[whyoh]

It's crazy how unreliable CPUs have become in the last 5 years or so, both AMD and Intel. And it seems they're all running at their limit from the factory, whereas 10-20 years ago they usually had ample headroom for overclocking.

[stavros]

That's good, isn't it? I don't want the factory leaving performance on the table.

[topspin]

I do. I've been buying Intel for the same reason as the author: I build machines that don't have glitches and mysterious failures and driver issues and all the rest of the garbage one sees PC assemblers inflict on themselves. Make conservative choices and leave ample headroom and you get a solid machine with no problems.

I've never overclocked anything and I've never felt I've missed out in any way. I really can't imagine spending even one minute trying to squeeze 5% or whatnot tweaking voltages and dealing with plumbing and roaring fans. I want to use the machine, not hotrod it.

I would rather Intel et al. leave a few percent "on the table" and sell things that work, for years on end without failure and without a lot of care and feeding. Lately it looks like a crapshoot trying to identify components that don't kill themselves.

[stavros]

So underclock your CPU.

[dahauns]

How about you "overclock" (overvolt, unlock TDP etc.)?

This is about sane, stable defaults. If you want the extra performance far beyond the CPUs sweet-spot it should be made explicit you're forfeiting the stability headrooms.

[stavros]

Good thing that's not the point I was making, then!

[dahauns]

Maybe not intentionally.

[topspin]

Because I'm not a CPU engineer, and neither are you. Neither of us can claim anything about fucking around with CPU clocks and voltages or anything else about any of this. If you want to screw around in BIOS settings and learn where all the sharp edges are and spend your time like this, enjoy. I've never done this nonsense and I never will.

[stavros]

I know enough to tweak the "voltage" slider down a few numbers, and that's enough to get more stability. Otherwise, I vote with my wallet, and don't buy CPUs that break, which is why companies don't generally make CPUs that break.

[dahauns]

>which is why companies don't generally make CPUs that break.

Well, that's the issue, isn't it? Both Intel and AMD (resp. their board partners) had issues in recent times stemming from the increasingly aggressive push to the limit for those last few %.

[bell-cot]

Depends on your priorities. That "performance on the table" might also be called "engineering safety factor for stability".

[makeitdouble]

TBF using more conservative energy profiles will bring stability and safety. To that effect in Windows the default profile effectively debuffs the CPU and most people will be fine that way.

[therein]

So now you're saying just accept the fact that they come pushed past their limits, and the limits are misrepresented. Factory configuration runs them faster than they could in a stable fashion.

That sounds terrible.

[stavros]

Given that there used to be plenty of room to overclock the cores while still keeping them stable, I think it was more "performance on the table".

[formerly_proven]

You could also get the idea that vendors sometimes make strange decisions which increase neither performance nor reliability.

For example, various brands of motherboards are / were known to basically blow up AMD CPUs when using AMP/XMP, with the root cause being that they jacked an uncore rail way up. Many people claimed they did this to improve stability, but overclockers now that that rail has a sweet spot for stability and they went way beyond it (so much so that the actual silicon failed and burned a hole in itself with some low-ish probability).

[devnullbrain]

Yep. Redundancy and headroom are antonyms of efficiency.

[techpression]

The 7800X3D is amazing here, runs extremely cool and stable, you can push it far above its defaults and it still won’t get to 80C even with air cooling. Mine was running between 60-70 under load with PBO set to high. Unfortunately it seems its successor is not that great :/

[williamDafoe]

The 7000 series of CPUs is NOT known for running cool, unlike the AMD 5000 series (which are basically server CPUs repurposed for desktop usage). In the 7000 series, AMD decided to just increase the power of each CPU and that's where most of the performance gains are coming from - but power consumption is 40-50% higher than with similar 5000-series CPUs.

[scns]

When you use EcoMode with them you only lose ~5% performance, but are still ~30% ahead of the corresponding 5000-series CPU. You can reduce PPT/TDP even further while still ahead.

https://www.computerbase.de/artikel/prozessoren/amd-ryzen-79...

[techpression]

I specifically singled out the 7800X3D though, it runs incredibly cool and at a very low power draw for the performance you get.

[mldbk]

> You know, I'm something of a CPU engineer myself :D

Actually almost everything what you wrote is not true, and commenter above already sent you some links.

7800X3D is the GOAT, very power efficient and cool.

[Numerlor]

The only reason the 7800x3d is power efficient is because it simply can't use much power, and so it runs at a better spot of the efficiency curve. Most of the CPUs won't use more than ~88w without doing manual overclocking (not pbo). Compare that to e.g. a 7600x that's 2 cores fewer on the same architecture and will happily pull over 130w.

And even if could push it higher, they run very hot compared to other CPUs at the same power usage as a combination of AMD's very thick IHS, the compute chiplets being small/power dense and 7000 series X3D cache being on top of the compute chiplet unlike 9000 series that has it on the bottom.

The 9800x3d limited in the same way will be both mildly more power efficient from faster cores and run cooler because of the cache location. The only reason it's hotter is that it's allowed to use significantly more power, usually up to 150w stock, for which you'd have to remove the IHS on the 7800X3D if you didn't want to see magic smoke

[hu3]

Same for 9800X3D here, which is basically the same CPU. Watercooled. Silent. Stupidly fast.

[k4rli]

7900X same. System uptimes of 1month+ often and nearly always runs at 5.0Ghz. Never goes above 80c or so either.

[mrheosuper]

we have unstable "code" generator, so unstable CPU would be natural.