Originally Posted by Kaltenbrunner
All the brands have gone for BLING instead of actually cooling stuff right. Personally I hate it. In the long run all these mobo's will die sooner too. I don't want to void my warranty yet by modding my new z370 mobo, but I really better add a fan or 2 just for the mobo.
My i5-9600k takes a lot of volts aswell, and runs really hot. Someone said it's the cache use in these when running P95 blend that really heats up something too much, and causes throttling, and/or errors.
These big companies love BLING, because then they don't have to sell products on facts or numbers, but pretty designs. And and quicker it wears out in 2-3-4 years, the more they sell.
I think you're mixing things up. Firstly Blender is a real application that can run just as hard and hot as Blend with the same power draw.
Fitting all instructions in the cache (FFTs in place) is what generally makes p95 hot becasue the CPU is rarely stalling where as most other applications stall out waiting on instructions.
Not running fixed FFTs allows more ram use and runs cooler...that's Blend plus it's cycling through all the FFT sizes as it runs.
Throttling and errors aren't P95's fault. If a CPU is OC'd it's the users fault for using too much or not enough vcore or using a CPU that draws a lot of power through a smaller VRM.
If it's doing that on a stock CPU it's likely becasue the board is using MCE and allowing MORE than Intel's TDP (95w) or becasue Intel has been upping the stock OC on 14nm for the last few years. A crappy cooler could also be a problem.
What's funny is that you have to exceed Intel's TDP to get the advertised boost specs, if you keep it with in the 95w range it will likely boost to around 4.4ghz and run pretty cool using p95.
None of this is P95's fault IMO, it's just an application running instructions. It's the CPU/board/users fault if it can't run ANY instruction that's passed through it. Anything past TDP and Intel's clock table is an overclock, boards typically allow MCE (single core turbo across all cores) and over current by default.
FWIW boards have better VRMs than they've ever had before. Look at the Phantom 6 and 9's VRM and the cost of the board! That's a heck of a deal especially on the 6 (though the 9 is a better board). Look at all the features a modern board comes with! Imagine the work it would take to integrate all that crap at such a low price point.
I hate all the garbage RGB plus the software and ugly shrouds but it's what the average consumer wants. No idea why but they really wan their PC to look like a riced out Civic complete with underglow and body kits. But in spite of that I appreciate all the work that goes into making a cutting edge VRM with Wi-Fi, 10Gb LAN, USB-C, lots of SATA and M.2, strong PCIE slots, fan headers everywhere, thunderbolt, a quality soundcard and so on and so on.
Boards today are awesome for people to EXPECT it to OC using MCE to force turbo and bypassing TDP by default and then to take it even further on their own. Many users complained when they were putting 8700k's on the cheapest board they could get and having limited OC potential. IMO all these problems are the users fault which they agreed to when they bought an unlocked CPU and an OC board. If the user doesn't know enough to get a board with a good VRM for overclocking that's totally on them. It's only Intel's or the board's fault when it fails to meet stock specs.
Hell look what the OP was able to do, 5ghz on an 8c/16t CPU using a mid-low end 8 phase VRM made for a 6c/12t CPU tops. My 4790k board was mid-range with 8 phases lol.
OP, if you're getting errors it's not stable, period. Especially on something as easy to run as RealBench. You need more vcore or lower speed, ignoring errors could (likely will) lead to crashes or data corruption. Could also be related to other things such as XMP or auto cache settings but I'm assuming you're following the typical OC guidelines.