Intel: 10nm Is Not Dead (It's Getting Better)

[jimmysmitty]

You seem preoccupied with desktop/laptop, but AMD's prize is actually cloud.

lol, no, those are some of Intel's least profitable market segments. The big money is 2-4S Xeons and FPGAs.

Okay, then it sounds like we agree.

And my point was that cost is actually a big driver, for that market. AMD's multi-die strategy gives it an unassailable cost advantage over Intel.

Just like Intels Core 2 Quads with multiple dies vs a single die. TBH I am surprised they haven't gone back to that, well except the performance across dies issues but that shouldn't be an issue with their new interconnects.

 

[bit_user]

And my point was that cost is actually a big driver, for that market. AMD's multi-die strategy gives it an unassailable cost advantage over Intel.

Just like Intels Core 2 Quads with multiple dies vs a single die. TBH I am surprised they haven't gone back to that, well except the performance across dies issues but that shouldn't be an issue with their new interconnects.

I think it's coming. They were just caught off-guard. I think Intel is the type of company where it's difficult to buck the status quo, until there's a need to react to new threats that arise from the outside.

Probably their next server socket will feature multi-die CPUs like EPYC (except using Intel's EMIB instead of AMD's interposer solution). That's just my hunch.

 

[InvalidError]

And my point was that cost is actually a big driver, for that market. AMD's multi-die strategy gives it an unassailable cost advantage over Intel.

Multi-die comes with some disadvantages too, mainly all of the challenges that come with basically stuffing quad-socket NUMA in a single socket where you have to worry about optimizing algorithms to compensate for the different distances between cores and memory pools even on a 1S setup.

Great if you can partition data sets to fit within memory channels attached to the die running the algorithm focusing on that subset, not so great if the algorithm's dataset requires access to a larger dataset than can be attached to each die or requires significant overlap.

A bigger single die allows you to organize your problem in fewer larger chunks, which reduces overhead and complexity.

 

[jasonelmore]

Charlie is going to be sued if he's not careful

 

[jimmysmitty]

And my point was that cost is actually a big driver, for that market. AMD's multi-die strategy gives it an unassailable cost advantage over Intel.

Just like Intels Core 2 Quads with multiple dies vs a single die. TBH I am surprised they haven't gone back to that, well except the performance across dies issues but that shouldn't be an issue with their new interconnects.

I think it's coming. They were just caught off-guard. I think Intel is the type of company where it's difficult to buck the status quo, until there's a need to react to new threats that arise from the outside.

Probably their next server socket will feature multi-die CPUs like EPYC (except using Intel's EMIB instead of AMD's interposer solution). That's just my hunch.

I am not sure Intel would really care to not do that. After all AMD made fun of Intels M2M system for Core 2 Quad stating a monolithic die is the only route, which they did with K10. Its kind of surreal to see the exact opposite yet similar effects. Intel has a big die that seems to be power hungry while AMD has a multi die solution that seems fine.

 

[InvalidError]

Its kind of surreal to see the exact opposite yet similar effects. Intel has a big die that seems to be power hungry while AMD has a multi die solution that seems fine.

AMD had to go with a modest size die to mitigate costs and risks of having both a new architecture, new process and betting the barn on it. A one-size-fits-all approach makes sense for that. Intel may be behind on 10nm but it has a very mature 14LP++++ process and architecture so its risk level on larger 14nm die isn't as problematic.

This may change at 10nm in whatever form Intel manages to bring it to market.

 

[urbanman2004]

Fvck Intel w/ their arrogance, price gouging and unethical business practices, monopolizing the industry w/ underhanded gatekeeping (remember VIA). After the whole 9900K fiasco I'm done w/ them.