News GlobalFoundries puts 7nm on indefinite hold

Something tells me that the prices of Zen 2 in 2019 won't be as competitive against Intel as they were for Zen 1.

 

7nm/10nm Self Aligned Quad Patterning:
- Intel: late, but with some small dies shipping for the last few months with sufficient volume for use in shipping products
- TSMC: No dies shipping yet, supposedly small dies coming soon (though any production is likely to be eaten by the giant fruit)

7nm EUV:
- Globalfoundries: WE OUT
- Samsung: 7LPE already dropped internally, "2nd gen 7nm" may turn into '5nm' like their 2nd gen 10nm became 8LPP, and internal scuttlebut is 2019/2020 for actual production.

 

Basically, we're struggling up a increasingly steep hill when it comes to "conventional" transistor shrinkage and if we don't quickly find a seriously alternative (Quantum, photonic, graphene, organic), we might end up stalling completely?

 

The stalling kind of happened around a decade ago (when gate oxide thickness hit ~1nm and voltages stabilised in the 1V range, and everything ended up noodling around in the max 4-5GHz range). And since 28nm, price/transistor has been going up rather than down.

 

Considering tsmc have 80% of there total 7nm capacity contractual bound to Apple where does that leave everyone else.

AMD will have to now prioritise CPUs or GPUs and we all know it will be cpu.

Where does that leave the gpu market. Cost implications could be huge going forward

Going Samsung assumes Samsung can make a big chip 7nm so far the reality is not. And re designing your chips to work on Samsung is not as easy as 1/2/3

They all use different tools

It’s why intel renting fab space never really got off he ground as all there tools are in house.

 

AMD has doubled down on what they told anandtech back in spring, there will be no relevant AMD consumer GPU any time soon:

https://www.techpowerup.com/247006/amd-7nm-vega-by-december-not-a-die-shrink-of-vega-10

 

It seems sub 14nm (probably even sub 28nm) is of questionable benefit to the likes of us, going smaller hasn't so far resulted in huge performance gains, it certainly gets you more performance per watt so benefits mobile and enterprises customers but if you only care about performance it seems larger nodes are more suited to higher clock speeds.

 

As former AMD CEO Jerry Sanders III put it:
"Real men have fabs"