Originally Posted by EniGma1987
Not initially. It was a very leaky pos node that took a year of tweaking to become any good. Towards the end of AMD's use of it though, it was an extremely good node
The 32nm PDSOI node was on-par with Intel's 32nm Bulk in perf/power in 2009, no GloFo tweaks. The node was built to be low leak, do to the massive buried oxide oven effect.
If anything if Bulldozer was leaky, it was AMD's fault not GlobalFoundries. For example, modern tapeouts of Cu-32 at GlobalFoundries are all low power/low TDP designs.
As far as timeline goes Bulldozer/Piledriver were with the old tools from Agena/Deneb/Thuban days. If Husky in Llano was bad, then Bulldozer/Piledriver were going to be bad.
While, Steamroller/Excavator were up to date following Bobcat, Jaguar, and later Zen design methodologies. If Jaguar in Kabini was good, then Steamroller/Excavator were going to be good.
It also helps that 28-nm didn't have box-derivied self heating. As well as nearly halving RVT/HVT transistor leakage, while LVT only nearly increased in 25% leak.
The current consensus in external research the problems derived from AMD focusing on idle leakage with Bulldozer. Which the design philosophy caused higher leakage in the active state. But, hey check out our cool fully gated modules, and low idle voltage. However, it does make sense as RAM bandwidth isn't increasing particularly fast. So, while waiting for the next load, might as well idle to reduce energy.