Apple Plans to Ditch Intel and Use Custom Mac Chips Starting in 2020

[cmaier]

Someone in this thread mentioned earlier that the strategy of DEC Alpha64 was the most excellent and it could emulate 3 different instruction sets simultaneously all within the same processor.

That was done in software
[doublepost=1523041193][/doublepost]

that's vliw, not risc

Was referring to the code morphing
[doublepost=1523041413][/doublepost]

I have. I remember them fairly well. They had a pretty good mobile processor for their day.

But I am not talking about code morphing. ARM and Intel both use μop translation, and ARMv8 has more registers than x86-64, so it seems like building a real-time hybrid might be a realistic possibility. There was rumor of a Taiwanese company that had a hybrid (32-bit ARM/64-bit Intel) a few years back, so it probably can be done effectively.

Ever hear of exponential? Ever wonder why there was a Texas team?

I wouldn’t call it translation, though. In x86 designs what is going on with the microcode rom is mostly dealing with situations where the instruction lengths are greater than the minimum. You still have to deal with all the crazy addressing modes and such. Microcode is mainly a way to trap all the complexity in the instruction decoder and not have to run lots of wires everywhere. Attempts by others to build x86 around a real risc core haven’t gone swimmingly. I was there.

 

[techwhiz]

That was done in software
[doublepost=1523041193][/doublepost]
Was referring to the code morphing
[doublepost=1523041413][/doublepost]


Ever hear of exponential? Ever wonder why there was a Texas team?

I wouldn’t call it translation, though. In x86 designs what is going on with the microcode rom is mostly dealing with situations where the instruction lengths are greater than the minimum. You still have to deal with all the crazy addressing modes and such. Microcode is mainly a way to trap all the complexity in the instruction decoder and not have to run lots of wires everywhere. Attempts by others to build x86 around a real risc core haven’t gone swimmingly. I was there.

People haven't been paying attention. If they had they would know we already talked about the SEG teams and one of them being the old Exponential team, which became Intrinsity, which was acquired by Apple in 2010 or so.
One of the techniques they used to make things run faster was something we did at Amdahl which was multiphase clocks.

Anyway.
They'll do some type of dynamic compilation and cache the results or something similar during the transition if they go to ARM. If they do the transition, they'll figure out how to make it work. It's just that legacy software will take a hit because emulation, translation (dynamic at runtime or static/cache) will never be as fast as a native instruction VM. Now if the performance is high enough, will the app be fast enough? Probably because most apps like Word or Excel spend more time waiting for you. It's the compute intensive stuff that will need to be ported to work well. Light Room. Photoshop, Premiere, Final Cut, etc....


There isn't any doubt that ARM can be made to rival Intel performance, so people calm down.
If they transition, we will get over the bump.

I still don't trust Bloomberg's accuracy.

 

[e-coli]

Try researching Apple' poor heat design in laptops, iMac, and mini. Apple does cripple the CPU when it starts to heat up. Apple does it in a way that does not show up in standardized performance tests. If you ever try to use Apple hardware for serious computing, it becomes obvious.

Thermal throttling is literally built into all CPUs and GPUs. All computers do this, otherwise they would incinerate themselves. How is this a black eye for Apple?

 

[cmaier]

People haven't been paying attention. If they had they would know we already talked about the SEG teams and one of them being the old Exponential team, which became Intrinsity, which was acquired by Apple in 2010 or so.
One of the techniques they used to make things run faster was something we did at Amdahl which was multiphase clocks.

Anyway.
They'll do some type of dynamic compilation and cache the results or something similar during the transition if they go to ARM. If they do the transition, they'll figure out how to make it work. It's just that legacy software will take a hit because emulation, translation (dynamic at runtime or static/cache) will never be as fast as a native instruction VM. Now if the performance is high enough, will the app be fast enough? Probably because most apps like Word or Excel spend more time waiting for you. It's the compute intensive stuff that will need to be ported to work well. Light Room. Photoshop, Premiere, Final Cut, etc....


There isn't any doubt that ARM can be made to rival Intel performance, so people calm down.
If they transition, we will get over the bump.

I still don't trust Bloomberg's accuracy.

Were you doing bipolar at Amdahl? We did that at exponential (and it was my Phd project before that). The intrinsity overlapping clock stuff with their 1-of-n stuff was weird, though.

 

[Jon L]

This image is making the rounds. Milan, Rome, and Naples are all AMD codenames for Zen processors. Maybe just a coincidence, but....

 

[twoehr]

Looks like I got off of Quicken for Windows just in time, or maybe too soon?

 

[nutjob]

Many of them have designed x86 before.

But x86 requires a license from intel, so unless intel is agreeable to all this I don’t buy it.

Apple could buy AMD with their pocket change, and it would fly through regulatory approval. Or they could enter a partnership with them where AMD implements and complements their designs. Neither would require Intel's approval.

 

[cmaier]

Apple could buy AMD with their pocket change, and it would fly through regulatory approval. Or they could enter a partnership with them where AMD implements and complements their designs. Neither would require Intel's approval.

Not true. There is almost certainly a change in control provision in the intel cross license. Such provisions are standard. In all likelihood if AMD is acquired it loses the license. Moreover such licenses frequently contain “have made” provisions, though this one may not.

 

[nutjob]

Sure you can run it on both, but do you honestly think future ARM chips can even beat today's Intel midrange chips, in 2020? If it can't come close to, or beat Intel performance, then it's literally like taking steps backwards.

If they had went with AMD on the other hand, I'm pretty confident this thread wouldn't have reached 59 pages, with many upset users that are ready to jump ship.

Watt for watt, ARM already has beaten Intel. ARM implementations focus on low power which is the only reason there haven't been more performance oriented implementations. Apple's ARM cpus are a demonstration that you can do both and do much better than Intel.
[doublepost=1523053246][/doublepost]

Not true. There is almost certainly a change in control provision in the intel cross license. Such provisions are standard. In all likelihood if AMD is acquired it loses the license. Moreover such licenses frequently contain “have made” provisions, though this one may not.

That provision would likely be invalid under antitrust rules, but even if it exists and holds, there is nothing stopping AMD entering a partnership with Apple that gives them a similar outcome.

 

[Music Ambulance]

I saw this coming when they released the benchmarks of the iPhone X Bionic chip.

I see their custom chips going into the consumer products...iMacs and Macbooks. But the pro lineup will probably still use intel chips. I can't see pro customers going for anything else.

It’s tricky though for some pros. As a graphic designer / photo retoucher I’m finding intels MacBook Pro and iMac chips absolutely underwhelming. But the Xeon chips are complete overkill for me. An Apple chip sounds intriguing, but Will I have software to run? Will I be able to jump into terminal to tweak things.

 

[cmaier]

Watt for watt, ARM already has beaten Intel. ARM implementations focus on low power which is the only reason there haven't been more performance oriented implementations. Apple's ARM cpus are a demonstration that you can do both and do much better than Intel.
[doublepost=1523053246][/doublepost]

That provision would likely be invalid under antitrust rules, but even if it exists and holds, there is nothing stopping AMD entering a partnership with Apple that gives them a similar outcome.

Where’d you get your law degree? Since when does antitrust law require licensing of patents? After all, patents grant a state-sanctioned time-limited monopoly.

As I also mentioned, there may well be a “have made” clause that prevents it, though that’s less likely.

 

[Jon L]

Not true. There is almost certainly a change in control provision in the intel cross license. Such provisions are standard. In all likelihood if AMD is acquired it loses the license. Moreover such licenses frequently contain “have made” provisions, though this one may not.

It is true that the agreement terminates upon change of control. So AMD loses its x86 license. However, Intel would likewise loses rights to AMD64. Both parties almost certainly would negotiate a new agreement.

 

[cram501]

It is true that the agreement terminates upon change of control. So AMD loses its x86 license. However, Intel would likewise loses rights to AMD64. Both parties almost certainly would negotiate a new agreement.

I would be surprised if Intel lost a license due to the action of another party (AMD selling themselves). Intel would have no interest in extending that license. Didn't they unsuccessfully try to invalidate it long time ago?

 

[techwhiz]

Were you doing bipolar at Amdahl? We did that at exponential (and it was my Phd project before that). The intrinsity overlapping clock stuff with their 1-of-n stuff was weird, though.

Yes, I started out doing ECL then bipolar at Amdahl.
The 1-n stuff was strange.
At Amdahl we used a custom clocking with the concept of a base clock with lots of phases of early and late. All flops were master/slave type and we had design rules with the maximum number of logic stages. You used the earliest version of clock required to meet your setup/hold. This allowed borrowing for a later stage in the pipeline. We were doing some very strange stuff to meet 5ns machine cycle times in the early 90's.

If you have thirty versions of the clock, then clock skew can be your friend.

 

[Jon L]

I would be surprised if Intel lost a license due to the action of another party (AMD selling themselves). Intel would have no interest in extending that license. Didn't they unsuccessfully try to invalidate it long time ago?

AMD gets access to x86 and Intel gets access to x86-64 through the cross-licensing agreement. Upon change of control, that agreement terminates, per the agreement. Intel losing access to x86-64 would be a bit of a problem because of the existing 64 bit ecosystem that has been built around AMD64 (remember Itanium? didn't work out so well)

https://www.kitguru.net/components/...nge-of-control-terminates-agreement-for-both/

 

[cmaier]

AMD gets access to x86 and Intel gets access to x86-64 through the cross-licensing agreement. Upon change of control, that agreement terminates, per the agreement. Intel losing access to x86-64 would be a bit of a problem because of the existing 64 bit ecosystem that has been built around AMD64 (remember Itanium? didn't work out so well)

https://www.kitguru.net/components/...nge-of-control-terminates-agreement-for-both/

Ok. Though it’s important to note that Amd is pretty free with the -86-64 license. They wanted to make it the standard.

Also, I designed parts of AMD64 so I remember Itanium
[doublepost=1523063710][/doublepost]

Ok

Yes, I started out doing ECL then bipolar at Amdahl.
The 1-n stuff was strange.
At Amdahl we used a custom clocking with the concept of a base clock with lots of phases of early and late. All flops were master/slave type and we had design rules with the maximum number of logic stages. You used the earliest version of clock required to meet your setup/hold. This allowed borrowing for a later stage in the pipeline. We were doing some very strange stuff to meet 5ns machine cycle times in the early 90's.

If you have thirty versions of the clock, then clock skew can be your friend.

Yeah, we did the same things with clocks and skew in some of our early designs (including the bipolar ones). Having worked both ways I prefer balancing the clocks and using edge triggered flops. There were so many opportunities for bugs caused by clock borrowing/hold times and manufacturing variations.

Though four phase non-overlapping clocks was kind of fun.

 

[nutjob]

Where’d you get your law degree? Since when does antitrust law require licensing of patents? After all, patents grant a state-sanctioned time-limited monopoly.

As I also mentioned, there may well be a “have made” clause that prevents it, though that’s less likely.

Did you get your law degree from Trump University? AMD has a flourishing semi-custom business which is exactly what Apple would need. I guess you think Intel is just holding back on suing MS and Sony for using AMD's CPUs, for now. You're confused about the direction of patent licensing - Apple need only licence theirs to AMD and have them produce the CPU they need. They can work together to design it.

If Intel tried to withdraw licensing based on who is on AMD's share register, then yeah, an antitrust suit would ensue. And although there is a change of control clause, it runs both ways, so Intel would lose access to the 64 bit arch patents. Renegotiating those in the context of AMD being acquired or recapitalised would indeed have antitrust overtones and Intel would be in a tight spot.

https://www.kitguru.net/components/...nge-of-control-terminates-agreement-for-both/

 

[cmaier]

Did you get your law degree from Trump University? AMD has a flourishing semi-custom business which is exactly what Apple would need. I guess you think Intel is just holding back on suing MS and Sony for using AMD's CPUs, for now. You're confused about the direction of patent licensing - Apple need only licence theirs to AMD and have them produce the CPU they need. They can work together to design it.

If Intel tried to withdraw licensing based on who is on AMD's share register, then yeah, an antitrust suit would ensue. And although there is a change of control clause, it runs both ways, so Intel would lose access to the 64 bit arch patents. Renegotiating those in the context of AMD being acquired or recapitalised would indeed have antitrust overtones and Intel would be in a tight spot.

https://www.kitguru.net/components/...nge-of-control-terminates-agreement-for-both/

I have no idea what the point of your first paragraph is. It’s irrelevant to anything I’ve said. Apple licensing AMD does no good if the resulting processor is x86 and there is a “have made” clause in the licensing agreement. MS and Sony may have their own license agreeements with intel. (Ms almost certainly does, or they’d have problems with x86-on-Arm emulation - intel has threatened to sue others for that).

Intel is not currently a monopoly, so where the antitrust concern? Doesn’t ARM dominate the cpu market? This isn’t 1990.

And I got my law degree and engineering degrees at accredited universities, but nice job getting a trump dig in.

 

[nutjob]

I have no idea what the point of your first paragraph is. It’s irrelevant to anything I’ve said. Apple licensing AMD does no good if the resulting processor is x86 and there is a “have made” clause in the licensing agreement. MS and Sony may have their own license agreeements with intel. (Ms almost certainly does, or they’d have problems with x86-on-Arm emulation - intel has threatened to sue others for that).

Intel is not currently a monopoly, so where the antitrust concern? Doesn’t ARM dominate the cpu market? This isn’t 1990.

And I got my law degree and engineering degrees at accredited universities, but nice job getting a trump dig in.

You didn't read my previous post (which you replied to). Apple has several alternative approaches. I was describing the simplest and least legally complicated route: Apple licences designs and patents to AMD and they produce CPUs based on that. Your "have made" clause issues are speculative, and you seem to have thrown that in in desperation. AMD wouldn't have a semi-custom business if every customer had to seek arch licences from Intel. Anyway, Apple and AMD could, for example, probably maneuver around such a clause by making the cost of licensing Apple tech (from AMD) prohibitive while offering it to all comers.

But then there are other possible approaches. Apple could buy AMD and then threaten to disrupt or destroy Intel's business if they don't extend the cross licence post purchase. Pretty sure Intel customers wouldn't like AMD64 to disappear. Intel would believe that Apple pockets the AMD graphics business and CPU patents and designs and discards the rest, and would quickly blink. Alternatively Apple may only need to licence the AMD64 arch from AMD depending on the scope of Intel patents and could discard x86 compatibility. Its full control of its own platform would make that possible. There are many ways to skin an uncooperative cat, you just need some imagination.

You should probably skim the relevant legislation becuase the antitrust laws have nothing to do with being an actual monopoly and everything to do with the abuse of market power to fend off competition. Something that Intel has already been found guilty of and has tacitly agreed it has done, though settlements. Regulators would be pleased if an aggressive competitor like Apple positioned itself against Intel and conversely very skeptical of Intel moves to stop AMD from strengthening its position. Also, ARM don't make x86 CPUs, so they're irrelevant.

Congratulations on your degrees, but here we win arguments with the weight of our words and ideas, not the bits of paper on our walls.

 

[mtneer]

Well how about Apple just buy Intel?

 

[Regime2008]

Well how about Apple just buy Intel?

Why would you want to gimp processors for the majority of desktop/laptop users? That sounds like such a bad idea.

 

[nt5672]

Thermal throttling is literally built into all CPUs and GPUs. All computers do this, otherwise they would incinerate themselves. How is this a black eye for Apple?

Because Apple's thermal design is so bad, that the CPU throttles more than necessary and yet they tout full speed specs.

 

[Regime2008]

Because Apple's thermal design is so bad, that the CPU throttles more than necessary and yet they tout full speed specs.

Apple choose form over functionality. Instead of giving a computer ample breathing room, proper ventilation and good fans, they just throw all the parts into a small container and call it a day. iMacs, Mac Minis, etc. all show signs of thermal throttling, and some much worse than others.

 

[gavroche]

Of course not, we even discuss that in the forums all the time. But its the only data I can find.

I'm definitely open to changing my mind. Since you hold claims to high standards, I'm curious what source you found that is the basis of your assertion.

What source do i base my assertion on? I base my assertion on the same thing i would base my assertion that the majority of people who drive cars don't subscribe to Car & Driver Magazine, that the majority of people that own homes aren't members of DIY Home Improvement Forum, or that the majority of people that fly in airplanes from time to time couldn't tell you what an Aileron is and does.... and that thing is called intuition. We people on MacRumors are people for whom computers is more than just a tool to do a job. It is a hobby, something we like.

Do you think that the average person who owns a 2017 Ford Mustang could tell you what next year's model is rumored to feature? How many horsepower its motor likely will have? There are forums where people passionate about that hang out, and they know the answers to those questions. And in those forums, you might find such talk as "If Ford decides to put that particular engine in it, i'm done.... i'm getting a XYZ. But i really doubt you would argue that most people who drive the car could give a rats butt what is rumored.

p.s. Why are you saying that i hold claims to high standards? Was it my statement that its "all about compromise, buy what meets your needs."? I know, thats a pretty out of line statement. Or my simple question asking if you thought that people on Macrumors represent the average user. I guess that was pretty out of line too.

 

[AppleScruff1]

What source do i base my assertion on? I base my assertion on the same thing i would base my assertion that the majority of people who drive cars don't subscribe to Car & Driver Magazine, that the majority of people that own homes aren't members of DIY Home Improvement Forum, or that the majority of people that fly in airplanes from time to time couldn't tell you what an Aileron is and does.... and that thing is called intuition. We people on MacRumors are people for whom computers is more than just a tool to do a job. It is a hobby, something we like.

Do you think that the average person who owns a 2017 Ford Mustang could tell you what next year's model is rumored to feature? How many horsepower its motor likely will have? There are forums where people passionate about that hang out, and they know the answers to those questions. And in those forums, you might find such talk as "If Ford decides to put that particular engine in it, i'm done.... i'm getting a XYZ. But i really doubt you would argue that most people who drive the car could give a rats butt what is rumored.

p.s. Why are you saying that i hold claims to high standards? Was it my statement that its "all about compromise, buy what meets your needs."? I know, thats a pretty out of line statement. Or my simple question asking if you thought that people on Macrumors represent the average user. I guess that was pretty out of line too.

You make some very good points, although I have never seen people on a car forum bragging about how much profit a company makes.