本帖最后由 埃律西昂 于 2023-3-5 09:48 编辑
明显只有两组GMI链路
Chips and Cheese主编@System360Cheese和Locuza的对话中提到了这一问题。
Things always change mate, in the beginning everything is open.
I can assure you that 3 CCDs was a possibility when I was talking about it, try to prove the opposite.
Checkmate, Atheist.
======
No no no, AMD was planning a 4 CCD version all along by connecting 2 CCDs to each IF link.... They were just going to round robin the bandwidth don't ya know.... Totally sane configuration......
======
Sure...
And just at the very end, the absolute beginning of it, they changed that to just 2x GMI links, because it was clear by then that the >20% higher core IPC was enough to destroy anything Intel anyway.
(Thumbs up for C&C of rethinking the essence of published material )
======
Yeah.......
I was so wrong when I published that..... Hell, I even had some others at CnC telling me not to...... But I did anyway and....... Well..... I regret being dumb enough to publish that article but I don't regret publishing it because it's what made CnC what it is today
Without the backlash from both inside and outside CnC (mostly from
@chiakokhua
), I would not have rethought what CnC should be
I published a retraction of it in our Zen 4 deep dive but I didn't remove it because I thought it would be disingenuous to and it serves as a reminder
======
In the last 15 years a lot of my expectations turned out to be wrong, in positive and negative forms.
Info I thought was correct, turned out wrong, things I thought were clearly bullshit turned out right.
To some extent there is also nothing wrong with it, as long as someone...handles it honestly, like you did!
There is nothing wrong or dramatic here in confirming fault.
If a person reflects and adjusts next time, it's just great.
I find it so annoying if people try to deflect screw-ups, moving the goal post or retroactively try to spin it around.
|
310112100042806
发表于 2023-3-5 09:40
