You think it'll be worth waiting for?
What I see AMD did, was enable 'eco-mode' on their 3000 series desktop CPUs, that basically gave them the performance tuning of the mobile chipsets.
I'm wondering if their upcoming 4000 series desktop CPUs is worth the wait, or should I just go with a 3900x instead?

Most of the mobile chips also have only 8 cores, 16 threads max.
While they might be good in terms of efficiency, I doubt a similar 4000 series desktop CPUs will launch near the $420 street price of a 3900x, as they're newer, work better, and have a Vega GPU built in.

From the technical documents I've skimmed through, they've improved on the design a bit, calling it 7+nm; meaning there's more in the chip that's made at 7nm, and less at larger lithography, like they did with their 3000 chips; resulting in those parts running cooler, and the CPU running somewhat cooler, which helps in maintaining slightly higher boost frequencies.
Perhaps +100Mhz boost speeds (perhaps now they finally do run 4+Ghz on all cores like promised, at the rated 105W TDP?)
Nah! That'd be impossible!
Considering that most of their 3900x and 3950x CPUs run at 145W, 3,85-3,95Ghz, I would estimate that their 4000-series CPUs will only be a minor improvement on current design, and speculate that they will run 4Ghz at ~150W, or run 3,7-3,7Ghz (non-boost frequency) at their rated 65/105W.

Been scouting the markets for most energy efficient chipset, and Ryzen CPUs make the most sense at 7nm.
Even more so, than ARM does, as they're still at 12nm at best, and while they're just as efficient, they're much slower and more costly than Ryzen.

Ampere would do well when they finally bring compatible ARM processors to the market (at sub 10nm).
Their 80 core CPU costs way more than a 3960x, and runs way lower frequencies too, while consuming about the same Watts as a 3900x.
The competitiveness is off. AMD made a 'good' call, to extinguish ARM for this time; as their x86 CPUs will, for a while, still be able to run inside servers.

The X86 architecture is old, and inefficient compared to ARM.
But for as long as X86 will have the upper hand in lithography, ARM won't ever play a role in server market.

The good thing is, that we're nearing the absolute minimum possible with current technology.
To make an x86/64 CPU below 5nm is a process of diminishing returns.
And it may take many more years than 2, to make a stable CPU at those sizes.
I guess it's at that point, that ARM actually makes sense.

ARM is basically beating against x86, x86 having core disabling features, as well as a variety of P-states, that ARM/Risc yet doesn't have.
If you look at their multi core CPUs, even the quad cores, they run usually around 80% of their TDP at idle; making them more suited for 100% load, 100% of the time (think Boinc).

I do hope in the near future, AMD or Intel will do more research in ARM/RISC.
RISC would be better, as it doesn't require a license.
AMD has abandoned their Opteron A-series processors, as they (at 14nm) weren't as efficient anymore as their newer 7nm x86 chipsets.
Too bad, because I wished they would re-vamp their Opteron A-series chips, at 7nm, with 32+ CPU cores!

At least AMD listened to its audience (whine) when they said that the next generation Ryzen needed a new motherboard and that X470/B450 wasn't going to be supported. They've now said they're looking into support for these chipsets, if the BIOS can hold all the information.

from what I've heard, they will provide support in the form of a Beta BIOS, that drops support for older CPUs to make room for the new CPUs. kind of a kludgey workaround but hey whatever works. and the BIOS will be up to the motherboard vendors to provide, not direct from AMD, so expect support timeline to vary from vendor to vendor.

---

One of the larger structural changes is that Zen 3 will have one 8 core CCX per compute die vs the 2x 4 core CCX that we’ve had previously.

This enables all 8 cores on each die to share the full 32mb L3 cache instead of two 16mb slices that are faster for 4 cores and slower to be accessed by the other 4 on the die. Zen cores cannot access the L3 cache of separate compute dies. This should increase L3 available per thread when lightly threaded and simplify the NUMA arrangement at the cost of slightly higher L3 latency.

Haven’t heard how else they’re getting to the claimed 15% IPC improvement, and even with it, performance per watt will not jump nearly as much as the 7nm process allowed for Zen 2 to bring with the enhanced 7nm being a small improvement.

I was actually reading up on those zen 4000 mobile cores,
And if there's a slight possibility that some companies will sell those soldered mobile versions of the 4000, for under $400 a board, I might actually purchase them!

Some industrial applications are about to release slightly larger than Pi-sized boards with a CPU on.
A $400 board like this, is cheaper than a Ryzen 3900x to maintain, for the same amount of cores.
Plus, they can all be stacked, and all use a regular 12V DC jack plug. So easy to power them as well!
And they're using much less power than a Ryzen 3900x (either 35W or 65W flat; no 95W or 145W).

Only con with those boards is they only run from a single SODIMM RAM slot.
Hoping that some manufacturers will at least allow 2x SODIMM slots for Dual Channel Mode.

Considering that most of their 3900x and 3950x CPUs run at 145W, 3,85-3,95Ghz,

I don't run anywhere that much power (wattage) on my R9 3950X using 28 of 32 cores running at 4250Mhz. ZenMonitor or Zenpower shows only around 105W of power being used across the SoC and the Cpu Cores.
SVI2_P_Core: 87.27 W
SVI2_P_SoC: 17.06 W

Considering that most of their 3900x and 3950x CPUs run at 145W, 3,85-3,95Ghz,

I don't run anywhere that much power (wattage) on my R9 3950X using 28 of 32 cores running at 4250Mhz. ZenMonitor or Zenpower shows only around 105W of power being used across the SoC and the Cpu Cores.
SVI2_P_Core: 87.27 W
SVI2_P_SoC: 17.06 W

Very unusual, considering that the majority of the posts record wattages well over 200W on the wall (I run 202-220W) on 3900x and on my 3950x, with only a mild overclock.
If you disable PBO, the CPU might stick to 105W, however, the memory controller, and other things are still on the soc as well, consuming power.
Once PBO is enabled, and XMP, power of the CPU increases by just under 50%; and that's with not even hitting 4Ghz on all cores!

That is why I don't run PBO or Performance Enhancer. Those mechanisms kick the hell out of Vcore in an attempt to boost 1 or 2 cores to the theoretical maximum clocks for gaming. But when you load all the cores up with crunching, your average all-core clocks plummet to ~ 4Ghz or lower depending on your cooling.

Since I don't game and my systems are primarily nothing but BOINC crunchers, I simply set a manual overclock on all cores at a MUCH lower Vcore than what PBO shoves and gain a MUCH lower operating temperature for the PC. I also gain a fixed overclock on all cores that does not waver or vary. So performance is very stable across all loads.

I also run an overclock on the memory. I run 3600Mhz CL14 Fast timings on my 32GB of TridentZ RAM.

Interesting news in the last few days of the Theadripper Pro processors with 8 channel memory support. I wonder if additional memory channels are a possibility with future Ryzen processors.