Closed Artoria2e5 closed 3 years ago
Wait, when does AMD start having logical/core differences anyways? Isn't it just dozer with the clusters? What is this function good for anyways "usually" then, if all these SMT usually gets better performance? Shouldn't a README change be enough?
Eff this.
This patch is wrong. It should return logical core count when family is greater or equal than 0x15 not less or equal. Ryzen has family 0x17 (23).
Example: https://www.cpu-world.com/CPUs/Zen/AMD-Ryzen%207%201700.html
This patch is wrong. It should return logical core count when family is greater or equal than 0x15 not less or equal. Ryzen has family 0x17 (23).
Example: https://www.cpu-world.com/CPUs/Zen/AMD-Ryzen%207%201700.html
That’s exactly what the patch does.... (0x15 <= getCpuidFamily())
cpu family is greater or equal than 0x15 for that expression to return true.
Sorry, You're right, I replaced sides in my mind ;-), sorry!
It probably doesn't make sense to kill the whole core detection completely, but we can still make everything a bit better by not limiting ourselves to dozer only. After all, Ken Mitchell 2017 does state that SMT is useful most of the time.
Fixes #2. Should be more in the spirit of @ThomasTheGerman's suggestion.