TCase and TJMax are not the same. TCase is the maximum temperature intel recommends that should be in the center of the CPU heat-spreader. There is literally no way of measuring this, so the spec, while important, Isn't very helpful. As long as you keep your temps in the mid 70s you are doing great, because the Thermal Junction point on the CPU is 100 degrees and you are well below that point.
Now the reason that Prime 18+ isn't working right is because the CPU, while mostly stable, is not stable when doing AVX intensive testing. Many people will say that using Prime 18+ is an unrealistic scenario and isn't valid for stability testing anymore, but others like myself still think it's an invaluable tool to make sure your system is 110% stable.
Here's a quote from the Haswell Overclocking guide over at OCN
'I must pass all stress tests!'
This kind of thinking might had merit in previous generation CPUs, but in Haswell at least, it is a load of bollocks. As you can see from my chart below, the range of temperatures vary wildy from test to test. We are talking about a 45C difference in temperatures. If I had stuck to Linpack or go home, I would be down from 4.6ghz to 4.1ghz. (This is backed up by testing.) This is insane. Linpack is so ridiculously hot, so completely out there, it's not worth counting. The mentality of passing all tests for the sake of stability is more irrational than you might presume at first glance. That kind of mentality means passing whatever test people happen to be able to make. If nobody made Linpack, then you would think your CPU is stable. If somebody made Linpack 2.0 that makes Linpack 1.0 look like child's play, then you might as well never overclock, because Linpack is throttling a few people at STOCK. Indeed, Linpack uses AVX2 which is a new instruction set, but so does x264, and that is one of the coldest benchmarks. Stressing AVX2 set doesn't nessesarily mean high temps and failing Linpack doesn't mean AVX2 instability. And how will you know when to stop stress testing under the original ideology? You can only estimate. Computers are built for using, not for stress testing. If you're running Linpack, and you're under the opinion that you must pass all possible tests, you need to update the math logic for Linpack and run it at MAX setting. That means using up all of your available ram for the largest problem size.
Run 2-3 different types of stressing programs, and then use your computer normally. If you crash, then it's not stable. What's stable for you might not be stable enough for me. Some people need 100% reliability because of their jobs. Some people can handle a Bsod once a week. NO, saying that you want to pass Linpack 'just in case you use your CPU to extreme limits' is complete hooey. Prime95 is already ridiculous. Linpack is ridiculous on top of ridiculous on top of unicorn blood powered by the core of the sun, worshipped by space aliens. What if there comes out a new normal application that uses as much CPU power as Linpack? Well, there is no hint of that happening, so this is just a 'what if'. Well, what if there comes out a new application that throttles you at stock? Then let's all downclock our CPUs! If you insist on passing every test just because, fine, just don't expect any half-decent overclock. If I hit 95C+ easily at 1.2v with D14, there is no way anybody can hit 1.25v+ with Linpack set to max even after delid and x60 Kraken. And guess what, the average voltage setting for the OC results chart is 1.3v, so what does this tell you? You'll be lucky to stay on 1.25v after delid and liquid cooling and having a stable setting because between Prime 28.3, which discovers stability issues like a god and Linpack at max which raises temps like a god, you will be severely hampered by the combination of both tests.
Don't give me that 'If you crash on anything, you're unstable, period' crap. Anything is decided by whatever program people decided to make. And if your definition of the word stable means not crashing in anything, ever, then I don't care about what you call stability. You will never know if something is stable by your own criteria because if you pass Prime for 500 hours, what's to say the 501th hour will be stable? That's right, you stop at some arbitrary time. I care about the computer not crashing often enough to annoy me. And that could be once a week, once a month, once a year, never, every 5 seconds. But as long as I'm fine with it, that's all that matters because it's MY CPU.
If you're ever Bsoding 'too much', all you have to do, if you are in the heat of the moment, is to lower the multiplier by one and BOOM, rock solid stability.
Read more: