News SMT4?!

[NTMBK]

Apparently we're going to be seeing SMT4... from Intel.

Intel Royal Core Project to improve x86 architecture efficency - VideoCardz.com

Intel Royal Core Project rumored to improve X86 architecture to better compete with Apple products Moore’s Law is Dead has a video covering a wide spectrum of Intel’s future consumer and data center products. According to the YouTuber, who has a very good track record on Intel leaks, the...

videocardz.com

At some point I might work up the energy to watch the 20 minute video, which will convey things that I could otherwise read in a 3 minute article. Sigh.

 

[igor_kavinski]

At some point I might work up the energy to watch the 20 minute video, which will convey things that I could otherwise read in a 3 minute article. Sigh.

In case you haven't watched it, save time by reading the transcript:

so i haven't exactly been coy about hinting that i know of a lot of insane intel architectures coming out over the next five years and that i feel competition is really going to heat up by the end of 2023 but i have been coy about telling everyone everything and i'm still going to be a little timid about that because frankly well there's a lot of key details i've been told by some of my best sources just to literally not tell people for competitive reasons for intel and simply because people could get in trouble but well i'm guessing a lot of you watching this are aware of that mysterious reddit post and jim's recent aero lake leak which let me confirm as far as i'm aware pretty much everything jim said in that aero lake leak is correct based on what i know in fact early this year i was told that it might not even come to desktop and it was really meant to take on mac so i really have no reason to doubt what he said it's actually because it's meant to take on mac i really didn't pay much attention to that architecture but i have been paying attention to other ones and hinting at kind of revolutionary upgrades that are coming and i've actually been saying a lot of quotes that are pretty similar to what that mysterious reddit poster said i suspect this reddit poster got a hold of a few key details and is an avid listener of jim and me on broken silicon and well decided to try to put it together and make some big post for some reason using a name similar to mine but he didn't know some real code names so he definitely knew something and yeah i mean i wish i said all the code names first i guess but i'm not sure how useful saying a bunch of code names is right for example when jim leaked his arrow lake information he tried to tell a little bit more than just the code name like tell you that intel really is building some of their stuff at tsmc and they really are seeing apple as a giant threat much like i've said before and honestly not just about intel but also about amd but clearly something's going on where it's becoming open season when it comes to confirming upcoming products from intel and so i've decided that i am going to leak a few things and at the very least although i still insist on not saying everything because well i just want to be careful with my sources i'd want to at least tell the story of what intel's doing because when i'm seeing some of these leaks coming out now about intel i think i think there's a greater narrative that people should know that tells you what intel's ultimate design goals are to maybe take the crown back at everything from amd but first i want to leak details about a product that isn't probably going to take a crown back but is coming soon enough that i think you should all know about it and that is emerald rapids which yes i am aware i am not the first to talk about it nor will all the details i'm about to confirm from my end be the first they're said now at this point everyone's getting really excited especially internally at intel for alder lake and that's because well let's just be honest this is the biggest increase in performance at well everything multi-threaded and single threaded for intel four years this is going to be a massive increase over rocket lake and tiger lake and i'm actually though kind of more excited for raptor lake just because of what it signifies no it's not gonna like double multi-threading performance over alder lake like alder lake should over tiger lake at the same power usage it's i don't want to get into what that actually means but it is adding eight more cores again as i leaked and bringing higher ipc and clock speeds than alder lake as i leaked a couple months ago and so for me raptor lake is the most more exciting because it shows alder lake isn't just like a one-off and then more 10 increases intel is looking to increase performance by like 30 or more year over year every year from now on and so that's really how you should think of emerald rapids you know not a doubling of performance or some massive increase like sapphire rapids really will be over ice lake x but it is an increase it's going to 64 cores at 350 watts on an intel 7 node that should be very mature by the time it launches in 2023 now it is planned to launch in quarter one and i actually think intel is going to start hitting their milestones when they say they will more often moving forward under pat gelzinger but i still have to say it's the goal it's not a hundred percent confirmed and it does bring faster ddr5 higher ipc it will have higher clock speeds too honestly when you look at that ddr5 speed you look at the higher ipc than what's in golden cove i would say this is clearly something like raptor cove or a similar architecture although i haven't had anyone specifically say it is and i still wouldn't be surprised if until came up with a new cove name for to replace what they're calling raptor code to also be used on emerald rapids and after that though all there is to say is that it is again just really an iteration of sapphire rapids design it's same node same four quartiles i do not know how many cores are in each tile but it's obviously at least 16. it would not surprise me if maybe it goes up to something like you know i don't know like 18 cores so that the total could technically be 72 cores but if i'm told the node is very mature i don't know they may have just been planning to use the full core count and it is just 16 cores per tile just can't say for sure but what i can also say is that my sources are talking about granite rapids which its placement on the road map is moving around now and diamond rapids and diamond rapids is the big one guys this is the one where i think intel could catch up to amd in server performance but it's probably not till 2025 and we'll get some hints of what could be in it when i talk about lunar lake later also there's sierra forest that i know other people have mentioned like adored i just want to say that sierra forest isn't some version of another thing as far as i'm aware that sierra forest is kind of a radical side project for different customers in that if you look at granite rapids coming next to sierra forest you look at arrow lake coming next to meteor lake stop thinking of everything as just linearly replacing everything before it that's how i knew that reddit post had so many things wrong in it it's it's because he acts like everything is just linearly replacing everything bergamo doesn't replace genoa either once we move past emerald rapids once we start getting out of zen 4 from amd's side there's going to start being multiple architectures per segment for different customers things are going to get complicated but they're also going to get really exciting and the last thing i will say is people aren't paying enough attention to these side architectures the stuff intel's working on is crazy but that's for another video and i'm about to talk about how intel will compete with threadripper next year and an early look at lunar lake but first an ad from a sponsor oh shoot what was that password god i forgot it reecey can you bring me my netflix password here puppy come here peacock what the heck no one needs to log into peacock ever having trouble managing your passwords is your dog categorizing system not working very well anymore well then you should use roboform password manager roboform is a password manager that allows you to easily store your passwords and log into websites with a single click that includes streaming devices credit cards and anything and they can be linked across multiple devices and browsers so you could share passwords for example between a chrome browser and your iphone and it even lets you share bookmarks between these devices as well which for me is personally an incredible feature when you consider how many electronics of different manufacturers operating systems and browsers i use every day it's super easy to get started you can have it up and running within minutes and they provided a special deal for my viewers you that allows you to get 30 off your roboform everywhere subscription by using the link on screen or in the description and if you do use this it's only 16 a year use my link so they know moore's law's dead sent you this really helps the channel and of course if you need to manage multiple passwords and bookmarks across multiple devices and browsers well you should be using roboform today what is for this video though is confirming that while i am excited for what intel has coming in some segments next year in 2023 and really in all segments 2024 and onwards i do not see emerald rapids is something that's really going to catch up to amd so they're still going to at least be behind in server in 2023. you see emerald rapids is yeah better than sapphire rapids probably like 30 percent better but genoa genoa is not going to be 30 better than milan genoa is going to be probably almost twice as good 50 more cores with massively higher ipc and it should come out before emerald rapids so that's not going to look very good and in fact bergamo will be out at about the same time anyways i've already leaked that threadripper is coming out at the end of this year with zen 3 and threadripper pro is coming out next year although i have been told recently that threadripper pro isn't coming out until july 2022 and this source is very sure of that so that's when threadripper pro comes out that's when threadripper comes out and i just don't see really intel being competitive in hdt until 2023 and they will only be competitive if they can price emerald rapids hedt chips really cheaply but things for smaller products that go into laptops and desktops things could be different different sooner and those products well they kind of spell out if you know what they are what intel's trying to do and when i see some of these mysterious leaks just saying little tidbits here and there and you know doing things like saying oh arrow lakes after meteor lake it replaces it and so on and so forth it misses the greater picture like intel's moving away from the idea of single architectures that linearly replace the ones before them and i've been leaking little details about meteor lake where i first confirmed redwood cove and lunar lake where i actually told people what the ipc roughly would be and people for some reason didn't care i don't know and so on and so forth for a while it's time to spell out what's going on here and it starts with something that i've been told by multiple people for a while now about alder lake and that's that it is a first step into intel's multi-tile architecture end goal now a lot of people would see that and go oh so that big little's the end go or oh so they're just gonna you know put one thing on a tile one thing on this top put them together that's the end goal it's actually kind of more ingenious and beneficial to how intel operates than that you see intel is a very large company with design teams making ip and parts of socs all over the world and these need to be combined and then tested and validated and well it's it's a lot and it's not strictly right now completely split up in a way where they could really pull like they're pulling different lego pieces from all over the world and put them together for very specific products for very specific customers that's what they're going to try to do by 2025 and once they can do that that's when they want to bring out their revolutionary lego pieces the revolutionary small big and gpu architectures that can be mixed and matched to hopefully take all performance crowns for that unquestioned leadership pat gelzinger says they're going to achieve so think about kind of what i'm saying here how alder lake is the first step alder lake brings out combining multiple types of core architectures and graphics together for well a mostly monolithic lineup though but then it's going to be iterated on with raptor lake and eventually disaggregated into meteor lake which let me give you some meteor lake quotes here meteor lake can almost be considered the final iteration of the alder lake raptor lake design into a disaggregated architecture with tiles meteor lake uses a separate z tile with 2.5 or three stacking meteor lake competes the first half of the journey to intel's revolutionary brand new architecture the first half is accomplishing heterogeneous cores in one tiled soc so you see what's going on here it's kind of like from alder lake to raptor lake to meteor lake getting good at the scheduling of big little of multiple core architecture splitting off the gpu into its own tile and then really as many things as you can where it makes sense into different tiles to mix and match in stack you see in fact when i saw jim's aero lake leak i said you know 320 execution units guys that seems like a lot not just a lot for being a die shrunk architecture to me you could almost suggest that though they're not getting multiple gpu tiles working for a gaming gpu with dg2 then it might even just share one of the gpu tiles with dg3 or dg4 you see that's what that massive gpu suggests to me that that it might be using tsmc for that gpu tile and sharing it with gaming graphics cards now i can't specifically confirm that but i can say that is intel's end goal and that's a big gpu to just include i don't think it's just to compete with apple and then once intel masters it once they master disaggregation and combining multiple tiles into a 3d stacked lineup that is when they're hoping to have well what's been called a royal cove project done and honestly i wasn't sure how much i was going to talk about this royal core project in this video because it is fairly sensitive but at this point i do have multiple sources from around intel who can confirm it and one of those sources is the one that told me about z super sampling so this is 100 true even if what goes into lunar lake changes and it will even if what goes into nova lake changes and it will it's years away guys it is not done things are going to change i at least now know that 100 multiple people in intel know it so it's not as sensitive as it was earlier this year and that this story is true and this is the story of jim keller building a zen 5 killer and i just think you guys are going to want to know about it but yeah so royal core is the project for the architectures that jim keller was working on to evolve past the core series the biggest uplift in decades at intel it aims to build a new series of architectures that ensure intel's x86 products can win in efficiency over arm and apple for the foreseeable future it was and i'm 100 sure of this initially planned to roll out in 2024 with lunar lake but much of this royal upgrade could really be coming out in nova lake remember that rdna 2 is kind of what the final rdna 1 was supposed to be but rdna 1 was good enough in the mid-range that they launched it early at amd things that are designed and being planned three four five years out things change what actually gets implemented into each thing is not going to be 100 known for a couple years but some pretty crazy things are coming guys what i will say is that intel wants to get as much of what i'm about to show you out in something by 2025 right so let's go through some of these quotes once disaggregated architectures are achieved in meteor lake lunar or nova lake this person was unsure though seemed to think it was actually lunarlike brings in the new architectures that should be the biggest leap in intel performance since the introduction of the core series up to a doubling of ipc over golden cove and lunar lake aims to bring at least a 30 ipc increase over meteor lake with lion cove and nova lake extends this with panther cove and i've actually seen more of an ipc increase over golden code but again i think you have to understand 2025 these designs are not done and people that tell you they're sure what they will be is wrong if they can get some of these massive ipc increases ready and into lunar lake before nova cove is out that's what intel's gonna do but it might have to wait for panther cove in nova lake sorry guys jesus and yeah they're considering doing some pretty crazy things four-way hyper threading is mentioned on multiple architectures after 2024 as potential ddr5 at 7 400 megahertz integrated machine learning accelerators and more are on the table by implementation in the next five years but first they need to master disaggregation scheduling and splitting up everything into three sac tiles stacking itself is hard after that is when they try to move to their crazy ambitious architectures that will be in lunar or nova lake and in things like diamond cove or in diamond lake sorry guys the code names confuse me sometimes and things like sierra forest that is what's going on that is what intel's end goal and it is not as simple as just more of this more of that but when you think about what they're trying to do and how good maybe they will get with communicating between tiles i think you can see what's possible here like i've said this on broken silicon multiple times if you can get big little to work then you can either it works or it doesn't and if big little does work well by raptor lake then why not just continue to give you 8 to 12 big cores every year with 10 to 30 percent more ipc every year and then just add a ton of little cores you know again either the multi-threading and scheduling with big little works or it doesn't if it does just give us a lot of ipc on eight or less cores or something like that and then a ton of little cords that take up significantly less die space which i outlined in a leak earlier this year is a more efficient way of scaling performance with die space so that's what intel is doing the possibilities constantly kind of sound endless here and a final thing i want to close on in this video is a word to the people at intel nothing i'm disclosing in this video and i did hold back on some pretty key stuff is stuff that amd doesn't know stuff that they're already saying behind the scenes the difference is that the people at amd don't really think you're gonna succeed at half of it so i guess it's up to you people that work at intel to know that you are being severely underestimated right now at least for the next few years and well i hope you proved them wrong and just so you guys know the people at amd are working on some pretty insane stuff as well and you should not hold back not just to prove them wrong but because if you don't succeed in the things i've outlined here i think they will have things around that anyways and i am so excited for the competition coming over the next few years and as you can guess there will be many more specific leaks as we get closer to those releases you know similar to how i outlined the alder lake lineup before it came out and gave key details on raptor lake we need to get closer to lunar lake to even be sure what lunar lake is versus nova lake because even intel does not know yet but i will be here to tell you about it so don't forget to subscribe to the channel moore's law is dead and ring the bell button and if you do like these videos consider sharing them it really helps a lot or supporting me dan girard the morris law's dead team on patreon where you can get exclusive podcasts every other week where you get ad-free versions of broken silicon the chance to ask me and guest questions that are read alive on air on broken silicon and access to a discord with people who i know would love to discuss the implications of what i've outlined in this video with you right now and as always though no matter what thank you for watching

Pasted that here: https://www.rapidtables.com/tools/notepad.html

and reading with Verdana font zoomed in. Much easier to read.

 

[igor_kavinski]

TL;DR

Intel feels seriously threatened by Apple.
Arrow Lake is meant to be a response to that.
Intel's plan is to deliver 30% performance increase year over year.
Emeralds Rapids launching in 2023 should be really exciting for the massive improvement it may bring.
64 cores on Intel 7 using 350W.
Diamond Rapids is supposed to be the BIG one that will finally catch up to AMD in server performance.
Not expected till 2025.
Emerald Rapids should be 30% better than Sapphire Rapids but still won't come anywhere close to Genoa.
Meteor Lake will be the finalization of the Raptor Lake design in disaggregated form.
320 GPU execution units so possibly using TSMC for that tile.
Multiple tiled 3D stacked design.
Something about Z super sampling???
Jim Keller built a Zen 5 killer in Royal Core.
Biggest uplift in decades that will be a new era after the Core series.
Some confusion from the source. Possibly Lunar Lake is the one with Royal Core?
Panther Cove in Nova Lake may bring more than double IPC over Golden Cove.
SMT4 in multiple architectures after 2024.
This gem: "we need to get closer to lunar lake to even be sure what lunar lake is versus nova lake because even intel does not know yet"

Wow. Thank God, I didn't waste time watching that video.

 

[TheELF]

and make a VAST improvement to MT while also improving peak ST scores for the whole thing by reallocating die space on the COVE cores for improving ST performance instead of supporting SMT. There would be no drop in total threads either.

Die space has no impact on any of this, they can increase the space for the CPU without cutting SMT or they can remove SMT and just save money on die space by shrinking it.
They don't need to reduce one to increase the other.

I expect we'll see the cove cores go away eventually and replaced by atom.

Nah, there will always be a reason to have a more power efficient core and one that is more performance oriented.
Maybe the archs will become similar enough that they will just merge them into one though but I don't think so.

And once a single thread on an E-core achieves or even surpasses the performance of one of two threads on an SMT enabled P-core the tradeoff between SMT enabling P-cores and just smaller more efficient E-cores is favoring the latter.

But that only ever could possibly happen when running heavy duty server level stuff, these are still desktop PCs we are talking about.
And even on server they are not all only number crunching systems so that everybody would want this.

It already does. Remember, SMT is only adding up to 30% or so more performance to MT scenarios, and that's under optimum conditions. Sometimes, it can be a few percent degradation if both tasks are competing for the same resources or are causing constant cache flushes.

No 30% is the worst case scenario when running server type stuff, close to 100% would be optimum conditions and those exist more than you think.
When you have a only few percent variation it's within margin of error and becomes irrelevant, especially to the end user.

 

[itsmydamnation]

Jim Keller built a Zen 5 killer in Royal Core.


Wow. Thank God, I didn't waste time watching that video.

Reading between the line on one of Jims most recent AI speeches, i interpreted it as his time at intel was a failure.

So i guess we will see.


edit:
in this one somewhere

i dont have time to find the time stamp

 

[Exist50]

There's really no sense referencing MLID for leaks, much less for Royal.

No 30% is the worst case scenario when running server type stuff, close to 100% would be optimum conditions and those exist more than you think.

No, 30-40% is a good average for the throughput gain from SMT. While there are occasional workloads that gain significantly more, it's more common to see even less.

Reading between the line on one of Jims most recent AI speeches, i interpreted it as his time at intel was a failure.

I feel like people are trying to read way too much into a joke. He said something like "There are a lot of great engineers at Intel, but sometimes it was like working in a computer museum. Working there aged me a decade."

 

[itsmydamnation]

I feel like people are trying to read way too much into a joke. He said something like "There are a lot of great engineers at Intel, but sometimes it was like working in a computer museum. Working there aged me a decade."

16 years @ DEC
5 years @SiByte
4 years @ PA semi
4 years @ apple
4 years @ amd
2 years @ tesla
2 years @ intel

So to me there is a clear pattern and we know CPU design has a long lead time.

sure we dont have all the pieces of the puzzle , but if you had to put your own money on it , which way would you go ?

 

[Geddagod]

Didn't Keller claim that his job at Intel was mostly just high level team organizational stuff, and not really architecture design or anything like that?

 

[Exist50]

16 years @ DEC
5 years @SiByte
4 years @ PA semi
4 years @ apple
4 years @ amd
2 years @ tesla
2 years @ intel

So to me there is a clear pattern and we know CPU design has a long lead time.

sure we dont have all the pieces of the puzzle , but if you had to put your own money on it , which way would you go ?

He made a vague joke in a talk. Actually, one of several (the TSMC/Samsung one was funny). For better or worse, I don't think there's any meaningful conclusion about Intel's products that one can draw from there, much less for Royal in particular.

It's worth remembering that Keller was not an architect on Royal, even if he did contribute to its creation. He's historically been very quick to point this out for products like Zen. He does a great job at getting the ball rolling, but the end result depends on the broader team, of which we publicly know very little.

I've been very open in stating I'm an optimist about Royal. Probably one of the most optimistic on this forum. But if it's really still years out, then certainly too soon to know either way.

 

[Geddagod]

I've been very open in stating I'm an optimist about Royal. Probably one of the most optimistic on this forum. But if it's really still years out, then certainly too soon to know either way.

Drawback of following tech leaks, info is sparse and very often unreliable. Hard to learn the intricate workings of what people are talking about, especially with lack of resources along with the innate complexity of the technology.

At the very least about SMT 4, we know the tech is possible and out in the real world. The rumor that this is in a major next architecture project is way more believable to me, IMO, than some of the crazier core fusion ideas floating around.

 

[Exist50]

Drawback of following tech leaks, info is sparse and very often unreliable. Hard to learn the intricate workings of what people are talking about, especially with lack of resources along with the innate complexity of the technology.

At the very least about SMT 4, we know the tech is possible and out in the real world. The rumor that this is in a major next architecture project is way more believable to me, IMO, than some of the crazier core fusion ideas floating around.

Oh, yeah, the core fusion thing is BS. SMT4 is, at least, a perfectly feasible technology. The only questions are around the value proposition and if/where it would actually show up in products. I lean towards the view that with small cores increasingly in the picture + side channel vulnerabilities, it doesn't make sense to double down on SMT, but that's a much more tractable issue than core fusion.

Anyway, I've also been very vocally annoyed at the nonsense around Royal, which is a shame given that it genuinely deserves to be talked about. Frankly, @Cardyak's observations and speculation in Lakes/Rapids thread has been far, far better than anything I've seen out of the traditional rumor mill.

 

[igor_kavinski]

I lean towards the view that with small cores increasingly in the picture + side channel vulnerabilities, it doesn't make sense to double down on SMT, but that's a much more tractable issue than core fusion.

SMT4 will improve branchy code through enhanced speculative execution or is it gonna make it worse?

Hot Chips 2020: Marvell Details ThunderX3 CPUs - Up to 60 Cores Per Die, 96 Dual-Die in 2021

www.anandtech.com

That table is for ARM ISA. Is there anything really bad about x86 architecture that would prevent Intel/AMD from seeing these 4-thread scaling gains?

 

[Doug S]

Intel feels seriously threatened by Apple.

Who is eating Intel's lunch right now? It isn't Apple, the damage there has been done and the magnitude of Intel's performance increases will have little effect since Macs are not going back to x86 regardless of how much Intel's chips may improve.

Intel is feeling seriously threatened by AMD. If AMD has stayed "in their place" a couple years behind Intel performance like they have for most of their history then Intel would be fine now. Profits would be high, they would still have process issues but Wall Street would totally ignore them and there would be no layoffs.

 

[moinmoin]

Who is eating Intel's lunch right now? It isn't Apple, the damage there has been done and the magnitude of Intel's performance increases will have little effect since Macs are not going back to x86 regardless of how much Intel's chips may improve.

Intel is feeling seriously threatened by AMD. If AMD has stayed "in their place" a couple years behind Intel performance like they have for most of their history then Intel would be fine now. Profits would be high, they would still have process issues but Wall Street would totally ignore them and there would be no layoffs.

If that part of the "leak" has any truth to it the long lead times can have had the effect that Intel internally actually reacted first to Apple moving to Apple Silicon (something Apple would have informed Intel about early on through reduced orders at the very least). For servers the roadmap may have not changed much anyway, there the primary problem have been all the delays and those keep adding up since.

 

[Exist50]

SMT4 will improve branchy code through enhanced speculative execution or is it gonna make it worse?

Eh? SMT doesn't help (actually, actively hurts) single thread execution. The idea is you can take advantage of underutilization from one thread to fit in more. Very loose correlation to speculative execution.

That table is for ARM ISA. Is there anything really bad about x86 architecture that would prevent Intel/AMD from seeing these 4-thread scaling gains?

Nothing fundamental, but that chart shows the problem. For Gracemont vs Golden Cove, just to use one x86 reference point, a 4 core Gracemont cluster provides ~2x the throughput of a Golden Cove core (SMT included) with approximately the same area and power budget. With the same kind of scaling from that chart, even if you added SMT4 to Golden Cove with zero incremental power or area overhead...you'd still be better off with Gracemont. So what exactly is the point?

 

[Exist50]

If that part of the "leak" has any truth to it the long lead times can have had the effect that Intel internally actually reacted first to Apple moving to Apple Silicon (something Apple would have informed Intel about early on through reduced orders at the very least). For servers the roadmap may have not changed much anyway, there the primary problem have been all the delays and those keep adding up since.

I think there's also something to be said for AMD being a much more "traditional" competitor. They're not doing anything fundamentally different than Intel; they're just doing it better. If Intel gets the process on par, and maybe some improvements to the big core PPA, then they'd do just fine vs AMD. Those are two areas where they've been strong historically, so perhaps they think it's a much more straightforward task.

Apple, however, is trickier. Low power, deep hardware-software integration, etc. These are things Intel's never really done well, or simply doesn't have any experience in. They'd require a much more fundamental shift in the company to tackle. And there's also the matter of fungibility. Someone who switches to AMD this gen probably wouldn't have any issue switching back to Intel in a few years, if the products were competitive. But with Apple's ecosystem lock-in, that might not be the case, even with a competitive lineup.

 

[Cardyak]

Anyway, I've also been very vocally annoyed at the nonsense around Royal, which is a shame given that it genuinely deserves to be talked about. Frankly, @Cardyak's observations and speculation in Lakes/Rapids thread has been far, far better than anything I've seen out of the traditional rumor mill.

Thank you, but you are far too kind! I dumped a load of patents and ideas in the Lake's and Rapid's thread here. But these are literally just microarchitecture ideas that have been explored or theorized in the past, they are most likely completely orthogonal to Royal Core, I'm just compiling some research into the more grandiose ideas into increasing Single Threaded performance and highlighting them for everyone to see.

Sometimes I think that, [in regards to Royal Core], what if it doesn't involve any of these revolutionary ideas at all? People like to focus on sexy new tech, but sometimes the best engineering solutions are more iterative and evolutionary rather than revolutionary. Royal Core could be the same approach as Sunny/Golden Cove, and the generations that came before it. Wider, Deeper, Smarter. Just dialled up to 11. Much wider, much deeper, much smarter.

It would be kind of ironic, after all this exciting speculation for Intel to just come out in 3 years time and simply say. "Forget SMT 4, Out of order retirement, Core fusion, and all that other crazy nonsense. Royal Core is basically Golden Cove x3. 18 Wide Decode, ~1500 entry reorder buffer, 15 ALU's. That's it, bye"

Honestly, you'd love to see it.

 

[DrMrLordX]

18 Wide Decode

At what, 2 GHz?

 

[Mopetar]

There's really no sense referencing MLID

Could have stopped right there.

 

[Hougy]

In case you haven't watched it, save time by reading the transcript:



Pasted that here: https://www.rapidtables.com/tools/notepad.html

and reading with Verdana font zoomed in. Much easier to read.

I always wanted youtube to have transcripts, it's so incredibly useful, but I didn't know they already had until now, so thank you so much!

 

[Doug S]

Apple, however, is trickier. Low power, deep hardware-software integration, etc. These are things Intel's never really done well, or simply doesn't have any experience in. They'd require a much more fundamental shift in the company to tackle. And there's also the matter of fungibility. Someone who switches to AMD this gen probably wouldn't have any issue switching back to Intel in a few years, if the products were competitive. But with Apple's ecosystem lock-in, that might not be the case, even with a competitive lineup.

I don't see Apple as being much of a competitor of Intel's. Not in any way. Sure at the margins there are some people who want a silent / long battery life laptop with high performance and might be willing to switch between Windows and Mac to get the best choice. Most people will not, they make the decision between Windows and Mac before they even begin looking at hardware.

Apple was already selling Macs that were competitively worse than Intel's offerings pretty much the entire time they were x86. Other than right after the release of a new model, if it got the latest and greatest from Intel, Intel always had something better than what was in a Mac. That didn't stop Apple from selling 20 million Macs a year. So even if Intel's products are superior to Apple Silicon, it isn't going to hurt Apple's sales. People who are buying Mac chose to buy a Mac first before looking at what else is out there. People are buying Windows are mostly unaware of what Apple offers and wouldn't consider them even if they were so clearly better it wasn't possible for anyone to come up with a benchmark saying otherwise.

 

[moinmoin]

So even if Intel's products are superior to Apple Silicon, it isn't going to hurt Apple's sales.

That doesn't make it better for Intel. By losing Apple as customer Intel lost a significant chunk of the market. They need to recover that loss somehow somewhere.

 

[Exist50]

Thank you, but you are far too kind! I dumped a load of patents and ideas in the Lake's and Rapid's thread here. But these are literally just microarchitecture ideas that have been explored or theorized in the past, they are most likely completely orthogonal to Royal Core

You identified a Royal architect and posted some of his known research topics. That's a far better method than the random guessing the so-called leakers seem to do. And better for discussion as well.

It would be kind of ironic, after all this exciting speculation for Intel to just come out in 3 years time and simply say. "Forget SMT 4, Out of order retirement, Core fusion, and all that other crazy nonsense. Royal Core is basically Golden Cove x3. 18 Wide Decode, ~1500 entry reorder buffer, 15 ALU's. That's it, bye"

I think that will end up being closer to the reality. I'm sure they will have some cool architectural features in store, but the bulk is going to have to come from elsewhere.

 

[Exist50]

I don't see Apple as being much of a competitor of Intel's. Not in any way. Sure at the margins there are some people who want a silent / long battery life laptop with high performance and might be willing to switch between Windows and Mac to get the best choice. Most people will not, they make the decision between Windows and Mac before they even begin looking at hardware.

Apple was already selling Macs that were competitively worse than Intel's offerings pretty much the entire time they were x86. Other than right after the release of a new model, if it got the latest and greatest from Intel, Intel always had something better than what was in a Mac. That didn't stop Apple from selling 20 million Macs a year. So even if Intel's products are superior to Apple Silicon, it isn't going to hurt Apple's sales. People who are buying Mac chose to buy a Mac first before looking at what else is out there. People are buying Windows are mostly unaware of what Apple offers and wouldn't consider them even if they were so clearly better it wasn't possible for anyone to come up with a benchmark saying otherwise.

I think people are a little more willing to switch that you make them out to be. Certainly Apple's Mac sales with the ARM transition have been quite favorable. There's only so long this kind of gap can remain.

 

[eek2121]

I think people are a little more willing to switch that you make them out to be. Certainly Apple's Mac sales with the ARM transition have been quite favorable. There's only so long this kind of gap can remain.

Apple has not moved the market by even a single percent with their move to ARM, FYI. Most M1/M2 buyers are repeats, with a small portion of college kids and wannabe CEOs buying a machine to look like they are doing important stuff.

I am actually hyper optimistic about RISC-V.

 

[soresu]

Sure at the margins there are some people who want a silent / long battery life laptop with high performance and might be willing to switch between Windows and Mac to get the best choice. Most people will not, they make the decision between Windows and Mac before they even begin looking at hardware

I think you have fallen into the trap of "most people" that many who visit these forums do.

More often than not people visiting or posting here are pretty enthusiast in hardware/software knowledge levels, and we will make purchase decisions differently from the average Joe (aka most people) walking in to a basic retail store, who very often will simply get what the salesman pushes to them because they do not know any better.

I convinced my nan once to go Android instead of iPhone, and after changing she had an easier time doing basic stuff on her phone. Then a year or 2 later she had an iPhone again and was having problems with basic stuff again. Why? Because she broke the Android phone by dropping it and instead of going to me to recommend a replacement she went into a local phone shop and immediately got recommended the new iPhone by the salesman 😱