By: Dummond D. Slow (mental.delete@this.protozoa.us), November 16, 2020 4:40 pm
Room: Moderated Discussions
Dummond D. Slow (mental.delete@this.protozoa.us) on November 16, 2020 4:29 pm wrote:
> Doug S (foo.delete@this.bar.bar) on November 16, 2020 12:32 pm wrote:
> > Dummond D. Slow (mental.delete@this.protozoa.us) on November 16, 2020 11:54 am wrote:
> > > Doug S (foo.delete@this.bar.bar) on November 16, 2020 10:24 am wrote:
> > > >
> > > > (...) and no one who has observed Intel's process problems over the past half decade could possibly suggest
> > > > they will improve performance faster than Apple will over the next few years. That doesn't even get into
> > > > the M1 being less expensive for them or the benefits of controlling their own roadmap instead of trying
> > > > to cajole Intel into making what they want and hoping they are able to deliver.
> > >
> > > Hmm, that is actually an intersting detail - if you think about it, Apple is as state-of-the-art
> > > in the process node as it gets, being like the first CPU/SoC vendor to jump on new TSMC tech.
> > >
> > > Intel on the other hannd (LOL)...
> > >
> > > But that hypothetically, if Intel suddenly became an agressive TSMC partner, that could actually
> > > give them opportunity to make gigantic advancements to their current stuff, because the would
> > > suddenly get that "state of the art process node" pillow between their seat and butt.
> > >
> > > So if that happened (and no, I don't think it will - or not soon), they could actually make bigger
> > > relative progress than Apple, because Apple already had that pillow put under themselves and their
> > > advancements would be just uarch compared to uarch + 'TSMC go brrr' on Intel's side.
> > >
> > > After that, the field would be levelled and we would just be watching
> > > a fight in who does better uarch advancements, again.
> > >
> > > Again, this is just thought experiments. It's not clear to me that Intel will really go TSMC. And if they
> > > will, we don't know if they could adopt new process nodes as quickly as Apple and not as slowly as AMD.
> >
> >
> > Well sure, but if Intel ended up doing that their decision would have been made far too late for
> > it to influence Apple's choice. I mean, I'm assuming they have not made such a design today and
> > are still working on getting their 7nm process ready and researching their 5nm process.
> >
> > If they went this route there would be a SERIOUS shortage
> > of Intel's new TSMC fabbed "high end" CPUs for years,
> > since TSMC can't just wave a magic wand and create enough
> > capacity to serve Intel nor could they expect getting
> > their process up and running in Intel's fabs (one would
> > assume they'd acquire those as a part of such a deal)
> > would not be a massive undertaking that probably takes longer than either party would wish.
> >
> > Apple can't worry about what Intel might do in five years, and presumably
> > would have (well justified) trust in their design team that they can compete
> > quite well with Intel even if both were on equal footing process wise.
>
> I didn't raise this scenario wrt to Apple'S decision. It's irrelevant because it's super theoretical.
> What I meant is that being behind in process tech is actually a hidden
> potentil for leap ahead IF you somehow manage to close that gap.
>
> Similar to how if you had memory controller in chipset and not in CPU, you kinda had that as a ace to
> pull once in the future and get a one-time bonus. AMD used up the card in Athlon 64 but Intel kept it
> till Nehalem where it suddenly gave a nice boost they had till then be holding back on. Of course, that
> makes it sound as if it is important when you play it which is a fallacy (instead you want all cards
> being always played ASAP), but you know what I want to say - it's a fruit hanging on the tree.
Maybe I could put it yet differently:
When one company is held back some huge setback (Intel's process, AMD's bad decisions with Bulldozer uarch, Apple's being stuck on badly aging PowerPC platform before 2005), you can't just take that as a definitive good thing for their competitors.
You also have to look out for the possibility that that one setback might suddenly disappear and the company could come back massively.
For example, while AMD had the Bulldozer problem AND at the same time process technology problem (Intel was the only player with FinFETs), they were forced to work on very advanced power and clock management. And in 2017, Zen and GLoFo/TSMC FinFETs suddenly appeared, AMD was suddenly pushed way way above its previous positions and that power management thing from before suddenly became an asset that Intel suffered from. Their boost was not as advanced and that allowed AMD to get closer to them than their core itself and their process tech would allow at that point normally.
Similarly, Intel's process shit is a huge setback for them, but what if the shackles are suddenly removed - they might be able to suddenly jump crazy high. Never rest on laurels when your competition seems to be in trouble.
> Doug S (foo.delete@this.bar.bar) on November 16, 2020 12:32 pm wrote:
> > Dummond D. Slow (mental.delete@this.protozoa.us) on November 16, 2020 11:54 am wrote:
> > > Doug S (foo.delete@this.bar.bar) on November 16, 2020 10:24 am wrote:
> > > >
> > > > (...) and no one who has observed Intel's process problems over the past half decade could possibly suggest
> > > > they will improve performance faster than Apple will over the next few years. That doesn't even get into
> > > > the M1 being less expensive for them or the benefits of controlling their own roadmap instead of trying
> > > > to cajole Intel into making what they want and hoping they are able to deliver.
> > >
> > > Hmm, that is actually an intersting detail - if you think about it, Apple is as state-of-the-art
> > > in the process node as it gets, being like the first CPU/SoC vendor to jump on new TSMC tech.
> > >
> > > Intel on the other hannd (LOL)...
> > >
> > > But that hypothetically, if Intel suddenly became an agressive TSMC partner, that could actually
> > > give them opportunity to make gigantic advancements to their current stuff, because the would
> > > suddenly get that "state of the art process node" pillow between their seat and butt.
> > >
> > > So if that happened (and no, I don't think it will - or not soon), they could actually make bigger
> > > relative progress than Apple, because Apple already had that pillow put under themselves and their
> > > advancements would be just uarch compared to uarch + 'TSMC go brrr' on Intel's side.
> > >
> > > After that, the field would be levelled and we would just be watching
> > > a fight in who does better uarch advancements, again.
> > >
> > > Again, this is just thought experiments. It's not clear to me that Intel will really go TSMC. And if they
> > > will, we don't know if they could adopt new process nodes as quickly as Apple and not as slowly as AMD.
> >
> >
> > Well sure, but if Intel ended up doing that their decision would have been made far too late for
> > it to influence Apple's choice. I mean, I'm assuming they have not made such a design today and
> > are still working on getting their 7nm process ready and researching their 5nm process.
> >
> > If they went this route there would be a SERIOUS shortage
> > of Intel's new TSMC fabbed "high end" CPUs for years,
> > since TSMC can't just wave a magic wand and create enough
> > capacity to serve Intel nor could they expect getting
> > their process up and running in Intel's fabs (one would
> > assume they'd acquire those as a part of such a deal)
> > would not be a massive undertaking that probably takes longer than either party would wish.
> >
> > Apple can't worry about what Intel might do in five years, and presumably
> > would have (well justified) trust in their design team that they can compete
> > quite well with Intel even if both were on equal footing process wise.
>
> I didn't raise this scenario wrt to Apple'S decision. It's irrelevant because it's super theoretical.
> What I meant is that being behind in process tech is actually a hidden
> potentil for leap ahead IF you somehow manage to close that gap.
>
> Similar to how if you had memory controller in chipset and not in CPU, you kinda had that as a ace to
> pull once in the future and get a one-time bonus. AMD used up the card in Athlon 64 but Intel kept it
> till Nehalem where it suddenly gave a nice boost they had till then be holding back on. Of course, that
> makes it sound as if it is important when you play it which is a fallacy (instead you want all cards
> being always played ASAP), but you know what I want to say - it's a fruit hanging on the tree.
Maybe I could put it yet differently:
When one company is held back some huge setback (Intel's process, AMD's bad decisions with Bulldozer uarch, Apple's being stuck on badly aging PowerPC platform before 2005), you can't just take that as a definitive good thing for their competitors.
You also have to look out for the possibility that that one setback might suddenly disappear and the company could come back massively.
For example, while AMD had the Bulldozer problem AND at the same time process technology problem (Intel was the only player with FinFETs), they were forced to work on very advanced power and clock management. And in 2017, Zen and GLoFo/TSMC FinFETs suddenly appeared, AMD was suddenly pushed way way above its previous positions and that power management thing from before suddenly became an asset that Intel suffered from. Their boost was not as advanced and that allowed AMD to get closer to them than their core itself and their process tech would allow at that point normally.
Similarly, Intel's process shit is a huge setback for them, but what if the shackles are suddenly removed - they might be able to suddenly jump crazy high. Never rest on laurels when your competition seems to be in trouble.