I suspect cooling for higher clocks is better on the 3900x due to the multiple CCDs.
Instead of 1 fully loaded CCD with 8 cores of heat in that area, the 3900x will have 2x 75% active CCDs. I.e., 75% of the heat to dissipate in each core complex die.
Following not in reference to wendell… but the OP
Also indicated in another thread on here, the 3900x has 2x the memory write bandwidth of the 3700x due to the two dies.
Of course, it is more expensive, but my advice (based on reading a bunch about them both this week) would be:
- If you are or are planning to do VFIO - go 3900x. Both for additional cores and memory write bandwidth.
- If you’re flush with cash, the 3900x will be faster for heavy multitasking.
The impact to total system BOM cost is not huge. Yeah its a more expensive CPU but your build cost won’t go up a huge percentage… you’ll get up to 50% more throughput for much less than 50% more money in terms of total system cost (motherboard/RAM/storage costs are constant).
If you already have an AM4 box and aren’t buying everything new, then the above cost vs. performance advice may differ…
The times the 3900x is getting beaten in gaming is likely (IMHO) due to cross-die communication like with previous gen Ryzen. If this is important to you, go 3700x.
But IMHO i think the tradeoff in those very rare cases is worth it for the potential VFIO memory bandwidth and more cores.
But either CPU will be a great choice, IMHO.
edit:
Also yes, as above future proofing is not a thing (though with AM4 at least your motherboard will likely get future CPU support).
Buy what you need today, etc. - but if you think your workload can benefit from more cores, i’d go 3900x. And if your workload can’t benefit from more cores… i’d suggest maybe 3600x (instead of the 3700x), and put the money away you’d spend on 3700/3900 for next year’s 4000 series.
I think i just kinda ruled out the 3700x to be honest. For simple gaming, etc. the 3600x is plenty. If you need more cores, then buy a 3900x - if you can afford it (and if you can’t, save a few more weeks or whatever). On core hungry workloads the performance improvement will be more than worth the additional spend - assuming you’re doing an entire system build.