AMD is already designing products on the 14nm node

[HEXUS]

Likely making use of GlobalFoundries 14nm FinFET.

Read more.

[zaph0d]

This could well be the Catchup Leap that AMD have been missing since Intel launched the Core 2 series of chips that delivered superior Work per Cycle to Power ratios.
AMD has been having to use the old Intel P4 tricks of increased Power Consumtion and Clock Cycles to remain performance competitive for too long now.
Good Luck AMD, hope this works and get's you back in the game again

[abaxas]

Owww.Hopefully we'll get a kabini style 8 core 2ghz chip for 50 quid.

Bring it on.

[scaryjim]

Hmmm, 14nm isn't suddenly going to make AMD's CPU cores perform better. There still seems to be a hard limit of 4Ghz - 5GHz for running a processor with acceptable TDP, and AMD would need a lot more than that to make a 4-core APU competitive with an i5.

Where this will really make a difference is in power efficiency (reducing process size tends to reduce power draw), and in GPU and parallel compute, because they'll be able to put more shaders/cores onto the same size chip. So great news for mobile/HPC/semi-custom/graphics. For mainstream CPU ... not so much.

[abaxas]

PCs are (and have been for the last 4-5 years) a race to the bottom.

Per core even the most modern core i3/i5/i7 is only 20-40% faster than a core 2.

Saying that, almost all intel's offerings are greater than the market wants or needs. They only exist due to near monopoly.

[yeeeeman]

They need to:
- redesign their high performance core and have it on the market as soon as possible (Zen)
- skip a process node and catch up with others. For milking customers it's good to prolong the life of a node as much as possible, but you can't do it if you have Intel as your competitor. (14nm)
- I would've skipped Carrizo alltogether and focus on Zen.
- entice OEMs to buy your products. It's not possible with kaveri, but beema and mullins are super good chips. Why aren't they as present as Bay Trail? Hmm, you answer that.

[scaryjim]

Per core even the most modern core i3/i5/i7 is only 20-40% faster than a core 2.

i7 4790k PiFast: 15.8s
QX9770 PiFast: 29.1s

That's just a little a bit more than 40% faster on *the* single core throughput test.

Even an A10-6800k matches a QX9650 on this one. It may not feel like it generation on generation, but Intel's latest Core i processors are massively faster than Core 2, even on single threaded tasks. Of course, the flip side of that is that AMD's latest core still isn't ahead of a 3GHz Core 2. They've made huge strides in core count and graphics, but in single threaded performance things haven't changed for AMD since the Phenom II days.

[sykobee]

They also need to disappear Intel from the planet...

[zaph0d]

Per core even the most modern core i3/i5/i7 is only 20-40% faster than a core 2.

i7 4790k PiFast: 15.8s
QX9770 PiFast: 29.1s

That's just a little a bit more than 40% faster on *the* single core throughput test.

Bear in mind the Clockspeed difference between those chips. 3.2GHz for the QX9700 and 4.4GHz (Single thread turbo) for the i7 4790k. That's a performance difference of 37.5% right there purely on clockspeed, when applied to the i7's PiFast time that's 21.725s (purely as a percentage) that leaves a 33.4% difference in per clock performance improvement over 6 years between launches, that's an average of 5% improvent per year.

[DanceswithUnix]

Why aren't they as present as Bay Trail? Hmm, you answer that.

Because Intel can afford to give their chips away, and what can AMD do in the face of that?

[shaithis]

This could well be the Catchup Leap that AMD have been missing since Intel launched the Core 2 series of chips that delivered superior Work per Cycle to Power ratios.

Lets hope but as others have said, shrinking on it's own isn't going to get huge gains, you need to make better use of your transistors.

Don't forget Sandy Bridge was 32nm, just like the current AM3+ CPUs....and the AMD CPUs aren't giving up real-estate to an iGPU.

[scaryjim]

Bear in mind the Clockspeed difference between those chips.

Had I been replying to a comment about IPC that'd be a fair comment. But I was replying to a comment about per core performance, and part of the enhancement in that has been the ability to run retail chips at stock speeds of 4GHz+. It's all part of the package.

And let's not forget the 4790k does that with 35% less heat dissipated. It's got (by your figures) 33% more IPC than the fastest Core 2, 37% higher clock speed than the fastest Core 2, and generates 35% less heat. I'd love to know when (outside of the big P4/Core 2 step) we've ever seen those kind of consistent upticks in performance and downticks in power consumption....

[abaxas]

Ermm, maybe you forget the pentium II/III ?

Huge leaps forward in not just performance but performance per watt.

From 450mhz to 1.3ghz in 3 years inside the same power envelope. Even more if your include the pentium II, 233mhz to 1.3ghz inside 5 years.

[abaxas]

Then what about the original pentium?

2-3x perf per watt in 5 years.

[scaryjim]

tdp of the original pentiums (1993) were in the mid-teens; tdps of the P III (1999) were in the high 20s/low 30s (and some into the 40s). The pure performance increase may have been greater (and that's down as much to the exponential increases in clock speed as the IPC increases), but it was at the expense of higher power draw/heat dissipation.

[DanceswithUnix]

It used to be that on each new process node, the wires were shorter and thinner and so had less capacitance. That made the wires easier to charge up and discharge, so clock speeds went up by up to double (specially if you go back to the 386/486 era). OK the power usage crept up, but then it comfortably could so what was the harm?

Now on a shrink the ability to drive the wires gets worse by about as much as the capacitance improves. Net result is no real automatic improvement in clock speeds just from shrinking.

However, what you do get from a shrink is more transistors to play with. That gets you deeper buffers, bigger caches, more registers, more entries in your predictor tables. Stuff like floating point performance is pretty much just down to transistor budget. That gets us a small increment in performance.

On top of that going to a new node is really expensive, so the automatic cost decrease that we used to get from shrinking to a new node isn't there any more or at least is diminished.

So small increment in performance for no real improvement in cost. There are rumours that there won't be any graphics cards made on 20nm, because it just doesn't help. Let's hope that AMD jumping from 28 to 14 is enough of a jump to make a real difference, and not just for laptops & tablets.