[Anandtech] GP107 fabbed on 14nm Samsung

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,398
8,567
126
No it's not. To tape out for a new process by a different manufacturer will cost a lot more then just doing a smaller version on the same well understood process from the manufacturer they have always used. They also got a slower card at the end of it which lowers the price you can sell it for. Financially it doesn't make sense to use Samsung.
TSMC is in demand from everyone. samsung isn't. dollars to donuts samsung is doing these at near its own cost, while TSMC is practically naming its price. so despite the additional upfront costs that nvidia will expend, they're almost certain to make up for it in a cheaper per wafer and even cheaper per chip samsung process.


edit: i see this has been covered
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com

HurleyBird

Platinum Member
Apr 22, 2003
2,811
1,544
136
I don't know whether or not it's cheaper in the long run for Nvidia, but my assumption is that, on a purely two dimensional level the additional engineering time, up front cost, and (likely) lower performance* outweigh the benefit of a somewhat lower cost from Samsung.

On a three dimensional level, Nvidia might have been looking to hedge their bets in case one supplier fails to deliver, or guard against supply constraints. A big part of the decision is probably to strengthen their bargaining position against TSMC by playing one foundry the other. There are plenty of good reasons to not want to be at the mercy of a single supplier.

*Assuming the Samsung fabbed chips do perform lower as expected, it's still up in the air how much is down to the process on Samsung's side vs. optimisation for the process on Nvidia's.
 
Last edited:
  • Like
Reactions: Headfoot