Topic on Glossary talk:Central Processing Unit (CPU)

I'm not talking about the usual just-so-frequent incompatibilities, crashes, or other obvious problems, and not even about the various hacked overclocking tools. But something a bit more in the shadow:

• Original Phenoms cpu had a very concerning bug with TLB. Gaming isn't as sensitive of a task as compiling tax returns though, so bold people could try to disable the software workaround.

• Llano has quite reduced DIV/IDIV performance (I'm really not the the guy recommending to undo errata fixes, but considering all the recent guarding around speculative execution.. maybe there is a chance?)

• Bulldozer has crippled x87 performance (there are no public errata regarding the feature, so this might be all free real estate?)

• Bulldozer (and I could swear Bobcat) CPUs are liable to system locking up on steam CEG games loading. Given the time window, and given some wine discussion involving debug registers, I believe this might have been erratum 671 (Trinity's AGESA has indeed provisions for it)

• Possibly every architecture before Ryzen could hit the TDP ceiling before even reaching the maximum frequencies (unclear whether this couldn't just be specific to some motherboards and/or workload though)

• ok this is probably just Turbo Core (whatever its features unclear progression) being dynamically adjusted according to the current load. It can be fooled then, and some OEMs may actually be enforcing All Core Turbo from the firmware (Turbo CPB, Turbo Performance Boost Ratio and Turbo Unlocker being some likely names?). Or at least I can't explain why else the same CPU getting downclocked for clear TDP reasons would be reported elsewhere happily running dozens of watts above the official limit.

• Steamroller CPUs are preemptively throttled when their iGPU is in use (which is not as bad as it sounds usually, even in games it should be hard to be CPU-limited without a dedicated graphics card after all, but it is not impossible). If a modded BIOS isn't available, editing p-states and disabling disabling APM may still do something, but not very much:

• this is unlike Trinity being able to sustain even turbo on both simultaneously (though with its idiot balls), Llano just barely knowing how to turbo its CPU, Jaguar somehow regressing with just one CPU model out of stone age (Puma came back with the very interesting concept of frequency sensitivity though) and Bobcat being a (very) mixed bag.

• speaking of APM, that's also necessary condition for turbo boost states (that are in turn a godsend for non-K-unlocked models). Unless sure it's holding the processor back (or overclocking some plutonium core) it's very much better off enabled.

• 2014 core designs were somewhat of a backstage trainwreck, on the other hand their power management started to rely more and more on the SMU. Nothing of interest in the HEDT market, but on laptops it introduced a buttload of hidden interesting variables.

• all PCIe 3.0 motherboards should support somehow Resizable-BAR (needed for Smart Access Memory), but before Zen 3 CPUs a critical instruction for such graphical applications was emulated slow enough to make it useless

• I couldn't find anybody on the net seriously reporting that forcing HPET helped them save for touching BCLK live or under W8(+?). Still for some reason something in the AMD drivers always does/did that.

And I could swear I knew more, but time was hard on my memory. I wonder if all of this couldn't be enough to resurrect the old AMD (CPU) page we had.

p.s. Ryzen is so stacked complex and wide that you'd need a manual just to understand the stock situation.