Globalfoundries files plans for second mammoth plant in Malta (450 mm)

[Idontcare]

How they got around that is by making a massive 550 mm² 32-nm CPU for 2014 called Warsaw.

At 16 cores and 550mm^2, that beast is going to have to be clocked at around 2GHz just to keep the power-consumption reasonable at the socket.

Still though, 16 cores, people will put that to good use.

 

[NTMBK]

At 16 cores and 550mm^2, that beast is going to have to be clocked at around 2GHz just to keep the power-consumption reasonable at the socket.

Still though, 16 cores, people will put that to good use.

There are already 16 core Piledriver Opterons, actually- one of them is clocked as high as 2.8GHz, with a 3.5GHz turbo. The Opteron 6386 SE. They are dual-die packages, though, not a monolithic die as Seronx is claiming (probably incorrectly).

 

[Ajay]

There are already 16 core Piledriver Opterons, actually- one of them is clocked as high as 2.8GHz, with a 3.5GHz turbo. The Opteron 6386 SE. They are dual-die packages, though, not a monolithic die as Seronx is claiming (probably incorrectly).

If they are not monolithic, why the name change?

 

[inf64]

If they are not monolithic, why the name change?

Because they are probably tweaked die (a la Richland). There is no conspiracy here, these are tweaked PD based MCM devices that work @ 2.8-3Ghz. So a minor perf. bump and probably somewhat better perf./watt.

 

[Ajay]

Because they are probably tweaked die (a la Richland). There is no conspiracy here, these are tweaked PD based MCM devices that work @ 2.8-3Ghz. So a minor perf. bump and probably somewhat better perf./watt.

Wasn't thinking about a conspiracy, but thanks for your input. It's really too bad AMD couldn't field something at least a step up from this in cores/die or perf/watt. I fear their remaining tiny share of the server market is going to be trashed by IVB-EP.

 

[inf64]

Well they are doing what makes sense from $ POV. When/if they have something worth mentioning (like SR or EX for servers) they will put it on the roadmap. Next year is going to be 16T PD max. in HPC segment. I guess they are aligning/adjusting their server roadmap with Haswell-E (if they intend to even compete in HPC next two years). The rest is low power/high core density push by AMD in new expanding server segment. Hopefully their ARM based chips will get some good market traction.

 

[Exophase]

Making a new die for a 16-core server part would be totally crazy. The revenue they could get from that would be too low to justify the new design, before you even consider how much worse yields would be vs the current solution.

We don't even know if the current topology could support a 16-core arrangement.. IIRC Thuban was pushing the crossbar design with 6 cores, and BD is similar except interfacing modules instead of cores. So going > 6 modules could be a problem. I do recall there being plans for 5 module parts but I think those were scrapped.

That's not to say they couldn't do something that's more like the current MCM but laid out internally but again.. why?

 

[inf64]

Yup, these are just refreshed PD MCM devices. Maybe RCM and some "Richland" tweaks.