Best possible scenario for post-Bulldozer AMD x86 CPUs?

[witeken]

Eh, 16 FinFET has smaller SRAM cell sizes, so it is denser.

Since when is a 15% higher density in line with Moore's Law? (20/16)² = 1.6

 

[SAAA]

OK you want data about Intel upcoming 14nm so you all stop derailing the thread?

Here: http://electroiq.com/blog/2014/01/intel-vs-tsmc-an-update/

http://electroiq.com/wp-content/uploads/2014/01/Chart-7.png

So it's 45nm metal pitch and 0.062 to 0.052um^2 of area, or in % numbers about 12 to 35% denser than TSMC 16nm node.
Possibly all the delays they had were because EUV went bad and to get the highest density they used 193i litho to the limits, also this increase over 0.09um^2 cell for 22nm is proven by the Broadwell die shots vs the Haswell ones (not in the link, Google it).

Of course the slide is two years old so something has changed, but still is the most relevant info about 14nm node. Wait two months to get all the updated infos at IDF.

 

[Hans de Vries]

So you're saying it's Intel's fault.

But it's not.

Intel's process engineers just developed the best processes they could
with the equipment they had available. They did a good job and did so
in all openness. You'll never hear me complain about that.

show me TSMC's or Samsung's 28nm transistor with a 28nm half-pitch and gate.

20nm first metal layer pitch: 64nm.

Not even close to 20nm,

This whole discussion wouldn't be necessary if people would just go one
little step further and looked at the actual meaningful numbers instead of
just the "node-naming-number'. The above remarks show you didn't.

So what does the (old) 2005 ITRS roadmap require as a first metal layer
pitch for a "16nm process" ? Well, look at the graph below: 2x32 = 64nm

Incidentally, TSMC's 16nm FinFet process has a first metal layer pitch of
64nm.... So based on that they are entitled to call it a 16nm process.

 

[SlowSpyder]

I do hope Excavator is good. Good CPU performance per watt, a strong GPU, and improved memory bandwidth. Maybe they can sell them for better prices than what they are able to sell their APU's for now.

Anyone know if AMD's plans are to continue to build from Excavator or go a different direction altogether for future APU's?

 

[Arachnotronic]

OK you want data about Intel upcoming 14nm so you all stop derailing the thread?

Here: http://electroiq.com/blog/2014/01/intel-vs-tsmc-an-update/

http://electroiq.com/wp-content/uploads/2014/01/Chart-7.png

So it's 45nm metal pitch and 0.062 to 0.052um^2 of area, or in % numbers about 12 to 35% denser than TSMC 16nm node.
Possibly all the delays they had were because EUV went bad and to get the highest density they used 193i litho to the limits, also this increase over 0.09um^2 cell for 22nm is proven by the Broadwell die shots vs the Haswell ones (not in the link, Google it).

Of course the slide is two years old so something has changed, but still is the most relevant info about 14nm node. Wait two months to get all the updated infos at IDF.

This isn't official information. Just guesswork.

 

[Qwertilot]

There's some sort of basically new design coming in the medium term - link to a thread about that the start of this thread before it all veered off widely. They can see what they've got to do so hopefully they'll get the efficiency rather better than the current stuff.

The iGPU stuff seems a bit sadly stuck until they get stacked memory

 

[Exophase]

K7 was more or less bought from NexGen, and K8 built on top of that.

It's K6 that was more or less bought from NexGen. K7 was legitimately developed in-house, albeit with a lot of ex-DEC Alpha engineers.

And contrary to what mrmt said, I don't think K6 was crap at all, it was a well designed uarch even compared with P6. The big hindrance was the socket which prevented them from having on-slot L2 cache, and the lack of optimized x87 with the later expectation that people would use 3DNow! (which was much better than even a high-quality x87 implementation)

 

[dbcoopernz]

I suspect that AMD will mostly focus on ARM based designs, and will find a nice niche in ARM based SOCs for servers. Then Samsung will acquire them.

 

[SOFTengCOMPelec]

I don't think K6 was crap at all

I long time ago, I really wanted a Barton (K6-3), it was really good at that time.

I think they were reasonably competitive, until Bulldozer, which unfortunately seemed to be a bit of a mistake.

But as regards overclocking, Intel was usually King. I've overclocked many Intels over the years (much less so now), but almost never overclocked AMDs, as they usually just either would take no overclocking, or so little that it was not worth the risk of instability.

I use to be looking forward to excavator, but it is not so clear to me, what the improved/FX is going to be like.

If only it could have more than 8 cores.

The current FXs use too much power for my liking, but an FX6300 is bearable, and the 65 Watt Apus (rather than the 100 Watt) ones are not too bad.

 

[ViRGE]

This thread has run its course. Not that it stayed even remotely on topic for the bulk of it.:|
-ViRGE