If I bought either the Z790 or X670E motherboard, then I could presumably look to upgrade the CPU and processor on the board at some point in the future if I wanted to increase the number of sticks of RAM. Is it worth thinking about other motherboards with a view to being upgradeable (or for any other reason)?
So far as we know, Z790 / socket 1700 is dead now, so I doubt there will be anything worth upgrading to from a 14700. There should be a new generation or two on AM5, as AMD have said they'll support it at least through 2025.
If the new CPUs and/or BIOS will offer better support for 4 sticks: I'd think so, it was the case with early AM4 versus AM4 now (e.g. people put in a new CPU and RAM that didn't work @ high speed before and it works fine now). But, I don't know for certain and I can't say if more expensive boards will be any better. I guess what you could do, is explore the boards they're using to demonstrate 192GB & 256GB, but we don't know if they're using cherry picked CPUs, RAM and motherboards.
Hopefully someone independent like buildzoid will do more exploring at some point:
My Patreon: https://www.patreon.com/buildzoidTeespring: https://teespring.com/stores/actually-hardcore-overclockingBandcamp: https://machineforscreams.bandca...
www.youtube.com
although it's hard to tell how significant a difference I'd notice compared to my current build. Do you know of any websites that could estimate/visualize the difference in performance between two component lists so I can make a comparison?
Well, the main problem is, that generic numbers like you get from Cinebench, PassMark and 3D Mark, aren't necessarily applicable to the specific tasks that you want speeded up, because it depends on how the software is written and optimised. For example: if the task mainly uses one core, then you could double your core count and see zero benefit. Equally, if graphics card is not your bottleneck, then upgrading to a 4090 may achieve nothing.
Most benchmarks are aimed at gaming and if they do include workstation tasks, it tends to be only for specific workloads that may not be applicable to your workload. That said, I'll try to give you an idea.
On PassMark, the single core/thread score of your threadripper is 2463, which is roughly comparable to a Ryzen 3600 (2569).
On PassMark, the multi core/thread score of your threadripper is 29404, which is a bit faster than a Ryzen 7600X (28645), even though the single core/thread of the 7600X is massively faster (4176).
In reality, your threadripper would only match the 7600X in long-run workloads that are fully multithreaded, in other situations the 7600X would feel much snappier and get the job done significantly faster. I wouldn't recommend the 7600X though, because I'd imagine that you really want a decent upgrade in both short-run single core/thread and long-run multi core/thread.
With this point of comparison, TPU include UE5 in their reviews (they call the benchmark: "Buiild, Cook & Release") and this is the time it took (in seconds):
- 14700K: 74.8
- 7900: 76.2
- 7600X: 110.9
By my numbers, that means in a fully multithreaded benchmark, I'd expect those CPUs to be around 45% faster than yours (if the 7600X is the stand-in).
If we compare the PassMark multithread scores, that's exactly what the difference is between the 14700K and your Threadripper (-45.0%).
For graphics:
I assume you looked at
these already, but relative FPS (UE 4.26):
- 2080 Ti: 30.1
- 4090: 91.1
Megascans Apartment RT on FPS (UE 4.26):
- 2080 Ti: 11
- 4090: 44
Looks like much higher performance to me (3x or more), but how relevant these tests are to you, I couldn't say. E.g. I don't know where you could find benchmarks for Lumen specifically.
between two component lists
It depends what the gap is between the components. If the release gap is relatively small, then it's not hard to find generic benchmarks, but benchmarks specific to UE? I only know Puget.
This channel does a lot of content creation testing, though I don't think UE is included:
E.g. the 7900 has double the number of cores/threads (12/24 vs 6/12).