Puff

Whether it is dual-core is not decided by its relative performance level to its predecessor, but whether it has a monolithic dual-core performance in certain workloads is. No matter how someone says, a Bulldozer module does not fit the definition of SMT, at least from the core pipeline perspective.

There is no certain definition for "core", but there is one for SMT. So Orochi being quad-core, eight-thread fails this way. But does it really perform as a eight-core? It depends on implementation and workloads then, and undoubtedly Bulldozer doesn't have a catchy single-thread performance to facilitate its niche multi-core scaling.

[ 本帖最後由 Puff 於 2013-1-9 16:42 編輯 ]

Puff

引用:

原帖由 dom 於 2013-1-8 17:31 發表


The problem is It doesnt perform !
Just like GCN Vs Kepler

GCN spec look good on paper, but in reality it need "cheat" driver to cheat performance

Simply because you just keep your eyes on the marketing figures, but not digging deep enough. Tahiti's fixed function does not have the level of GK104's. This is the fact. So cheat? No. It performs how it should be.

And it has the compute capability as AMD claimed. AMD does not give guidance on SI's graphics performance before launch either, only the benefits to compute.

[ 本帖最後由 Puff 於 2013-1-9 16:03 編輯 ]

Puff

引用:

原帖由 dom 於 2013-1-9 17:07 發表


Well thats why its call FAILDOZER

And there are eight Faildozer cores.