Discussion AMD's Future CPU-APU Gone ARM !!!

[SpudLobby]

The "wild cooling setups" enabled it to hit 4000, but it was already well over 3800 without any cooling at all. So that's like 3-4% "benefit", which is to say not much. If that's all PC overclockers were able to get no one would bother with it.

His point is that the thermal enclosure of the M4 has limited it beyond normal, and so it’s possible power has gone up significantly. It doesn’t really matter what the benefit was, a few % at that tipping point can be quite a lot of heat or power. He isn’t saying it’s as bad as AMD/Intel heat is right?


Idk I think you guys are talking past one another, I happen to agree with Uzzi the M4 is probably drawing like 9-12W max ST. Way better than AMD/Intel @ iso-power perf? Sure. But they’re probably pumping it a bit

 

[uzzi38]

The "wild cooling setups" enabled it to hit 4000, but it was already well over 3800 without any cooling at all. So that's like 3-4% "benefit", which is to say not much. If that's all PC overclockers were able to get no one would bother with it.

3-4% benefit on average, not across all workloads. As noted in the other thread, not all workloads saw any gains, whereas others saw much larger gains, and the reasoning behind that is extremely simple: not all workloads stress a CPU core to the same degree.

The fact that those workloads are seeing much more significant clock gains from the additional cooling, so clearly passive cooling isn't enough. That very clearly indicates >8w power draw in the right workloads.

If you don't like me taking numbers like that as a basis, then you should have no problems with me claiming that Zen 4 is a 15w core at most if you look at lighter workloads or benchmarks. In reality under heavy AVX2 load you can push a Zen 4 core to 20-25w, and that's what how we refer to Zen 4 generally speaking.

 

[StefanR5R]

Wouldn't it be nice if people who desire to make quantitative comparisons of power efficiency would straight away

• state the task energy spent on a given benchmark — that's a figure in J or Ws or Wh (Joule, Watt Seconds, Watt Hours) — as well as
• provide footnotes on which subsystems of the computer were measured, and by which method and apparatus?

What we usually get presented instead is a performance figure combined with a mostly uncorrelated single power draw figure.

(I do admit that I have personally, among fellow dc hobbyists, provided throughput-per-Watt figures much more often than task energy figures. But this was just for the purpose of workload tuning on the own computers rather than vendor/architecture comparisons, and it was for workloads with very steady power draw, such that the claimed power draw times task duration comes very close to task energy indeed.)

 

[marees]

Microsoft's weird ARM fetish:

https://twitter.com/x/status/1792631641107739120

 

[uzzi38]

Agreed.

X5’s average though in Specint and SpecFP was still not 10W. It was like 5.7 and 7 where Apple’s were also as high just with much performance. So yeah while that’s true, I could quote outliers in individual loads for AMD and Intel as well and it’d be fairly unpretty, they’re still ahead of what Phoenix etc would pull down (and remember those workloads measured full power including DRAM etc).

But I agree about the direction it and Apple has gone even if still more efficient than AMD/Intel from a system perspective — though they (Apple especially) get very upset about this because they believe their “core” really only consumes 4-5W for the whole machine lol. It’s hilarious and reminiscent of like, the other guys quoting package or core power. Everyone is completely full of it and it’s so tiresome.

Indeed RE: X5. They have as much IPC as the A14 almost, yet they’re pulling down much more power even on superior nodes. I’ve said this elsewhere, but Arm will need more than just improved IPC — they’ll need a better architecture RE: power, and more cache.

Did you see the M4 iPad GB6 test from a Chinese reviewer? Not the fake 7W BS, the 11W one, which makes far more sense to me, and was measured externally.

https://twitter.com/x/status/1790354528770084945

Except this is averaged over the entire load, and more telling. The performance of course is good, but the direction they’re going is clearly pushing power up more than nodes can afford.

I was reminded of where the ~10W I was thinking of comes from yesterday by Cheese, it's what happens when an X4 sustains 3.6GHz, which is the max clock ARM was advertising. I think the numbers you're thinking of are for 8Gen3, which clock the "prime" cores at 3.2GHz (3.3GHz on the binned Samsung variant iirc).

 

[SpudLobby]

I was reminded of where the ~10W I was thinking of comes from yesterday by Cheese, it's what happens when an X4 sustains 3.6GHz, which is the max clock ARM was advertising. I think the numbers you're thinking of are for 8Gen3, which clock the "prime" cores at 3.2GHz (3.3GHz on the binned Samsung variant iirc).

Thanks! Yep that would make sense. We haven’t seen that in an actual design though, which is my quibble, but given Arm’s IP model it’s completely fair to say it’s a 10W core. The peak of those voltage/F curves are really steep, almost comically so.

TIL. I haven’t even seen that info from Arm, didn’t know they showed that. But will check it out.

 

[Tigerick]

The first ARM core designed by AMD would be the now-shelved K12, which was literally just Zen but with an ARM front-end.

It never came to market and no future development has happened on the design either. AMD are not interested in designing their own ARM cores currently, they're fully focused on x86.

(Btw, the design team that worked on K12 later worked on Zen 3).

Like I said earlier in this thread (probably multiple times by now, I actually don't know on that but I know I've said it at least once): what you're looking at right now is a semi-custom product with standard ARM core IP put together with AMD fabrics (I think, I'm not 100% sure on this one) and GPU IP for an APU. That's it, just a semi-custom product.you really need to strop trying to read into this any more than that.

I am really surprised by your lack of logic thinking and imagination. You should at least link to current event to have an idea what is going on in ARM SoC:

• Microsoft launched new Surface lineup using standard X-Elite. Why don't Microsoft use custom name like before? Cause Surface volume has dropped significantly last year, Qualcomm and Microsoft don't bother to custom name it. Think about future NV and AMD's ARM SoC..
• Here is Qualcomm's list of partners using X-Elite and Plus. Do you really think AMD will limit itself by semi-custom SoC for MS? Think...
• 
• Dell has mentioned they are going to use NV's ARM SoC. This indicates NV going to push heavily in PC market. Do you still think AMD will limit itself and let Qualcomm and NV flood the market? Think.

I have changed the title of this thread; instead of question mark, I am putting exclamation mark. Cause I am pretty sure Zen 7 is using custom ARM core. Unless you have concrete info about Zen7 using x86, please stop limiting yourself about future which is not in the roadmap yet. Lisa Su says AMD is on track to a 100x power efficiency improvement by 2027. Do you really think x86 core could achieve 100x power efficiency?

 

[inquiss]

I am really surprised by your lack of logic thinking and imagination. You should at least link to current event to have an idea what is going on in ARM SoC:

• Microsoft launched new Surface lineup using standard X-Elite. Why don't Microsoft use custom name like before? Cause Surface volume has dropped significantly last year, Qualcomm and Microsoft don't bother to custom name it. Think about future NV and AMD's ARM SoC..
• Here is Qualcomm's list of partners using X-Elite and Plus. Do you really think AMD will limit itself by semi-custom SoC for MS? Think...
• View attachment 99613
• Dell has mentioned they are going to use NV's ARM SoC. This indicates NV going to push heavily in PC market. Do you still think AMD will limit itself and let Qualcomm and NV flood the market? Think.

I have changed the title of this thread; instead of question mark, I am putting exclamation mark. Cause I am pretty sure Zen 7 is using custom ARM core. Unless you have concrete info about Zen7 using x86, please stop limiting yourself about future which is not in the roadmap yet. Lisa Su says AMD is on track to a 100x power efficiency improvement by 2027. Do you really think x86 core could achieve 100x power efficiency?

The ISA doesn't really make any difference to power consumption. Whether AMD can achieve 100x, well that's what they say they can do. ARM Vs x86 wont make a difference on that front though. It's a red herring.

 

[carancho]

How much significant rework would AMD need to do to switch their CPUs to ARM?

 

[gdansk]

How much significant rework would AMD need to do to switch their CPUs to ARM?

Rumor has it they already offer semi-custom SOCs with ARM cores if the customer wants.

 

[uzzi38]

I am really surprised by your lack of logic thinking and imagination. You should at least link to current event to have an idea what is going on in ARM SoC:

I can easily. It goes like this:

1. Microsoft is desperate to compete against Apple M silicon

2. Microsoft decides the only way to do that is by providing ARM silicon of their own (to match Apple) and then focus on AI as the big selling point for Windows.

That leads us to the current situation, pretty much.

• Microsoft launched new Surface lineup using standard X-Elite. Why don't Microsoft use custom name like before? Cause Surface volume has dropped significantly last year, Qualcomm and Microsoft don't bother to custom name it. Think about future NV and AMD's ARM SoC..
• Here is Qualcomm's list of partners using X-Elite and Plus. Do you really think AMD will limit itself by semi-custom SoC for MS? Think...
• View attachment 99613
• Dell has mentioned they are going to use NV's ARM SoC. This indicates NV going to push heavily in PC market. Do you still think AMD will limit itself and let Qualcomm and NV flood the market? Think.

Oh I'm sure AMD would love to sell this semi-custom product more generally to the market, but what will they offer? They're using standard ARM IP (because there is no point to AMD creating an ARM core), and I'll discuss why in a second. AMD's power management and internal fabrics are fine, but not market leading. Their GPU IP is fine, but not market leading. What does AMD really offer vendors aside from being a third option?

I have changed the title of this thread; instead of question mark, I am putting exclamation mark. Cause I am pretty sure Zen 7 is using custom ARM core. Unless you have concrete info about Zen7 using x86, please stop limiting yourself about future which is not in the roadmap yet.

What I have, is common sense and reasoning. That's frankly enough.

Lisa Su says AMD is on track to a 100x power efficiency improvement by 2027. Do you really think x86 core could achieve 100x power efficiency?

Firstly: that claim is for compute nodes, not for client products of any kind. It's a claim based upon power efficiency of CPU, GPU, networking, memory - everything that would fit within a compute node in a datacentre. It has nothing to do with CPUs alone. In most compute nodes CPU is just one portion of the puzzle, and there's nore headroom for power efficiency gains in the short term in everything outside of networking/memory etc that most of that 100x figure will come from there.

Now for the wakeup call: ARM is not some kind of magical silver bullet that gets you tons of power efficiency for free. Whether or not Zen 7 is x86 or ARM will have absolutely no real impact on power efficiency. There is nothing intrinsic to ARM that makes it lower power than x86: the only thing that gets you there is damn good engineering and specific targets for power efficiency.

Unlike the magical touted power efficiency gains from switching ISA, there is a very real difference in software moats that makes sticking with x86 significantly more valuable than switching to ARM. There is no reason for AMD to switch their main core design to ARM, only downsides.

 

[Cheesecake16]

I can easily. It goes like this:

1. Microsoft is desperate to compete against Apple M silicon

2. Microsoft decides the only way to do that is by providing ARM silicon of their own (to match Apple) and then focus on AI as the big selling point for Windows.

That leads us to the current situation, pretty much.



Oh I'm sure AMD would love to sell this semi-custom product more generally to the market, but what will they offer? They're using standard ARM IP (because there is no point to AMD creating an ARM core), and I'll discuss why in a second. AMD's power management and internal fabrics are fine, but not market leading. Their GPU IP is fine, but not market leading. What does AMD really offer vendors aside from being a third option?



What I have, is common sense and reasoning. That's frankly enough.



Firstly: that claim is for compute nodes, not for client products of any kind. It's a claim based upon power efficiency of CPU, GPU, networking, memory - everything that would fit within a compute node in a datacentre. It has nothing to do with CPUs alone. In most compute nodes CPU is just one portion of the puzzle, and there's nore headroom for power efficiency gains in the short term in everything outside of networking/memory etc that most of that 100x figure will come from there.

Now for the wakeup call: ARM is not some kind of magical silver bullet that gets you tons of power efficiency for free. Whether or not Zen 7 is x86 or ARM will have absolutely no real impact on power efficiency. There is nothing intrinsic to ARM that makes it lower power than x86: the only thing that gets you there is damn good engineering and specific targets for power efficiency.

Unlike the magical touted power efficiency gains from switching ISA, there is a very real difference in software moats that makes sticking with x86 significantly more valuable than switching to ARM. There is no reason for AMD to switch their main core design to ARM, only downsides.

Irony being that by Zen 7, or whatever it is called, launches, AMD could have added the APX extension to the core which would make the move to ARM make even less sense IMO....
Heck, they could also have moved to x86S which removes alot of the legacy stuff that folks like to complain about....
So yeah... I don't see any point for AMD to move to ARM for their cores...
Nor would AMD move to RISC-V either... too fragmented...

 

[Doug S]

1. Microsoft is desperate to compete against Apple M silicon

I know that's the theory, but I still haven't heard a great explanation as to WHY they should care so much how they compete against Apple Silicon.

Apple switching from x86 to ARM hurt only Intel. It didn't affect Microsoft, those people weren't running Windows before and they aren't running Windows now. There may be some element of "wow I wish I could buy something as quiet and long lasting on battery as a Macbook but for Windows" but as far as people running Windows actually switching there just aren't that many. Otherwise we would have seen a big sustained boost in Mac market share. Instead we saw a big boost when M1 was released - because every Mac owner knew that new Macs were coming and if they were thinking about upgrading most wanted to wait for new ones to do so. Then a reversion basically to the mean. Maybe they gained a little, but it was hardly enough to hit Microsoft in the wallet.

Likewise, if Microsoft sees something running Windows (whether theirs or some other vendor's) that can match or even beat Apple's best, Microsoft is not going to get Mac users to switch. Heck Microsoft doesn't even care whether that competitor is ARM or x86. Some may think that competitor has to be ARM and believe x86 is yesterday's news, but maybe Zen 5 has some lower power versions that are right there with M4 in a fanless design, or Intel gets their process house in order and takes process leadership from TSMC and matches Apple as a result.

Microsoft's turn to ARM is just hedging their bets. If x86 offerings get better they won't feel bad about abandoning ARM support. They've left ARM in the lurch before. Also Alpha, MIPS, and Sparc.

 

[Tigerick]

I can easily. It goes like this:

1. Microsoft is desperate to compete against Apple M silicon

2. Microsoft decides the only way to do that is by providing ARM silicon of their own (to match Apple) and then focus on AI as the big selling point for Windows.

That leads us to the current situation, pretty much.



Oh I'm sure AMD would love to sell this semi-custom product more generally to the market, but what will they offer? They're using standard ARM IP (because there is no point to AMD creating an ARM core), and I'll discuss why in a second. AMD's power management and internal fabrics are fine, but not market leading. Their GPU IP is fine, but not market leading. What does AMD really offer vendors aside from being a third option?

So, you think AMD won't release Sound Wave to the general market because their ARM core is using standard IP, namely Cortex-X5/X6 which won't be competitive compared to others? Then why would Microsoft choose AMD ARM SoC in their future lineup? Why not continue using market leader like Qualcomm and NV? Have you really thought through? And do you know Mediatek and Samsung are using standard IP as well?

Thanks for your lack of confidence of AMD being market leader in PM, internal fabrics, GPU IP and so on. Have you seen any benchmark with AMD ARM APU to make such conclusion? Or it is just your speculation?

What I have, is common sense and reasoning. That's frankly enough.



Firstly: that claim is for compute nodes, not for client products of any kind. It's a claim based upon power efficiency of CPU, GPU, networking, memory - everything that would fit within a compute node in a datacentre. It has nothing to do with CPUs alone. In most compute nodes CPU is just one portion of the puzzle, and there's nore headroom for power efficiency gains in the short term in everything outside of networking/memory etc that most of that 100x figure will come from there.

Now for the wakeup call: ARM is not some kind of magical silver bullet that gets you tons of power efficiency for free. Whether or not Zen 7 is x86 or ARM will have absolutely no real impact on power efficiency. There is nothing intrinsic to ARM that makes it lower power than x86: the only thing that gets you there is damn good engineering and specific targets for power efficiency.

Unlike the magical touted power efficiency gains from switching ISA, there is a very real difference in software moats that makes sticking with x86 significantly more valuable than switching to ARM. There is no reason for AMD to switch their main core design to ARM, only downsides.

Your common sense and reasoning told you it is x86 server platform will achieve 100x power efficiency by 2027.

My common sense and reasoning told me it is ARM server platform will achieve 100x power efficiency by 2027. As explained here two month ago, AmpereOne's 160 core server CPU will only need 60% of the Turin's TDP, does that sound like better power efficiency?

We don't know what TDP AMD would require to house 256-core, we do know AMD have to use N2 process to achieve it. AmpereOne have just announced their 256-core CPU fabbed by N3E process. Can you imagine if AmpereOne is using N2 process to manufacture their 256-core, how much better power efficiency would AmpereOne achieve? Not to mention how many more cores would AmpereOne fits in.

If you still believe x86 server platform would achieve 100x power efficiency by 2027, I respect that. After all, the thread is opened for speculation and discussion. We will see how the outcome in the coming years.

Next year we will see army of ARM SoC coming strong and fast.

 

[branch_suggestion]

So, you think AMD won't release Sound Wave to the general market because their ARM core is using standard IP, namely Cortex-X5/X6 which won't be competitive compared to others? Then why would Microsoft choose AMD ARM SoC in their future lineup? Why not continue using market leader like Qualcomm and NV? Have you really thought through? And do you know Mediatek and Samsung are using standard IP as well?

Because Microsoft is paying them to, it really isn't very complicated.

Thanks for your lack of confidence of AMD being market leader in PM, internal fabrics, GPU IP and so on. Have you seen any benchmark with AMD ARM APU to make such conclusion? Or it is just your speculation?

That is how things currently stand, the future is anyone's guess.

Your common sense and reasoning told you it is x86 server platform will achieve 100x power efficiency by 2027.

The 100x target is all about Instinct and full stack server architecture, x86 vs ARM is such a small slice of that overall equation.

My common sense and reasoning told me it is ARM server platform will achieve 100x power efficiency by 2027. As explained here two month ago, AmpereOne's 160 core server CPU will only need 60% of the Turin's TDP, does that sound like better power efficiency?

Not all cores are made equal, Ampere has been talked about in great length and to put it simply, they just aren't competitive. Not to mention availability of AmpereOne is... not as good as Bergamo.

We don't know what TDP AMD would require to house 256-core, we do know AMD have to use N2 process to achieve it. AmpereOne have just announced their 256-core CPU fabbed by N3E process. Can you imagine if AmpereOne is using N2 process to manufacture their 256-core, how much better power efficiency would AmpereOne achieve? Not to mention how many more cores would AmpereOne fits in.

Once again, the cores are not remotely even, each AMD core has SMT and way higher IPC. Not to mention EPYC is the more robust platform.
Once again, where is the thing?

If you still believe x86 server platform would achieve 100x power efficiency by 2027, I respect that. After all, the thread is opened for speculation and discussion. We will see how the outcome in the coming years.

I beg of you, watch the full ISSCC presentation and read the updated ITF World slides properly.

PS: I remembered you kept saying no OEMs would make X-Elite notebook before. I hope above picture taught you some lessons and stop limiting yourself in the x86 world. Next year we will see army of ARM SoC coming strong and fast.

You have yet to make a single post that has anything insightful to say, really bad speculation and still believing that there is a noticeable difference between ISAs in capability and other lacks of knowledge.
Not exactly what you should be doing on a technical forum but I digress.

 

[uzzi38]

So, you think AMD won't release Sound Wave to the general market because their ARM core is using standard IP, namely Cortex-X5/X6 which won't be competitive compared to others?

I just don't think there's a good reason for OEMs to want to adopt it over standard AMD APUs. That's all. I don't see a way in which AMD has a competitive advantage here (at least right now, we are talking about a 2026+ product after all), and I especially don't see how that competitive advantage also just wouldn't spill over to their x86 parts.

Then why would Microsoft choose AMD ARM SoC in their future lineup? Why not continue using market leader like Qualcomm and NV? Have you really thought through? And do you know Mediatek and Samsung are using standard IP as well?

If you can switch between suppliers, why wouldn't you? I mean why use Nvidia over Qualcomm for that matter, same story applies to them. Where do you think Nvidia has a competitive advantage wrt ARM SoCs vs Qualcomm. Only real spot is the iGPU, but since when have Microsoft cared about iGPU performance?

Thanks for your lack of confidence of AMD being market leader in PM, internal fabrics, GPU IP and so on. Have you seen any benchmark with AMD ARM APU to make such conclusion? Or it is just your speculation?

You are aware that neither power management nor internal fabrics are related to the core IP, right? Doesn't matter if it's an x86 or ARM core, if AMD can do it for one they can do it for the other.

Your common sense and reasoning told you it is x86 server platform will achieve 100x power efficiency by 2027. My common sense and reasoning told me it is ARM server platform will achieve 100x power efficiency by 2027. As explained here two month ago, AmpereOne's 160 core server CPU will only need 60% of the Turin's TDP, does that sound like better power efficiency?

Power efficiency has two metrics to it: performance and power consumption. You're only looking at the power consumption part.

I don't even think the 256 core version of AmpereOne will be competitive on power efficiency with Turin-Dense, much less the 160 core version.

We don't know what TDP AMD would require to house 256-core, we do know AMD have to use N2 process to achieve it. AmpereOne have just announced their 256-core CPU fabbed by N3E process. Can you imagine if AmpereOne is using N2 process to manufacture their 256-core, how much better power efficiency would AmpereOne achieve? Not to mention how many more cores would AmpereOne fits in.

Why would they need N2? Turin Dense uses less CCDs and less CCXs than Turin for more cores (192). Equalize the CCD and CCX count and you get 256 cores.

Power would be a realistic concern, but again, performance is on two totally different tiers.

If you still believe x86 server platform would achieve 100x power efficiency by 2027, I respect that. After all, the thread is opened for speculation and discussion. We will see how the outcome in the coming years.

As @branch_suggestion wrote, most of the gains are going to come from the GPU and other parts of the system.

The CPU in a server is just a fraction of the overall power budget. In modern platforms even memory can eat up hundreds of watts these days. Even if the CPU were 100x more power efficient, it would only account for a small portion of the total power of a compute node.

PS: I remembered you kept saying no OEMs would make X-Elite notebook before. I hope above picture taught you some lessons and stop limiting yourself in the x86 world. Next year we will see army of ARM SoC coming strong and fast.

Now you're just plain lying to prove a point: I never claimed anything like that. The only thing I've been concerned about for X Elite was pricing, but I've been pleasantly surprised there so far

 

[FlameTail]

The only thing I've been concerned about for X Elite was pricing, but I've been pleasantly surprised there so far

It iz very good, my friend.

 

[yuri69]

AmpereOne

Oh well... AmpereOne was unveiled in 2022, "released" in May 2023, and promised to be available as Oracle's OCI Compute A2 instance during 2023. Yet, there is no trace of A2 as of now.

 

[Tigerick]

I just don't think there's a good reason for OEMs to want to adopt it over standard AMD APUs. That's all. I don't see a way in which AMD has a competitive advantage here (at least right now, we are talking about a 2026+ product after all), and I especially don't see how that competitive advantage also just wouldn't spill over to their x86 parts.



If you can switch between suppliers, why wouldn't you? I mean why use Nvidia over Qualcomm for that matter, same story applies to them. Where do you think Nvidia has a competitive advantage wrt ARM SoCs vs Qualcomm. Only real spot is the iGPU, but since when have Microsoft cared about iGPU performance?



You are aware that neither power management nor internal fabrics are related to the core IP, right? Doesn't matter if it's an x86 or ARM core, if AMD can do it for one they can do it for the other.

You are saying Microsoft will switch to AMD in 2026 not because of better features but because Microsoft is asking AMD to make inferior APU? Really? I may not technical competent but if you think Microsoft will downgrade to compete....

Power efficiency has two metrics to it: performance and power consumption. You're only looking at the power consumption part.

I don't even think the 256 core version of AmpereOne will be competitive on power efficiency with Turin-Dense, much less the 160 core version.


Why would they need N2? Turin Dense uses less CCDs and less CCXs than Turin for more cores (192). Equalize the CCD and CCX count and you get 256 cores.

Power would be a realistic concern, but again, performance is on two totally different tiers.



As @branch_suggestion wrote, most of the gains are going to come from the GPU and other parts of the system.

The CPU in a server is just a fraction of the overall power budget. In modern platforms even memory can eat up hundreds of watts these days. Even if the CPU were 100x more power efficient, it would only account for a small portion of the total power of a compute node.

I have laid my points, and we should check out the announcement of BlackHawk in the coming days. Microsoft's WoA Server 2025 should be coming next year as well, then we shall see.

Now you're just plain lying to prove a point: I never claimed anything like that. The only thing I've been concerned about for X Elite was pricing, but I've been pleasantly surprised there so far

Maybe I remembered wrongly, I have removed that sentence for the gesture of goodwill.

 

[Gideon]

Do you really think x86 core could achieve 100x power

You really should look what AMD's own engineers think of the subject:

Here is his original thread:

https://twitter.com/x/status/1770963537675911365

And here followups:

https://twitter.com/x/status/1789148247971287165

https://x.com/philparkbot/status/1793695487046086775?t=fzYXkUXtOk5jzanSt2RmUw&s=19

https://x.com/philparkbot/status/1781745230439420345?t=lqEhFCjez3qQHd68-7C2wg&s=19

He was one of the core designers on the initial X3D cache btw.

And it helps to take a look at the quoted articles in the threads (from reputative oeople thst know their stuff)

Does x86 Need to Die?

A recent article claimed that "x86 needs to die". Does it? I was on ThePrimeagen's live stream last week to talk about it.

www.computerenhance.com

Why x86 Doesn’t Need to Die

Hackaday recently published an article titled “Why x86 Needs to Die” – the latest addition in a long-running RISC vs CISC debate. Rather than x86 needing to die, I believe the RISC vs C…

chipsandcheese.com

 

[inquiss]

I know that's the theory, but I still haven't heard a great explanation as to WHY they should care so much how they compete against Apple Silicon.

Apple switching from x86 to ARM hurt only Intel. It didn't affect Microsoft, those people weren't running Windows before and they aren't running Windows now. There may be some element of "wow I wish I could buy something as quiet and long lasting on battery as a Macbook but for Windows" but as far as people running Windows actually switching there just aren't that many. Otherwise we would have seen a big sustained boost in Mac market share. Instead we saw a big boost when M1 was released - because every Mac owner knew that new Macs were coming and if they were thinking about upgrading most wanted to wait for new ones to do so. Then a reversion basically to the mean. Maybe they gained a little, but it was hardly enough to hit Microsoft in the wallet.

Likewise, if Microsoft sees something running Windows (whether theirs or some other vendor's) that can match or even beat Apple's best, Microsoft is not going to get Mac users to switch. Heck Microsoft doesn't even care whether that competitor is ARM or x86. Some may think that competitor has to be ARM and believe x86 is yesterday's news, but maybe Zen 5 has some lower power versions that are right there with M4 in a fanless design, or Intel gets their process house in order and takes process leadership from TSMC and matches Apple as a result.

Microsoft's turn to ARM is just hedging their bets. If x86 offerings get better they won't feel bad about abandoning ARM support. They've left ARM in the lurch before. Also Alpha, MIPS, and Sparc.

Some excellent points.

I've been mulling over two theories. One is that Microsoft wants to have windows be an apple-like relatively locked down evo-system and in the long term will make their own chips using the traditional vendors to do the hardware leg work initially while the platform matures.

The second, more likely, is along what you said. I think it's a reaction to Intel's stumble at 10nm and realising that if progress in CPU exists within arm but not within x86 (not because of the ISA itself. But because of both Intel and AMD stumble), then they don't want to be caught with their pants down.

 

[FlameTail]

Microsoft is playing Game of Thrones with the hardware vendors.

Intel vs AMD vs Qualcomm vs Nvidia vs Mediatek vs etc...

Microsoft is playing off each of these vendors against each other. And they (Microsoft) stand to benefit from the increased competition.

 

[adroc_thurston]

He was one of the core designers on the initial X3D cache btw.

He's not the V$, no.
Fabric stuff.

Microsoft is playing Game of Thrones with the hardware vendors.

no.

Intel vs AMD vs Qualcomm vs Nvidia vs Mediatek vs etc...

They're just unmedicated.

And they (Microsoft) stand to benefit from the increased competition.

No, they're desperate to monetize all the money they torched on openAI.

 

[Doug S]

Some excellent points.

I've been mulling over two theories. One is that Microsoft wants to have windows be an apple-like relatively locked down evo-system and in the long term will make their own chips using the traditional vendors to do the hardware leg work initially while the platform matures.

The second, more likely, is along what you said. I think it's a reaction to Intel's stumble at 10nm and realising that if progress in CPU exists within arm but not within x86 (not because of the ISA itself. But because of both Intel and AMD stumble), then they don't want to be caught with their pants down.

Well I think the WHY is pretty simple. Microsoft's leadership is jealous of all the free (and positive) press Apple got over their move to ARM, and wished someone would talk about Microsoft or Windows in such glowing ways.

What they don't understand is that no one loves Windows in the way that some love their Mac. Part of that is because no one loves an OS (people might love their Mac, and they might like macOS, but they don't love macOS) and part of that is because most people associate the Windows OS and Windows PCs in general with work and with homework. You can built at best a tolerant relationship on that basis, but you'll never get to love.

 

[gdansk]

In Microsoft's threat analysis ChromeOS must be more important than MacOS. There's nothing they can do about MacOS (since its adoption is driven by iOS).
Even if that's nowhere near where they are targeting now - that's where it started. Qualcomm pivoted up market because of Apple and bought Nuvia to do the dirty work (two months after M1).