Reply 60 of 66, by Jasin Natael
- Rank
- Oldbie
cyclone3d wrote on 2021-06-30, 03:01:Bulldozer/Piledriver was just a bad design overall.
Even before the release it was super easy to tell that the design was not good.
AMD tried to use automation and simulation in their design process instead of using experienced CPU design engineers and it bit them in the gonads pretty hard.
It wasn't really that it was a bad design, per se. At least not the general idea. But it was poorly executed.
The shared FPU method as well as the cache sharing was a bad way of implementing what was essentially supposed to be alternative to SMT.
It was in it's own way a unique way of parallel processing that AMD was betting their future on, but it really didn't play out that way.
i do think it was ahead of it's time though.
Clearly higher thread count dependent applications were on the horizon.
But Piledriver was such a let down in IPC and overall performance that AMD didn't have the market share to cause any developers to push for optimizing any code for it.
I like to think of it as similar situation to the 3DNow! fiasco of days gone by...
Just look at the impact that Ryzen made when it hit the scene. It was impressive enough that it even caused Microsoft to make changes to Windows thread scheduling. Because the chips were selling, of course because they were good.
In short, AMD pulled a Icarus and flew to close to the sun. It COULD have paid off, but it was immature and not that well executed so it failed.
But it wasn't nearly so bad as people make it out to be, it certainly wasn't a flawed as other architectures I could name (cough....Netburst...cough)
