Intel 'tick-tock' finally dead; introduces POA 'Process-Architecture-Optimization'

[tenks]

Looks like Intel is officially labeling and recognizing the new 3 element cycle. Tick-Tock was money for so many years ;(

"Intel’s Tick-Tock strategy has been the bedrock of their microprocessor dominance of the last decade. Throughout the tenure, every other year Intel would upgrade their fabrication plants to be able to produce processors with a smaller feature set, improving die area, power consumption, and slight optimizations of the microarchitecture, and in the years between the upgrades would launch a new set of processors based on a wholly new (sometimes paradigm shifting) microarchitecture for large performance upgrades. However, due to the difficulty of implementing a ‘tick’, the ever decreasing process node size and complexity therein, as reported previously with 14nm and the introduction of Kaby Lake, Intel’s latest filing would suggest that 10nm will follow a similar pattern as 14nm by introducing a third stage to the cadence."

Source

 

[MongGrel]

Hickory-Dickory-Dock process, possibly ?

 

[tortillasoup]

I would have preferred if Intel had sold improved processors that allowed users to retain their motherboards like some of the processors back in the day. Instead, Intel always found a way to upgrade the processors so that they were never backward compatible with older boards. It was rare to see the user being able to continue to use a motherboard through two iterations of a processor design. Maybe we will get that now with their new cadence.

 

[MongGrel]

I would have preferred if Intel had sold improved processors that allowed users to retain their motherboards like some of the processors back in the day. Instead, Intel always found a way to upgrade the processors so that they were never backward compatible with older boards. It was rare to see the user being able to continue to use a motherboard through two iterations of a processor design. Maybe we will get that now with their new cadence.

Most mother boards would need to evolve to just continue going forward, of course there were things in the past.

Still using a used OD'd X5680 in the main and a X5650 in the HTPC, I guess they might have steered away from that.

But I seriously doubt being backward compatible really a thing for Intel, resales are not one of their worries I would think.

 

[ShintaiDK]

Backwards compatibility is usually more of a hinderance than good.

3 on a node is the new standard. GPUs already did it on 28nm. Well essentially 4.

 

[NTMBK]

Backwards compatibility is usually more of a hinderance than good.

Yup, the outdated platform probably didn't help AMD's FX processors. (On top of the obvious performance/heat issues.)

 

[exar333]

Tick-tock made sense as we moved very quickly from 90nm to where we sit now. It makes perfect sense to reevaluate your strategy when you face changing physics. I think this makes a lot of sense based of what we have seen recently from Broadwell and Skylake. Intel needs to have a solid schedule in place for it's partners so they know when refresh cycles are planned, and if those are ideal for their products or not. Especially with their focus on getting all products on the same node, at a closer timeframe (Atom not trailing 2-3 nodes back, for example).

 

[Eug]

So, how much improvement do people think there will be going from Kaby Lake (3rd generation 14 nm) to Cannonlake (1st generation 10 nm)?

 

[cytg111]

So, how much improvement do people think there will be going from Kaby Lake (3rd generation 14 nm) to Cannonlake (1st generation 10 nm)?

Perf/watt? A whole lot.

Peak singlethreaded performance? Nothing.

 

[showb1z]

I think the more interesting question is how much performance will improve for the "optimization" architecture. I was under the impression that Kaby was just going to be IGP improvements and not much else.

 

[Eug]

Perf/watt? A whole lot.

Definitely in theory, but we do have any evidence for this right now? Just wondering. And I wonder what "A whole lot" means.

Peak singlethreaded performance? Nothing.

Well, considering I'm looking at Core M and the ultraportable market, I'm thinking maybe it would make a difference. Perhaps it will be easier to design machines that will maintain peak performance for longer with Cannonlake in those ultrathin fanless (or tiny fanned) laptops.

And given that there are rumours that Kaby Lake Y and Kaby Lake U won't necessarily get HDCP 2.2 and HDMI 2.0, I may just have to wait anyway.

EDIT:

I think the more interesting question is how much performance will improve for the "optimization" architecture. I was under the impression that Kaby was just going to be IGP improvements and not much else.

Ah I see. Yes, I'm interested in the IGP improvements - HDCP/HDMI - but if they aren't applied to all Kaby Lake chips across the board, and there isn't much CPU performance/Watt improvement, then it's kind of an irrelevant upgrade over Skylake for me. See my comments above.

 

[moonbogg]

If Zen ends up not sucking, Intel will be right back to Tick, Tock! That's because the clock will be ticking regarding their sales opportunities if people are buying Zen instead. So that tock better come quicker than #$@! and none of this milking 3rd stage nonsense.

 

[Qwertilot]

Physics is a much, much bigger obstacle than any competition might be at this stage!

 

[poofyhairguy]

I think the more interesting question is how much performance will improve for the "optimization" architecture.

Agreed. Maybe the GPUs will be the real difference.

 

[frozentundra123456]

Physics is a much, much bigger obstacle than any competition might be at this stage!

That is true, but intel could unlock (sanctioned) more lines, and add more cores to the mainstream, or make edram more common. I know more cores is rumored for Cannonlake, but I have seen no official confirmation. And maybe a specially binned, better TIM/solder product like Devil's Canyon. But yea, I think AMD does have an advantage now (relative) when it comes to catching up or at least being quite competitive. Intel 14nm was difficult, and 10 nm is probably going to be the same. So AMD has a better chance of catching up (relatively) now that intel has pretty much hit the wall, at least as far as absolute performance is concerned. But *if* Zen cpu performance is good and they can get HBM onto igpus at a reasonable cost, intel could be in for some serious competition. But who knows, it is too soon to say.

 

[AtenRa]

Higher node R&D for each new node plus lower wafer volume due to decrease in PC (Both mobile and Desktop) sales and Intel cannot maintain a 2 year cadence anymore.
The reason they tried entering the Tablet/Phone market was this, they needed the added volume to sustain a 2 year cadence. Since they failed in to the Tablet/Mobile they have to go to a 3 year cadence and i believe we may come to a 4 year cadence after the 14nm era, much like GPUs.

 

[IntelUser2000]

So AMD has a better chance of catching up (relatively) now that intel has pretty much hit the wall, at least as far as absolute performance is concerned.

I don't believe this presents as an opportunity for AMD to take the lead, since everyone is struggling.

But if the gap between AMD and Intel was a meter wide, maybe with post-Zen generation it'll be 10cm.

Ultimately it seems that architectures are converging to a spot. Even more radical talk like saying CPU and GPU will converge isn't being said anymore. In fact, in their DX12 presentation Microsoft even stated that gap between CPUs and GPUs will remain.

That may mean that rather than one vendor having everything, in the future you would choose a chip based on your needs.

I think the more interesting question is how much performance will improve for the "optimization" architecture. I was under the impression that Kaby was just going to be IGP improvements and not much else.

"build upon the foundations of the Skylake micro-architecture but with key performance enhancements."

PLENTY to expect that we'll see more than "spurious" changes like HDMI advancements. We don't know what went wrong with Skylake but we know that they struggled with yields on 14nm, and likely impacted Skylake's availability. Yields are related to how a product will turn out. PCWatch has speculated on the possibility that Intel knew about 14nm issues and Skylake had to be architected to account for that. By "Optimization" time the process will be more stable and just by the virtue of having a year more the architects can further squeeze more out of the chip.

 

[frozentundra123456]

I never said AMD would take the lead, or even catch up. It will just be easier for them to be at least competitive, since intel is advancing very slowly now. At least they aren't shooting at a moving target. (or at least a fast moving one).

 

[IntelUser2000]

I never said AMD would take the lead, or even catch up.

No you did not. But posts are directed towards EVERYONE. Sorry if it sounded otherwise.

I don't believe that a company with meagre resources will suddenly catch up because their far bigger(and so far more nimble) competitor struggles massively, and add to that them not being the only one to struggle.

 

[ninaholic37]

So, how much improvement do people think there will be going from Kaby Lake (3rd generation 14 nm) to Cannonlake (1st generation 10 nm)?

I'm guessing less than Sandy Bridge -> Ivy Bridge (32nm -> 22nm) and Haswell -> Broadwell (22nm -> 14nm) because... physics.

 

[Hulk]

Since in the old "tick-tock" model "process" always involved some pretty significant optimization I'm wondering if in the new model "process" will be more of a strict die shrink and very few if any architecture optimization.

On one hand this would seem to reduce the complexity of the die shrink which we all know has been getting more and more difficult and will continue to do so.

But on the other hand it will be kind of strange to have a new part that is nearly identical from a raw performance point of view (not per watt) to it's predecessor.

Intel would also have the issue of the new process being "exposed." And by that I mean power savings could be accurately quantified as stemming solely from the new process.

Seems as though Intel is getting caught between a rock and a hard space here. They have to slow down the process shrinks because they are so hard to get the yields going and there probably aren't too many left. And on the architecture side of the coin the low hanging IPC improvement fruit was picked long ago. So really how much IPC improvement is left. I'm very interested to see what Kaby Lake will bring over Skylake.

I can imagine people running around at Intel like chickens without head, scrambling here and there. Kind of like when they realized they took a wrong turn with NetBurst and AMD caught them with their pants down. But this time they have competition coming from all sides!

 

[poofyhairguy]

One thing I don't understand is why physics matters to value. I mean I get that is how IPC or perf/watt or whatever improves, but I don't get why that is the only limit to CPU value that matters.

People are hesitant to upgrade because Sandy->Skylake is like 25%, and the next gen is something like 5% again or whatever, but it doesn't have to be that way. Intel could bump CPU power 50% tomorrow by throwing another two cores in consumer i7 products and we would all celebrate. I would upgrade past Sandy finally even if I don't need it just to finally bask in a decent leap in performance. But screw paying $200+ for a decent mobo that can support six cores today for a chip Directx12 makes us need less.

I look at an Intel CPU die and I see so much space dedicated to a GPU that many of us don't use, so it's not like it couldn't get done if that was the priority. Quite frankly ANY desktop chip with a better than a decent integrated GPU is a waste because general business machines just need Windows and 2D apps to work and gamers will put a real card it. There is no point of better integrated GPUs in our desktop chips but year after year that is the only real improvement that we are sold.

I feel like at this point Intel is just coasting in the desktop consumer CPU space, using us as part of an economy of scale go make better devices for portable and smal form factor computers that CAN need a better integrated GPU. I know the real answer is they lack competition, but what I don't understand is why people pretend to get excited about 5% here, 10% there or perf/watt on a big desktop with a 600w PSU. Just admit they stopped catering to us, hug your current CPU, and spend your money on a better GPU or monitor or whatever. I have stopped caring what Intel does in socketed CPUs until we get hex cores that run on z level mobos. The G3258 is the only fun chip Intel has released in half a decade.

 

[Eug]

I have cared far more about the feature set and low power utilization than raw performance for the last decade. Once I have two cores, I'm happy.

I don't generally buy discrete GPUs these days and in fact don't generally even buy upgradable desktops anymore. And I don't even generally buy separate monitors.

Much of the market is going that direction, in the form of laptops, particularly ultrabooks.

 

[tortillasoup]

Intel is definitely coasting and it's AMD's fault. The death of AMD will mean practically no progress on the CPU front though with the progress ARM has made, that could be Intel's new competition.

 

[VirtualLarry]

The G3258 is the only fun chip Intel has released in half a decade.

LOL. Pretty-much true. Sig-worthy, also.