To reach these high core counts (relative to China’s current manufacturing capabilities), Loongson is using a quad-chiplet layout interlinked with its Loongson Coherent Link (LoongLink) technology to achieve a 64-core configuration. LoongLink is Loongson’s equivalent to Intel’s mesh interconnect, Nvidia’s NVLink, and AMD’s Infinity Fabric.
This is the strategy AMD used with their Epyc server chips to take marketshare from Intel in the server market, and it works.
AMD also used it for Threadripper, and AMD has taken the HPC market completely away from Intel.So this is absolutely an excellent strategy to compensate for being behind on manufacturing process.
It depends on the use case though. Not all workloads benefit from lots of slow cores. And both Intel and AMD have comparable solutions (Intel’s E-core only server CPUs and AMD’s Zen 5C Epic server CPUs).
Except that because AMD uses TSMC, AMD had better fab technology than Intel. AMD did a double (or triple) whammy to Intel. Intel has been behind since like 2015 or so. I swear Intel was on 14nm forever.
AMD used Global Foundries when they started this strategy, and technologically they beat intel because of it, despite an inferior production process.
So as I wrote, the strategy is solid.Intel has been behind since like 2015
This is just stupid, AMD was way behind Intel until the arrival of Ryzen in March 2017, and Epyc came later.
When AMD was later released from the GloFo agreement, they could stave off Intel with better production process from TSMC too.2015 is probably around the time Intel lost their production process advantage, but they were not way behind yet at that point.
Awesome results from sanctions. I wish EU would follow suit.
Let’s not forget that the game is also about consistency. If only one of these CPUs per wafer actually achieves these specs and the rest has lots of dead cores, this is still impressive but makes scaling production very hard.
