True but that’s two processors from the same family where the main difference is the fabric - 14nm going to 7nm with very little change in the instruction architecture so, yes, ipc will be the same and the improvement is clock.

@BobbyB
I am in the same situation like yours since I planned one year ago to replace my two 10 years old Phenom II x6 with Ryzen.
At first, the pandemics strongly disturbed my plan and caused a significant delay.
Secondly, since my plan was delayed, I decided to wait for the release of the Zen3 CPU which occurred in early November 2020.
Unfortunately the Ryzen 9 5900 (105 W TDP, probably very similar to the Phenom II x6 TDP) is still not really available in Europe at this time. So I am waiting again.
Definitively the Ryzen 5900 (12 cores / 24 threads) is a good choice, in particular because all cores shared the same cache and the cache has the same size (i.e. 64 MB) like the Ryzen 5950 with 16 cores / 32 Threads.
In terms of cost / performance ratio, based on the current prices, the Ryzen 5900 represents for me a good choice.
Cheers,
Yves
---
PS: Usually, I use only Asus motherboards for my machines.

The other reason to ask these questions is that on the OPN forum someone is making a case for using Raspberry Pi4s: a tower with 12 Pi4s. I estimated the cost of that setup to be about $1200 CAD.
https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,43161_offset,20

The info I can't get is how much work that setup puts out (points). My estimate based on that thread is 1 Pi4 puts out about 6000-7000 points/day which is what my 12-year old Lenovo puts out except the Pi uses about 4-5 times LESS wattage. 12x6500 = 78000 points/day.

So @FlynRyan:

How many points/day for each machine Ryzen 3950 and Ryzen 5900x

Points are my measuring stick.

It's no contest. I can't even guess why there is that much enthusiasm for the Pi4s. One device was my preferred way.

Don't want to start a war on the other forum so I'll just follow my instincts and start shopping.

Probably the single core at 2 GHz (maybe 10% or 20%) will better perform because multi-/hyper-threading is not perfect and it generates a limited CPU overhead.
However, I do operate all my multi-/hyper-threading capable machines with multi-/hyper-threading.
The other important aspect is the size of the Level 3 cache (see my previous message) because, as more concurrent threads, as higher the memory need. If the Level 3 cache is too small, the CPU will have to wait a lot for memory access and in this case the performance will decrease significantly.
This is one of the most significant trouble caused by MIP1 which has huge memory needs. For this reason, it is meaningful to define a balance between concurrent projects in order to generate the best throughput as possible but at the same time being able to support various projects with various expectations.
Generally spoken, "as many core as possible at the higher frequency as possible with the largest per thread Level 3 cache as possible" represents the most optimal configuration from a performance point of view; additionally the work/watt - i.e. credit/watt - represents the economical aspect since energy is not for free.
Cheers,
Yves