> I don't know much about how virtualbox is implemented, but that doesn't grok.
Yeah, it doesn't - but that doesn't mean it's untrue.
The "standard forum advice" of "not more than half your core count" - and I really want to emphasise that it is *forum advice*, meaning it's based almost entirely on experience rather than strong technical knowledge - is both correct and incorrect, depending entirely on your use case.
There are genuine scenarios where e.g. assigning 4 VCPUs to a VM on a 4C/8T system *will* destroy the performance of *some* operations (say, networking, for example),
regardless of the fact that the host remains 50% idle.
However, there are also scenarios where assigning 6+ VCPUs to a VM on the same system will produce hugely more performant results within that VM, with absolutely no measurable downside to any aspect of the entire system, including the host.
The "forum advice" IS fundamentally good advice, as it prevents users from running into unexpected issues. That doesn't mean it applies in all situations though, nor does it mean that you cannot choose to ignore that advice and be correct to do so IF your situation is one of those. But since the other forum members can't know what your scenario is, suggesting the "safe" path is the right thing for them to do, as it minimises the number of problems that other users could potentially end up asking for help with here.
If you have the appropriate technical background, an analogy may help you understand why:
Once upon a time, when hyperthreading was new, a lot of CPU cores had two INT pipelines but only one FMA per core. So on a 4C/8T CPU of that era, you WOULD get "full" 8-core performance for integer workloads, but you'd still only get 4-core performance for floating-point ops. Think of VCPUs that way, where the *core itself* is either in Host mode or Guest mode, but cannot be in both at the same time. If all 4 of your cores are in Guest mode, it *doesn't matter* that you still have 4 half-cores "free": all of your *FLOP* resources are in use, so those "free" cores are still execution-blocked behind the Guest workloads, and can't actually do anything until that stall is cleared.
(socratis, feel free to split these last few posts out of the 6.0.14 thread please, as they're not release-specific).