Prime95 crashes everytime on same core

[chalsall]

Originally Posted by CRGreathouse

But I must disagree with the video's claim that "a sample size of 2 is absolutely meaningless from a statistics perspective".

I agree with you.

A sample size of one is meaningless; a sample size of two simply has a very large margin of error...

[VictordeHolland]

Originally Posted by M344587487

Reviews from big players also need to be taken with a pinch of salt, sending out golden review samples is probably common practice: https://www.youtube.com/watch?v=pl0we6-ZiQY

Thanks for sharing that video, I always assumed it was done (since lower temperatures/ higher OC/lower power draw get more favourable reviews) but he has clearly put a lot of time in it to prove it.

The key for the thread starter is: only 88% of the 8700K's are able to hit 5GHz AFTER de-lidding. As de-lidding lowers temperatures by 12-25C and increases the max OC by 100MHz or sometimes 200MHz you're probably need a chip that is better than >50% of the chips to be able to do it without delid. So it is not a shame if your chip doesn't hit 5GHz.

[tServo]

Originally Posted by chiro

Well, I have Kraken x62 cooler. After OC my temps were 90-91°C on all cores so that's pretty bad. But I'm planning on doing delid but I'm still trying to find some retailer in Europe that has it in stock or ships it into my country.

I tried prime without OC and my temps were all 80+°C exept one core which was 90°C. That's even worse I guess.

As for memory I have G.SKILL 16 GB KIT DDR4 3200 MHz CL16 Trident Z RGB.

By "normal" I meant I didn't change clocking or voltage of CPU manually but via my mobo's 5Ghz profile.

Your cooler and memory are excellent. G.Skill Trident Z is known for being the best for OC.

It would be interesting to use an AVX offset of -2 as Mackerel suggested in reply post #5 to see if and how much that helps.

[mackerel]

Originally Posted by M344587487

Reviews from big players also need to be taken with a pinch of salt, sending out golden review samples is probably common practice: https://www.youtube.com/watch?v=pl0we6-ZiQY

That guy is salt, and attacks anything he perceives is either negative to AMD, or positive to Intel.

Overclocking is never as simple as set desired clock and enjoy the performance. There will be different stability criteria and use cases that need to be considered.

For what it's worth, my Coffee Lake i3-8350k does 5.0 GHz stable for non AVX workloads, but it requires a voltage not considered safe for 24/7 running. I use that for competitive benching, and drop it down again afterwards. In my specific case, I didn't find cooling a problem with high end air (Noctua D15) before de-lid. Temps did drop after de-lid but didn't get me any more OC headroom. The 50% extra cores of the higher models I guess will mean more power to be dealt with...

[M344587487]

Originally Posted by mackerel

That guy is salt, and attacks anything he perceives is either negative to AMD, or positive to Intel.
...

I don't know, I get the impression that he could be overzealous but I've tended to agree with the point of everything I've seen him say. I have an obvious bias for AMD but I do try to moderate my opinion, if you have a particularly egregious example of him being a fanboy I'd like to see it.

[mackerel]

Suffice to say I had a negative personal interaction with him some years back. I have noticed he's getting more mainstream airtime since around Ryzen's launch, but I've decided not to spend my time on him beyond the inevitable references like these when hanging around various tech forums. His vids do end up promoting discussion, for better or worse, but I'll leave the analysing of details to others.

Edit: to clarify, there are factual elements, and there are opinion parts. I guess it is how the first part leads to the 2nd, and their applicability that is more open to interpretation.

[mackerel]

Some additional thoughts on the cherry picked CPU samples thing... if nothing else, it has stimulated debate in the tech community.

Hardware Unboxed just did two-part video where they got 10 new retail 8700k CPUs (two boxes of 5) and overclocked them to see how far they got. They said their original review sample from Intel seemed to be a golden sample, hitting 5.2 GHz at under 1.4v. The retail samples varied with most hitting 5.0 or 5.1, with one sample reaching 5.2. So Intel did give them a great sample, but that doesn't answer in itself if it was intentional.

Part 1: https://youtu.be/jXYPrS9WCOE
Part 2: https://youtu.be/GRviKkVUAa0


An interesting comment in there was that Gamers Nexus' original sample was an OC lemon in comparison. They said it did 4.9 ok, 5.0 runs but wasn't stable in all tasks at 1.425v.

https://youtu.be/oCSkyNHXIAE

If Intel were cherry picking samples, how does that one fit in? I'd rate GN well above average in technical ability compared to the average tech-tuber. If the CPU was capable of more, they would have got it. Some may come up with theories like, maybe Intel slipped up in the samples, or they gave him a bad one as revenge for some past event. I'd still think it more likely they weren't cherry picked and that's silicon lottery for you.

Don't know if anyone has done it, but go over all the 8700k launch reviews again, and see who got what?

[retina]

My personal thoughts on overclocking are that it isn't worth it. Not for me anyway.

My personal list of things rated from most important and going down is this:

1. Ability. Can it do the tasks I require? Enough RAM? Enough HDD space? Enough register width?
2. Correctness. Are result correct? Can it detect when the results are bad? That calls for ECC.
3. Reliability. How often do I get bad results?
4. Efficiency. How much power does it need to complete each task?
5. Performance. Will it do the task quickly?

Throughput can always be improved by adding more machines, not by trying to increase the performance.

If I try to improve the performance by overclocking then the other, more important, factors will suffer.

Many of my machines are underclocked. The reliability is awesome and the efficiency is improved. For the same power usage I can have more machines and increase the overall throughput.

If I overclock I'd have to restart many tasks, and the overall throughput, efficiency and reliability would fall.