It seems like the next GCN will be, like its forebearers, manufactured at TSMC, c.f. http://wccftech.com/amd-arctic-islands-next-generation-graphics-16-nm-finfet-tsmc/. This all the while AMD uses GloFo for their APUs.
I assume there would be some gains for AMD if they were able to target a single node pr generation. Instead, GCN has to target both GloFo and TSMC. If they only targeted GloFo, possible benefits would be:
- Fewer issues fulfilling their wafer supply agreement with GloFo
- Not having to port GCN from TSMC to GloFo
- Being able to optimize GCN more towards a single node
- More experience with the targeted node
Obviously, these and other pros do not outweigh the cons, as AMD is again using both GloFo and TSMC. However, what are these cons; why are they insisting on using TSMC for their GPUs?
I assume there would be some gains for AMD if they were able to target a single node pr generation. Instead, GCN has to target both GloFo and TSMC. If they only targeted GloFo, possible benefits would be:
- Fewer issues fulfilling their wafer supply agreement with GloFo
- Not having to port GCN from TSMC to GloFo
- Being able to optimize GCN more towards a single node
- More experience with the targeted node
Obviously, these and other pros do not outweigh the cons, as AMD is again using both GloFo and TSMC. However, what are these cons; why are they insisting on using TSMC for their GPUs?