Question UCIe, universal chiplets interconnect

[Thibsie]

See Anandtech's homepage...

Universal Chiplet Interconnect Express (UCIe) Announced: Setting Standards For The Chiplet Ecosystem

www.anandtech.com

Just huge IMO

 

[Saylick]

Lol @ the lack of Nvidia (at the moment). We'll see if they end up joining, but the fact that the consortium promotes an open standard must have scared off Jensen.

 

[nicalandia]

Impressive and Huge

 

[moinmoin]

Lol @ the lack of Nvidia (at the moment).

Never going to happen with Nvidia. You could as well ask Apple to participate there (and I tend to give that a higher chance than Jensen stopping his dream of complete vertical integration).

 

[igor_kavinski]

It would be expensive and stupid of Nvidia not to join. Sure, it suits them great to do their own thing as long as their proprietary tech is leaps and bounds above the rest. But they will suffer when all these high value companies come up with something truly revolutionary and cost effective that beats Nvidia's solutions. At that point, they will have to join.

 

[Saylick]

It would be expensive and stupid of Nvidia not to join. Sure, it suits them great to do their own thing as long as their proprietary tech is leaps and bounds above the rest. But they will suffer when all these high value companies come up with something truly revolutionary and cost effective that beats Nvidia's solutions. At that point, they will have to join.

I agree. Capital expenditures are only going to go up as we hit diminishing returns on scaling. It may have been previously possible for one company to develop proprietary standards for every aspect of their business, but good luck doing that now when development costs are increasing rapidly and it's forcing the whole industry to collaborate to share those costs. For example, NVLink is likely going to suffer a long and slow death at the hands of CXL. CUDA isn't currently suffering the same fate due to it's market dominance, but if the biggest cloud providers / HPC clients banded together and pushed for an open standard, CUDA would effectively have a shelf life.

Never going to happen with Nvidia. You could as well ask Apple to participate there (and I tend to give that a higher chance than Jensen stopping his dream of complete vertical integration).

Apple doesn't need to participate because they are sufficiently vertically integrated. Nvidia isn't, so I think there should be no reason why Nvidia should not join as many standards/consortiums as possible. At some point Nvidia products have to interface with something not Nvidia, and if they refuse to play ball or support the latest industry standards for those non-Nvidia products they will be left behind.