I wonder if cooling/power is really the key here. I've got a 5950x that ended up getting the water loop I'd intended for my next threadripper - only to find they were not selling the blasted things to anyone but a few companies. With the cooling sized for almost twice what the 5950x could put out, it has been a very stable machine for some crazy workloads. That old dog will likely keep the setup when a zen 5 TR gets swapped in.

For a long time, my Achille's heel was my Bride's vacuum. Her Dyson pulled enough amps that the UPS would start singing and trigger the auto shutdown sequence for the half rack. Took way too long to figure out as I was usually not around when she did it.

> I think a lot of it boils down to load profile and power delivery

You said the right words but with the wrong meaning! On Gigabyte mobo you want to increase the "CPU Vcore Loadline Calibration" and the "PWM Phase Control" settings, [see screenshot here](https://forum.level1techs.com/t/ddr4-ram-load-line-calibrati...).

When I first got my Ryzen 3900X cpu and X570 mobo in 2019, I had many issues for a long time (freezes at idle, not waking from sleep, bios loops, etc). Eventually I found that bumping up those settings to ~High (maybe even Extreme) was what was required, and things worked for 2 years or so until I got a 5950X on clearance last year.

I slotted that in to the same mobo and it worked fine, but when I was looking at HWMon etc, I noticed some strange things with the power/voltage. After some mucking about and theorising with ChatGPT (it's way quicker than googling for uncommon problems), it became apparent that the ~High LLC/power settings I was still using were no good. ChatGPT explained that my 3900X was probably a bit "crude" in relative quality, and so it needed the "stronger" power settings to keep itself in order. Then when I've swapped to 5950X, it happens to be more "refined" and thus doesn't need to be "manhandled" — and in fact, didn't like being manhandled at all!

If you have a double conversion UPS that is complaining about less than 100W deviation, I would recommend you check the UPS for a component that is out of spec or on the way to failure.

I'm sure there are spec's for how fast a PS should be able to ramp up in response to spikes in demand, how a motherboard should handle sudden load changes, etc.

But if your UPS (or just the electrical outlet you're plugged into) can't cope - dunno if I'd describe that as cratering your CPU.

Also be careful with overclocking, because the usual advice of "just running EXPO/XMP" often results in motherboards setting voltages on very sensitive components to more than 30% over their stock values, and this is somehow considered normal.

It looks completely bonkers to me. I overclocked my system to ~95% of what it is able to do with almost default voltages, using bumps of 1-3% over stock, which (AFAIK) is within acceptable tolerances, but it requires hours and hours of tinkering and stability testing.

Most users just set automatic overclocking, have their motherboards push voltages to insane levels, and then act surprised when their CPUs start bugging out within a couple of years.

Shocking!

Yea, but unfortunately it comes attached to a Mac.

Yeah. If Asahi worked on newer Macs and Apple Silicon Macs supported eGPU (yes I know, big ifs), the choice would be simple. I had NixOS on my Mac Studio M1 Ultra for a while and it was pretty glorious.

>And my stupid MSI board resets everything (every single BIOS setting) to MSI defaults when you upgrade its BIOS.

I had the same issue with my MSI board, next one won't be a MSI.

I don't think I've ever had a BIOS that didn't reset things to default after a firmware upgrade.

I'd bet you don't care that it's attached to a Mac. I bet you don't want to switch OS. Which is understandable. In a couple of years, when Microsoft finally offers Windows as an ARM purchase, Linux is finally full-fledged done implementing support, and Apple resuscitates Boot Camp, I think a lot of people like you will look at Macs like the Mac Mini differently.

Crash on idle, interesting. Must be some timing issue related to down clocking, or maybe a voltage issue related to shutting off a core.

Have you tried using powerprofilesctl to change the power profile to 'performance' instead of 'balanced' or 'power saver'? I think this would prevent the lowest idle states at least. Just a guesss, never had this problem myself.

My modern CPU problems are DDR5 and the pre-boot timing thing never completing. So a build of a 9700x that I did that WAS supposed to be located remotely from me has to sit in my office and have its hand held thru every reboot cuz you never know quite know when its doing to decide it needs to retime and randomly never come back. Requires pulling the plug from the back and waiting a few minutes then powering back, then waiting 30 minutes for 64gb of ddr5 to do its timing thing.

> they destroy x86 lineage chips on performance per watt.

This isn't really true. On the same process node the difference is negligible. It's just that Intel's process in particular has efficiency problems and Apple buys out the early capacity for TSMC's new process nodes. Then when you compare e.g. the first chips to use 3nm to existing chips which are still using 4 or 5nm, the newer process has somewhat better efficiency. But even then the difference isn't very large.

And the processors made on the same node often make for inconvenient comparisons, e.g. the M4 uses TSMC N3E but the only x86 processor currently using that is Epyc. And then you're obviously not comparing like with like, but as a ballpark estimate, the M4 Pro has a TDP of ~3.2W/core whereas Epyc 9845 is ~2.4W/core. The M4 can mitigate this by having somewhat better performance per core but this is nothing like an unambiguous victory for Apple; it's basically a tie.

> I have an M1 Max, a few revisions old, and the only thing I can do to spin up the fans is run local LLMs or play Minecraft with the kids on a giant ultra wide monitor at full frame rate. Giant Rust builds and similar will barely turn on the fan. Normal stuff like browsing and using apps doesn’t even get it warm.

One of the reasons for this is that Apple has always been willing to run components right up to their temperature spec before turning on the fan. And then even though that's technically in spec, it's right on the line, which is bad for longevity.

In consumer devices it usually doesn't matter because most people rarely put any real load on their machines anyway, but it's something to be aware of if you actually intend to, e.g. there used to be a Mac Mini Server product and then people would put significant load on them and then they would eat the internal hard drives because the fan controller was tuned for acoustics over operating temperature.

> I have an M1 Max, a few revisions old, and the only thing I can do to spin up the fans is run local LLMs or play Minecraft with the kids on a giant ultra wide monitor at full frame rate. Giant Rust builds and similar will barely turn on the fan. Normal stuff like browsing and using apps doesn’t even get it warm.

This anecdote perfectly describes my few generation old Intel laptop too. The fans turn on maybe once a month. I dont think its as power efficient as an M-series Apple CPU, but total system power is definitely under 10W during normal usage (including screen, wifi, etc).

I'd rather it spin the fan all the time to improve longevity but that's just me.

One of the many reasons why snapdragon windows laptops failed was both amd and Intel (lunar lake) was able to reach the claimed efficiency of those chips. I still think modern x86 can match arm ones in efficiency if someone bothered to tune the os and scheduler for most common activities. M series was based on their phone chips which were designed from the ground up to run on a battery all these years. AMD/Intel just don't see an incentive to do that nor do Microsoft.

Afaik they are the fastest cores in raw speed. They’re just not available in very high core offerings so eventually fall behind when parallelism wins.

Just buy an AMD CPU. One person’s experience isn’t the world. Nobody in my circle has had an issue with any chip from AMD in recent time (10 years).

Intel is just bad at the moment and not even worth touching.

I wouldn't be so hopeless. Intel and AMD CPUs are used in millions of builds and most of them just work.

It's either bad luck, bad power quality from the mains, or something in the air in that particular area. I know plenty of people running AM5 builds, have done so myself for the last couple of years, and there were no problems with any of them apart from the usual amdgpu bugs in latest kernels (which are "normal" since I'm running mainline kernels — it's easy to solve by just sticking to lts, and it has seemingly improved anyway since 6.15).

> I thought of AMD Ryzen 7 5700

Definetly not that one if you plan to pair with a dedicated GPU! The 5700X has twice the L3 cache. All Ryzen 5000 with a GPU have only 16MB, 5700 has the GPU deactivated.

3 of my last 4 machines have been AMD x NVDA and I have been very happy. The intel x NVDA machine has been my least stable one.

If you are on Linux, there are long time known problems with low power cpu states. These states can be entered by your CPU when under low/no load.

A common approach is to go into the BIOS/UEFI settings and check that c6 is disabled. To verify and/or temporarily turn c6 off, see https://github.com/r4m0n/ZenStates-Linux

You could try using powerprofilesctl to change the mode from 'balanced' or 'power saver' to 'performance' since i think this may prevent the cpu from ever entering the throttled down low idle states that your freezing happens in. they are controlled with powerprofilesctl. You may also be able to add some flugs to grub config file. assuming you are using linux i guess.

I wonder if it's related to the main board... Are you running X or B series chipsets? I've found that less is more when it comes to stability. Vendors always add features to justify the $200-300 price of X series.

I have always run B series because I've never needed the overclocking or additional peripherals. In my server builds I usually disable peripherals in the UEFI like Bluetooth and audio as well.

Is there some tricky edge case here? I thought the “3” and “4” just denoted generations. The Ultra chips are like Apple’s equivalent of a workstation chip, they are always bigger right? It is like comparing the previous generation Xeon to a current gen i5.

I haven't moved on from AM4 yet but the way XMP is advertised you'd think it was guaranteed instead of overclocking that technically voids your warranty.