsink.
It quickly goes past it's ~125*c rated maximum temp, the hotter it gets the worse the efficiency gets, it's a positive feedback loop.
Like all positive feedback loops, it ends with destruction, eventually the mosfet looses it internally, and odds are it starts arcing inside.
So it goes POP! And then most likely it gives your GPU core 12v to think about.

Alternatively, your card may be designed such that in 3d loads, even nasty hardcore 3d loads, it's at 50a of 55a allowable, but furmark's math load puts it up to 56a. Pop!
Click to expand...

https://www.overclockers.com/forums...-video-cards?p=6869254&viewfull=1#post6869254

Log in or Sign up

Graphics Would you return this 1070?

boiled_elephant Merom Celeron 4 lyfe

GeorgeK Swinging the banhammer Super Moderator

boiled_elephant Merom Celeron 4 lyfe

Share This Page

Log in or Sign up

Graphics Would you return this 1070?

boiled_elephant Merom Celeron 4 lyfe

GeorgeK Swinging the banhammer Super Moderator

boiled_elephant Merom Celeron 4 lyfe

Share This Page

Useful Searches