Discussion in 'Article Discussion' started by bit-tech, 22 Nov 2019.
So who decided the "a second core in the same CCX that’s nearly as fast’" is something that is considered optimal for Zen 2 performance, AMD or Microsoft?
AMD. They're twiddling the reported core priority to force threads to be distributed between CCXes rather than to fill one CCX and then the other (optimum for threads that need to access the same data, not optimum for thermal management). Similar to how they fiddled with the temperature readouts for X-suffix Ryzens to force fans to run faster on those CPUs by pretending the temperature is 20°C higher than it actually was. Once it was noticed they tweaked Ryzen master to then show the correct values (previously it reported the 20°C offset values).
The non-bodged soltuon would be to work with Microsoft to update the Windows scheduler, as occurred with the WX-series Epycs. The perils of heterogenous CPU layouts.
Just noticed, in the third paragraph below the summary:
"...arbitrary selection that removes the potential for rounding areas when the scale is then interpreted..."
I believe the intended word was, "errors".
I would say it's another example of how AMD dropped the ball, yet again, on what's essentially a good product but I'm worried about incurring the wrath of AMD advocates.
TBH I don't really think the whole thing is in the ballpark of being an issue, let alone being anywhere near a ball being dropped.
N.B. That was neither said in wrath nor in a grumble grumble fashion
Yea, sorry. I didn't mean issue in that way, as in causing problems for customers, i meant more of an issue in that they didn't have to fiddle with things so much after releasing something.
While it's nice to make changes to improve things it would be nicer if there wasn't a need for improvements after somethings been released, if what's released, and reviewed, was largely what customers buy some six months or so down the line.
Separate names with a comma.