DISCLAIMER : I am the OCCT developer, know this before you read what I'm saying
While I appreciate the idea of protecting the users against their own problems, I think I need to put some of the things you said right.
First, OCCT doesn't use furmark in its 3D test. It's another furry donut, yeah, but this test is fully made by myself, I've got the source code, and share nothing with Furmark, but the idea behind the stress test.
Next, frying a GPU nowadays is pretty tough. They're pretty well protected, and will throttle as soon as anything goes past the limits of the GPU. OCCT adds to that another safety measure, as it you can set a temperature threshold - if the GPU temp exceeds that threshold, the test will instantly stop. I'm adding another layer on this.
I'm even adding a third layer on it, as you can throttle my 3D test (unlike Furmark), and decide if you want it to run at 100% power, 50%, 74%... whatever you pick.
I've had one occurence of a fried GPU recently, to be honest. Someone edited their BIOS and removed the built-in protections, and ran my test. This resulted in a fried card. Whose fault is it ? I'll let you judge.
Using Furmark, or OCCT, doesn't void your warranty. That is pure imagination. We are both standard programs, following the rules of what a software can do, and there's no way this voids your warranty.
Also, jayztwocents is using OCCT in some of his videos Just as Linus is.
Finally, my test is used... by AMD themselves. It's one of my customers (and I'm pretty happy about it) And others manufacturers use it daily to test their computers before sending them to their customers, and use my 3D test extensively (I can tell by the bug reports I get sometimes). So yeah, using OCCT is pretty common.
So all in all, unless you don't go all the way in removing the protections on your GPU and in the program, the chances you'll fry a card are close to zero.
Now that I'm done with the "fry cards", let's get to the gist of the issue : How come a 3D test which is pretty standard in itself (i am only doing regular Direct3d / Shader stuff, nothing else) can make a card reach its limits ? The answer is simple ; your GPU isn't meant to work at full capacity. It's meant for games, or perhaps mining, with a 50 to 70% workload in mind. So, when you buy a card boasting 3 bazillions of pixels/vertex/whatever, in reality, you're never going to reach those values, as the card will throttle way before reaching this value.
Is it normal ? I do not think so. Should this happens on a CPU, we would all riot, as that would mean encoding a movie, or doing CPU intensive task, would make your CPU throttle. On GPUs, as we think a game is a very intensive task (which really isn't most of the time), we tolerate that as we don't reach those thresholds.
What if a game reaches those thresholds ? You would have a shiny, brand new GPU that costed you 1000$, and yet... cannot run a particular game. This is going closer to happening (looking at you, New World). Is it new world's fault ? HELL NO. They're doing normal 3D stuff. The hardware design is at stake here, make no mistake. They follow standards, and should be able to run any program that uses those standards legitimately.
Now, the problem is my test is too perfect, just as Furmark is. It is pure load, optimized in every way possible, and probably too much. Adding a limit slider was the first step in trying to gain a load that's closer to what the card manufacturers impose on you, and i'm currently working behind the scenes on a 3D test update to actually be able to automatically find the sweet spot.
I do that because I'm forced to - frankly, it's against all my principles as a stability tester for 18years+. But I am forced to follow the manufacturers in their moves towards a 50-70% load max. Having a way to start low, and find the sweet spot for GPUs .
Having OCCT being accused of being a Power virus is pretty common, i'm used to this to be honest. It's always the same discussion, it's always the same answer, sadly.
Do you want to know when OCCT started getting called a power virus/ being accused of frying cards ? It's in 2008, when my test found a flaw in an AMD GPU ( HD4870 ), where the power limit caused the reference design card to shutdown instantly. Calling OCCT a power virus was a PR move to mitigate the damage back then, as the card was advertised as "highly overclockable" at that time, when my test showed it wasn't. When you're saying this, you're actually relaying the PR opinion from back then... really. When we've come to terms now, and AMD themselves are even using OCCT internally
