The past few years of hardware have left me feeling so frustrated; I wish, I wish someone, whether it's AMD or Intel, would acknowledge the fact that there is a market for extreme single-thread performance, and provide us some sort of "real" server platform; unfortunately, not everything parallelizes well — be it due to things I theoretically have control over [1], proprietary software in which I have no control over [2], or software/algorithms that simply inherently do not scale to dozens, let alone hundreds of threads [3].
It feels like the only step-up from incredibly I/O-limited Ryzen/Alder Lake is Threadripper or Sapphire Rapids WS; I was actually super excited about the latter — I mean, really, wow up to unlocked, 24 P-cores on a monolithic die, what's not to love? — until I felt totally gutted by the fact that, as far as I can tell, they intentionally crippled SPR by removing QAT and pretty much everything else from the WS chips that made Sapphire Rapids an exciting platform, all whilst a w5-2455X + motherboard is, at least, 2-3x the price of an equivalent AM5-based platform simply for the virtue of PCI-E + memory.
Similar feelings towards Threadripper Pro, and even more egregious, the latest EPYC 4004 series, skipping out on AMD SEV; it's so disappointing not being able to offer even the tiniest of guarantees that, "Yes, I promise I cannot snoop on your VM", because there is nothing currently available that provides what you'd actually expect from server-grade infrastructure while still offering at least similar performance to commodity desktop hardware.
I wish someone would please just offer a low-medium core-count, unlocked server chip with the bells and whistles you'd expect; I don't mind paying 3x the price if they don't intentionally screw you over in an effort to prevent non-existent cannibalization of the high-core-count market (how many people are really going to be pushing 600W+ through their CPU unless they have no other choice? I can't imagine any hyperscalers want to pull 1000W per rack unit).
While I know things like the 3930K did not have ECC, how I miss the X79-era; it was beautiful having more-or-less the equivalent of an unlocked 6-core Xeon, minus ECC, with quad channel memory, and non-crippled PCI-E connectivity for less than a current high-end desktop chip -- not to mention the fact that my "absurdly expensive" Rampage IV Extreme was... ~$350? And it at least still included a 7-segment display!
Meanwhile, the last time I upgraded my personal system, though to be fair, this includes the Russian 20% VAT, god, my X570 Aorus Ultra was something like ~$450 USD equivalent and does not offer anything in the realm of I/O, and apparently that is no longer enough money to get you a 3 cent post-code speaker, let alone a <$1 USD 7-segment display.
I wanted to rip my hair out when it died; all I had was a useless pair of debug LEDs that would constantly fluctuate between "CPU" and "DRAM" -- I swapped out literally every component before realizing it was a broken motherboard trace that I had to resolder.
[1]: There's loads of open-source software that one "could" refactor and potentially alleviate some of these bottlenecks. But often times, this is not actually practical; many of these are some combination of decades-old ancient codebases that are very difficult for even the core maintainers to work on; monstrously complex engineering feats with insufficient tests that would be incredibly difficult to re-write from the ground-up while maintaining compatibility; and/or incredibly critical software that you maybe really shouldn't be mucking around in, unless you want to cause some sort of disaster.
[2]: Game servers are a common one. Things like SRCDS are essentially entirely single-threaded programs that you realistically cannot easily modify nor do you have permission to modify; trying to push 100+ players on something like Garry's Mod, where not only do you have to do real-time updating for ~100 players and all the other entities in the game, but you have constant physics interactions and such -- I was literally colocating overclocked, liquid-cooled servers at one point to try and keep up.
[3]: Many video compression algorithms fall into this category, as well as tons of other things. Personally, I work on some server emulators for an older MMORPG; despite being "greenfield", in the sense that it's all code written from scratch, many of the game mechanics are designed for a mostly single-threaded environment, so there is a limit as to what can be improved without fundamentally breaking the game. Minecraft seems to be in a similar boat; there are higher-performance server mods out there, but they come at a sacrifice of accurate emulation.
