Ryzen Mobile 5 4500U-hexacore review appears online

#5758330 Posted on: 02/07/2020 04:29 PM

---

I know and for me that´s sufficient. I only need powerful parts for my gaming rig, for everything else even a modern dual core is more than enough.

#5758335 Posted on: 02/07/2020 04:41 PM

---

Are you saying that HT gave you a 10-20% performance yield vs it being off? Because with modern mitigations, that sounds about right. Without migitations, HT yields an overall benefit of 30%, and there isn't much room for it to go any faster.
AMD's solution is a bit more hit of miss, because it relies on a competent scheduler and it's not as flexible about workloads as HT is. When used properly, it can result in a 50% performance gain (theoretically, even higher). For the average workload, it's not really any better than HT.

The way I like to look at it:
Intel's solution is better if you like to multitask. The idea behind HT is to fill up the pipeline with queued instructions, which works great when running many independent single-threaded tasks (or in the case of games, a thread for logic and a thread for rendering, for example). But because HT doesn't truly run tasks in parallel, this is where it can actually hurt your performance, because such an application has to wait for the logical threads to sync with the main process. In the early days of HT, a lot of gamers deliberately disabled HT.
AMD's solution is better if you run individual tasks that are multithreaded. This is because a pair of integer and FP units share the same fetch, encode, decode, etc. Unlike Intel, AMD's solution has physically separate transistors, allowing truly parallel workloads per clock cycle. But, because of the shared resources, this makes AMD CPUs very inefficient at running many unrelated tasks. So using games for an example again, a single AMD core isn't ideal to run a logic thread and a rendering thread because their instructions are very different.

Like Intel, AMD's CPUs can lose performance with SMT enabled, depending on the workload. However, now that getting many-core CPUs is becoming so affordable, AMD's method begins to look much more appealing. What would normally make AMD's SMT yield worse performance can be easily overcome by just using a physically separate core. Intel's method was great for its time, because it allowed Intel to squeeze in more performance with fewer transistors, back when a single core took up a LOT more physical space.