My 13700k went 9C over normal 100C tjmax without throttling.

[forman313]

View attachment 300017

A CPU running at 115C means more heating of the socket and area around. Adding 5-10c to already hot capacitors is a bad idea. Even high quality capacitors with 105C rating takes a huge hit going from 85 to 95C. Expected lifetime is probably reduced by 40-50%. At 85C lifetime might be somewhere around 25.000 hours and 15.000 hours for 95C. Add 10 more and you´re at 5000.

Im not so sure its good advice to just "enjoy the high frequencies". Running AUTO settings in BIOS can be very bad, and OP getting 1.4v @ 5200Mhz proves it.


BTW: "Normal solder" like 60/40 melts at ~185C. The solder used on CPUs, somewhere around 140-160. Assuming its lead-free, its most likely Indium with 3% Silver. Melting point 141C.

 

[phanbuey]

--

Tried a few things including raising SA voltage, but after some hours I noticed what appears to be dram voltage droop, 1.35v set in bios, 1.31v in hwinfo, boosted it and I think might be the issue.

It was failing within 20 seconds on actual memory tests, extremely quick, working at 3000, but now past that immediate at 3200 failure after fixing the voltage.

Is this failing at stock settings? Can you test each stick? you might have a bad stick.

 

[chrcoluk]

Is this failing at stock settings? Can you test each stick? you might have a bad stick.

It was failing at stock CPU combined with XMP, but its working now, the bios is still a little dodgy I think, there is dram voltage droop so when set to 1.35v it wasnt actually 1.35v it was a little over 1.31v, so I set it to 1.38v to compensate and the ram errors have stopped.

 

[phanbuey]

It was failing at stock CPU combined with XMP, but its working now, the bios is still a little dodgy I think, there is dram voltage droop so when set to 1.35v it wasnt actually 1.35v it was a little over 1.31v, so I set it to 1.38v to compensate and the ram errors have stopped.

I had a somewhat similar experience with some B dies -- they could pass 12 hours of stress tests but then crash on sleep / wakeup etc once every 5 - 10 days or so. Turns out one of the sticks was bad and once I tested just that stick it threw a ton of errors - but since I also run VM's and had at that time 32GB 4x sticks, that bad one made it seem like there was a low power voltage issue, but really it was just a dodgy dimm.

Drove me nuts and took forever to find -- ended up using Ramtest and doing it for every stick to 1000%, the bad stick would fail at less than 100%, but with all 4 in it would be very intermittent. But would randomly crash after long periods of stability.

 

[chrcoluk]

Yeah I was getting to the point I was going to throw the cheap hynix ram back in, but luckily I noticed the voltage issue, and no issues since.

--

Also now each dimm has been tested individually as you suggested, no behaviour changes between each one, but the voltage drop off increases when more dimms are populated.

Also I forgot to mention about the cache clock speed observations, I observed and recorded cache clock speeds in hwinfo whilst testing various tasks, idle load, cinebench etc, and even though there is a trigger stating cache is being throttled, the cache clock speeds dont change with and without cache undervolt, e.g. cinebench has the same cache clocks whilst running. The cache throttle trigger still appears at stock voltage as well, just not as much during idle.

 

[chrcoluk]

Fully switched over now not running on the test OS, I am running my cache and core undervolts, there was no issues at all aside from the ram which is now fine, ram is running at XMP clocks although I tightened some timings as XMP sets some really loose.

Thought I might need to reinstall windows at first, but booted into safe mode, then back into normal mode and it was fine, cleared out the old ghosted devices. Thanks to anyone who responded and offered input. TJMAX of course is manually set to 100, took no chances on it, and my pl1/pl2 are 175w, those might change up or down but I set to that for now.

 

[chrcoluk]

Got some serious performance boosts happened on NVME drives.

The expected sequential on the gen 4 980 pro, the Intel DC P4600 now matches the performance it gets in my AMD machine (was slow for some reason on the 9900k).

Both NVME have some serious jumps on 4k i/o. Wonder if thats enough to fix the stutters I got when playing ff7 remake from NVME drives. (I mitigated before by playing the game of SATA).