qcmadness

引用:

原帖由 Puff 於 2017-3-11 16:54 發表

It needs to be aware of the cache topology, not NUMA.

Both

qcmadness

引用:

原帖由 Sandbo 於 2017-3-11 23:11 發表

It might end up locking everything to a single CCX and uses only four core that case, this isn't ideal.

There is a new article from PCProspective :
https://www.pcper.com/reviews/Processors/AMD-Ryzen ...

There are other claims that Windows 10 misreports the cache size for Ryzen CPUs

qcmadness

引用:

原帖由 Puff 於 2017-3-11 18:09 發表

The chip is UMA, and doesn't statically partition the physical memory address space betweeb CCXs. So essentially apps just need to be aware of the topology, e.g. porting console CPU scheduling code t ...

Cat. it as NUMA simplifies the problem.
But don't treat NUMA is an advertisement as NUMA reduces efficiency and possibly performance.

qcmadness

http://www.xtremesystems.org/for ... eople-to-buy-Nvidia

第2日就走去接UAE d AMD stock

qcmadness

引用:

原帖由 BlackBird 於 2017-3-13 16:49 發表

同Deneb差不多

Deneb低頻好多

qcmadness

引用:

原帖由 BlackBird 於 2017-3-14 09:22 發表

Frequency scaling 都差不多係3.7後就要瘋狂上電4G就玩完

3.7要好後期的Deneb