Cross-posting this from another thread, with more added here on AM5 boards killing CPUs - because its echoing what we've seen on AM4, as well.
There's two separate issues:
One: Asus and gigabyte decided that to be the OC'ing champions, they'd raise SoC values really high.
Two: As we've seen on AM4, board makers just don't bother enabling any safety features and rely on the CPUs hard coded limits.
Conclusion: A new CPU comes out that relies on the boards having safely set limits and shit hits the fan.
PBO has "motherboard limits" - yet i've NEVER seen or heard of a board that ACTUALLY IMPLEMENTS THAT. They just run straight to infinite values, and rely on the CPU to implement some kind of limit, which as we all learned from the 5800x seems to be some generic maximum for the entire lineup - 8 cores? 16 cores? Who cares! Have The same values!
Here's an ASUS Tuf x570 board and shows "motherboard" as a choice with no explanation what the values actually are
You don't know what auto or motherboard uses - only disable and manual
This particular motherboard has pretty decent VRMs, but imagine these settings on a board without that? Oh wait, you don't need to imagine.
When reviewers test the boards, they compare with PBO disabled or static all core overclocks and don't expose themselves to random PBO values
They also don't re-test with newer CPU's released years later, or if they do they use default PBO off settings, with XMP if you're lucky.
View attachment 293976View attachment 293973
View attachment 293974
These are the default manual limits... which are likely what auto and motherboard use as well.
That video creator goes "Uhhh.... I like to start with the default limits of the CPU involved" and then manually enters them
View attachment 293975
You can see why this would disable a whole bunch of safety features on the board, since those values are higher than I Dunno, every AM4 CPU ever made?
It's a board that when released, the top CPU's had a 110W peak wattage -
and then next gen CPU's could shoot to over 220W with those exact same BIOS settings
View attachment 293977
TDC is AMD's throttle safety feature and seems to drop amperage if the CPU is overheating, like setting performance to 65% to lower temps, instead of outright dropping to minimum clocks.
This only really happens on low end cooling, like the stock wraith coolers.
TDC will not limit or save your CPU if your VRM's are overheating, because the CPU cools rapdily when its got no power to it so the full amperage is still being sent and the VRM's never recover from this throttling state.
This is what causes stuttering for people running motherboards designed for a 65W CPU that can spike to 220W+ if PBO is enabled without manual settings.
Everything i'm seeing in that GN video now that i've had time to watch all of it, shows something i've seen in AM4 and argued with a lot of users here on TPU and on facebook about.
Board makers are not enabling or using their OCP and OVP safety features If you don't manually set limits.
The biggest issue i keep seeing is people
enabling PBO to adjust curve undervolting which enables all other PBO values to effectively unlimited values that relies entirely on the CPUs hardcoded limits.
Some CPU limits like TDC will never enable, if you have sufficient CPU cooling.
They had a non overclockable CPU receive 400W in the socket and desolder itself. They had multiple boards throw unsafe SoC voltages into the SoC, as well.
The board that sent 400W in, had no VRM heatsinks -
they should have been thermal throttling, but it was disabled on the board.
In all the various threads on TPU about people with stuttering issues, it always turns out they've got a motherboard with budget VRMs, PBO enabled with "auto" settings and the assumption that "motherboard limits" and VRM thermal throttles will actually do their job, when it seems to be entirely upto the CPU resulting in stuttering performance as the CPU throttles when the voltages go out of spec
PBO's got EDC (electrical max) values and TDC (Thermally constrained max) values assuming the VRM's will communicate when they're hot, and lower to a safer limit to run a middle-ground, but because theres no communication happening at all between these components the CPU seems to have to guess what the hell is going on based on the voltage it's receiving and is being forced to throttle itself to minimum clocks, rather than to TDC.
It'd be exactly like intel motherboards changing PL2 to 800W with no time limit if you enabled XMP
