WCCFAMD Zen Architecture Could Feature in APUs in 2016, ASUS a Key Player

[DrMrLordX]

F. Design; EV9 "Tarantula"
M. Design; T2/T2+ "Niagara 2"

Huh, looks like you've answered the question there. Niagara 2. Still, Oracle did dump a lot of talent when they took over Sun, so it is conceivable that AMD poached one of the UltraSPARC team members that worked on T1/T2.

Buldozzer bad performance is not CMT fault.

No, in fact it isn't. Process/transistor budget and corner-cutting were more to blame, if memory serves me correctly. The decision to use CMT under those circumstances may not have been the best in the world. In defense of BD/PD, PD seems to finally be coming into its own, albeit far too late.

Still, CMT has a pretty lousy track record between Sun's and AMD's implementations. Third time's the charm, eh?

 

[NostaSeronx]

In my opinion, the actual fate of Cluster-based Multithreading will be determined by Excavator(2014) and Excavator+(2015).

Excavator => 28-nm unnamed/misnomer node.
Excavator+ => 20-nm unnamed/misnomer node.

Excavator will be using modern techniques rather than old outdated techniques. There will be some custom work like Jaguar, but majority of it will be synthesized cells. Hand schematic designs like Bulldozer/Piledriver/Steamroller are limited in scaling. That and we will be seeing a shrink of the magnitude of UNB+GPU(Trinity/Richland) to UNB+GPU(Kaveri). For the CPU side of the die, which is a ~1.85x shrink.

Steamroller to Excavator is equal to K8 to Greyhound.

 

[positivedoppler]

In my opinion, the actual fate of Cluster-based Multithreading will be determined by Excavator(2014) and Excavator+(2015).

Excavator => 28-nm unnamed/misnomer node.
Excavator+ => 20-nm unnamed/misnomer node.

Excavator will be using modern techniques rather than old outdated techniques. There will be some custom work like Jaguar, but majority of it will be synthesized cells. Hand schematic designs like Bulldozer/Piledriver/Steamroller are limited in scaling. That and we will be seeing a shrink of the magnitude of UNB+GPU(Trinity/Richland) to UNB+GPU(Kaveri). For the CPU side of the die, which is a ~1.85x shrink.

Steamroller to Excavator is equal to K8 to Greyhound.

Where did you hear about Excavator+

 

[NostaSeronx]

Where did you hear about Excavator+

Excavator on a 20 nanometer node was never cancelled.

Excavator (GF28A) - Carrizo
Excavator (GF20AN) - Basilisk

After that it was sub-connecting the dots. I'm essentially pulling the plus out of Puma(28-nm) to Puma+(20-nm).

Just remember that AMD is never on time.

Carrizo is going onto existing platforms with DDR3 and PCI Express 3.0.
Basilisk is going onto new platforms with DDR4 and PCI Express 4.0. <== "Zen" platform uses "Excavator" first.

 

[positivedoppler]

Excavator on a 20 nanometer node was never cancelled.

Excavator (GF28A) - Carrizo
Excavator (GF20AN) - Basilisk

After that it was sub-connecting the dots. I'm essentially pulling the plus out of Puma(28-nm) to Puma+(20-nm).

Just remember that AMD is never on time.

Carrizo is going onto existing platforms with DDR3 and PCI Express 3.0.
Basilisk is going onto new platforms with DDR4 and PCI Express 4.0. <== "Zen" platform uses "Excavator" first.

Carrizo is 2015. Zen is 2016. Seems almost a r&d waste to release carrizo 20nm in between within a year

 

[NostaSeronx]

Carrizo is 2015. Zen is 2016. Seems almost a r&d waste to release carrizo 20nm in between within a year

It's not the first time for this to happen.

http://i.imgur.com/ZZvt5o8.gif
Greyhound(65-nm) -> Greyhound+(45-nm) -> Bulldozer(45-nm)

Sometimes, that is the most hopeful path.

http://i.imgur.com/SxuOzOb.jpg
Greyhound(65-nm) -> Greyhound+(45-nm) -> Greyhound+(45-nm) -> Greyhound+(45-nm) -> Bulldozer(32-nm)

I'm simply telling those to expect;
Excavator(90-nm M1 Pitch) -> Excavator+(64-nm M1 Pitch) -> Zen(64-nm M1 Pitch).

Definitely, not to expect;
Excavator(90-nm M1 Pitch) -> Excavator+(64-nm M1 Pitch) -> Excavator+(64-nm M1 Pitch) -> Excavator+(64-nm M1 Pitch) -> Zen(48-nm M1 Pitch).