Question What will the CPU industry marketshare look like in the future? The Rise of ARM?

[Mahboi]

I will not deny that I have a preference for ARM, but I strive to be factual and open-minded.

You DO? Cause lord knows that's not what I'm seeing.

Last time I checked, as per the leakers, Zen5 is said to bring a 40-50% ST performance improvement.

Geekbench 6
Hawk Point: 2700
M3 : 3150
M4 : 3900
Strix Point : 3700-4000

So Zen5 in Strix is going to match M4 ST performance at best, but it will most certainly use significantly more power to do so.

Two weeks. Stop projecting.

Attributing Apple's success to them being on an advanced node is a ridiculous presumption.
The node is just the foundation. It's not magic. The microarchitecture is where the magic is.

https://www.reddit.com/r/hardware/comments/1cxq7em/comment/l54rzba

Case in point: Ryzen 7840HS vs Apple M2

Both score about 2600 points in GB6 ST, but the Ryzen uses about 2x the power to do so. That 2x difference isn't coming from the node, and the irony is that Apple doesn't even have a node advantage here (N4 vs N5P). You can't use the newer uarch excuse, because both Zen4 and M2 are 2022 uarches.

Unless Zen5 is bringing like a 2x ST performance improvement or even better- 2x the performance-per-watt (both of which are unlikely), I don't see any issue for Apple. They'll be fine.

I will not repeat myself.

 

[FlameTail]

It is more accurate to compare the M4 P-core to Z5c as both are N3E and aim to offer a strong balance of performance, power and area.
Z5 classic is N4X and is built for pure performance.

I see no issue in comparing Zen5 (Strix Point) and M4 P-core. Both are laptop chips.

 

[FlameTail]

Two weeks. Stop projecting.

The Zen5 hype train has reached truly dangerous velocities. I have been recieving some threatening messages for trying to dispute this (in another platform, that is).

 

[branch_suggestion]

I see no issue in comparing Zen5 (Strix Point) and M4 P-core. Both are laptop chips.

Good thing there is the Strix Z5c and the Turin Z5c to compare the node diff.
Any comparison should be across as many tests as possible.
And well, Zen5 has maximum per thread performance no matter how many threads are loaded with only power/thermals holding it back.
M4 only scores 3900 with a single thread due to SME, which is a single core using a cluster wide resource, any more than that and the SME effect is bifurcated.
Not a huge difference but it does matter compared to Z5 not sharing any resources from other cores, just a design choice but it does bear mentioning.

At the end of the day Zen5 is still a server first core and platform, with desktop and mobile being secondary.
Zen6 should be fully tuned for specific domains, so finally the priority will be the same.

 

[poke01]

It is more accurate to compare the M4 P-core to Z5c as both are N3E and aim to offer a strong balance of performance, power and area.
Z5 classic is N4X and is built for pure performance.

There’s nothing wrong with comparing both at least for science. After all both Zen cores and M4 P will end up on laptops.

Edit: flame said it before me

 

[branch_suggestion]

There’s nothing wrong with comparing both at least for science. After all both Zen cores and M4 P will end up on laptops.

Yes, I did remember that Strix has both.
At the end of the day AMD isn't bothering with a tablet level chip yet, extremely low power operation is a future design goal.

 

[Nothingness]

@Mahboi I don't disagree, in tech logically at the current rate (that is Zen 5 keeping all the promises) AMD with x86 should be the clear winner even with all the added competition.

But we are mainly talking about the mainstream laptop market, which is dominated by Intel up to now and AMD even with all its clearly superior tech had a rather hard time getting into. Microsoft (mostly through its Surface brand) is now heavily pushing ARM which will mix that up some more. Intel is bound to be the biggest loser from this, and while I don't expect ARM to endanger x86 it's likely to take a significant share.

Beyond laptops, hyperscalers use of Arm is another thing to consider. Desktop is another discussion (I wouldn't consider anything but AMD here), but all other markets are seeing a push for Arm. Except for Apple, Arm laptops might prove a complete failure, but I doubt we'll see hyperscalers drop Arm in the medium term.

Arm provides the possibility for hyperscalers to design their own SoC using a good enough core from Arm. That's a definitive advantage over AMD and Intel both from the flexibility point of view and from the cost point of view (there likely is a TCO advantage but given that no official figure for hyperscaler Arm SoC power consumption is known, it's hard to say). And if anyone thinks the hyperscalers are buying the "RISC hype", they have no clue what they are talking about.

So sorry for Arm bears and x86 apologists, I doubt Arm will decline soon. And that applies to x86 too, it's alas here to stay for way too long.

 

[Nothingness]

ARM rose through phones.

And Intel/x86 rose through PC. Everyone in the workstation and server market laughed at them for a dozen of years. And I was one of them. I was proven very wrong, a good lesson in humility some lack.

 

[Mahboi]

And Intel/x86 rose through PC. Everyone in the workstation and server market laughed at them for a dozen of years. And I was one of them. I was proven very wrong, a good lesson in humility some lack.

Not really comparable now is it? ARM isn't getting a massive 10 year higher income vs an x86 world that would be slowly losing its edge. Which was the case for workstations/servers back in the day. x86 just managed to cover everything well enough and had a bigger TAM and income. Today, ARM's market isn't crushingly higher than x86, so they don't have that kind of strong advantage.

But sure, in 10 years ARM could have broken through every tier of the x86 bunker. It's 10 years away though.

 

[poke01]

Yes, I did remember that Strix has both.
At the end of the day AMD isn't bothering with a tablet level chip yet, extremely low power operation is a future design goal.

True but Intel is with Lunar Lake. It would be interesting to see M3/M4 vs Lunar.

 

[branch_suggestion]

True but Intel is with Lunar Lake. It would be interesting to see M3/M4 vs Lunar.

Yep, same design goals, Lunar will lose but not by as much as people expect. At the end of the day to compete in power you need to xtor bloatmaxx like Apple does.

 

[Nothingness]

Not really comparable now is it? ARM isn't getting a massive 10 year higher income vs an x86 world that would be slowly losing its edge. Which was the case for workstations/servers back in the day. x86 just managed to cover everything well enough and had a bigger market. Today, ARM's market isn't crushingly higher than x86, so they don't have that kind of strong advantage.

Do you happen to know how many CPU Arm sells in the embedded and hyperscaler markets? I don't.

Of course Arm (as a company) revenue is not high as they don't sell any chip contrary to Intel, AMD or NVIDIA. But they seem to have enough revenues to pump out cores that are good enough for many of their customers.

But sure, in 10 years ARM could have broken through every tier of the x86 bunker. It's 10 years away though.

Or Arm could have just disappeared. I don't make any bet that far away, we might be at a turning point... or not

 

[poke01]

Yep, same design goals, Lunar will lose but not by as much as people expect. At the end of the day to compete in power you need to xtor bloatmaxx like Apple does.

yep Nvidia does this too but let’s not go there. Don’t want to die yet😂

 

[poke01]

I will talk moee about Lunar vs M3 later in the year and will make a new thread for it closer to launch. As it’s the first time (I think) in history that Intel and Apple used the same node for the CPU. Both on N3B and both target same design goals.

I will also be looking forward to the hot chips keynote about lunar. It’s honestly the most interesting thing since Skylake from Intel.

 

[soresu]

Qualcomm vs Nvidia vs Mediatek vs AMD vs Samsung vs Huawei

Huawei will be artificially limited to China, or at best Asia.

That's not to say that the potential market there isn't huge, but it is a limit never the less.

 

[Mahboi]

Do you happen to know how many CPU Arm sells in the embedded and hyperscaler markets? I don't.

Of course Arm (as a company) revenue is not high as they don't sell any chip contrary to Intel, AMD or NVIDIA. But they seem to have enough revenues to pump out cores that are good enough for many of their customers.

No clue how many, but it's more a question of margin and volume isn't it?
Phones have insane volume, they're way more sold than PCs nowadays (most people can internet enough with a phone), but it's a highly competitive market where margins for chips cannot be too high. We're miles from the top tiers where duopolies are common.

As for server/hyperscaler, honestly, I don't think it's that much. Every time someone mentions ARM server CPUs, it seems to be some IHFH (in house for house, take that semiconductor industry, I can make mad acronyms too), or it's designed for a fairly specific task. Or it's cheaper than x86 for that task. Or all three. I've yet to see ARM server presented as "taken because it was the best for us". It's always "taken because it's cheaper" or "takent because it's not x86".

Or Arm could have just disappeared. I don't make any bet that far away, we might be at a turning point... or not

I think they'll be fine unless their niches get strongly munched by RISC-V, but really no clue, we'll have to see.

 

[soresu]

Unless Zen5 is bringing like a 2x ST performance improvement or even better- 2x the performance-per-watt (both of which are unlikely), I don't see any issue for Apple. They'll be fine.

Different markets for the most part with Mx being fixed to MacOS.

Especially when you account for server/datacenter too.

 

[branch_suggestion]

yep Nvidia does this too but let’s not go there. Don’t want to die yet😂

Let's wait for BW vs RDNA4 PPA comparisons.
NV's extra bloat is really for stuff that isn't gaming, the 2x render perf/gen is their main design goal since Turing.

 

[branch_suggestion]

Different markets for the most part with Mx being fixed to MacOS.

Especially when you account for server/datacenter too.

Yep,, x86 cores cannot show any weaknesses across the whole shebang of workloads or the other x86 vendor will punish you hard.
Apple Si just needs to do things on MacOS and Geekbench and they can call it a win basically.

 

[SarahKerrigan]

Yep,, x86 cores cannot show any weaknesses across the whole shebang of workloads or the other x86 vendor will punish you hard.
Apple Si just needs to do things on MacOS and Geekbench and they can call it a win basically.

Apple's cores have been well-characterized across a wide range of workloads. They are serious cores, with superb general-purpose performance. This isn't a great take IMO.

 

[Nothingness]

No clue how many, but it's more a question of margin and volume isn't it?
Phones have insane volume, they're way more sold than PCs nowadays (most people can internet enough with a phone), but it's a highly competitive market where margins for chips cannot be too high. We're miles from the top tiers where duopolies are common.

Again Arm doesn't sell chips. They sell licenses and get royalties. You can't compare their business model and revenues with AMD, Intel or Arm customers.

As for server/hyperscaler, honestly, I don't think it's that much. Every time someone mentions ARM server CPUs, it seems to be some IHFH (in house for house, take that semiconductor industry, I can make mad acronyms too), or it's designed for a fairly specific task. Or it's cheaper than x86 for that task. Or all three. I've yet to see ARM server presented as "taken because it was the best for us". It's always "taken because it's cheaper" or "takent because it's not x86".

The best for hyperscalers is "cheaper for them". Where cheaper is a good enough chip which TCO is less than competition, and that satisfies their customers. Definitely not a question of x86 vs Arm as ISA; as I wrote above, they don't care about the now obsolete RISC hype.

I think they'll be fine unless their niches get strongly munched by RISC-V, but really no clue, we'll have to see.

Yeah, big question mark here in the medium term for R-V.

 

[poke01]

Yep,, x86 cores cannot show any weaknesses across the whole shebang of workloads or the other x86 vendor will punish you hard.
Apple Si just needs to do things on MacOS and Geekbench and they can call it a win basically.

Might I remind you that Apple take CPU performance very seriously, so much so that they switched vendors/architectures if they felt their platform lacked CPU performance in laptops and desktops for their designs and “prestige”.

macOS has a wide range of workloads that take advantage of the CPU performance but M4 was released on iPad which doesn’t really have any and if Apple wants to sell new Macs to existing Mac users then they need to improve their cores.

 

[Mahboi]

Again Arm doesn't sell chips. They sell licenses and get royalties. You can't compare their business model and revenues with AMD, Intel or Arm customers.

Don't get semantical, you know I meant the ARM-based SoCs.

The best for hyperscalers is "cheaper for them". Where cheaper is a good enough chip which TCO is less than competition, and that satisfies their customers. Definitely not a question of x86 vs Arm as ISA; as I wrote above, they don't care about the now obsolete RISC hype.

As I said in another discussion in this thread, that is an irrelevant debate. If you boil down to "it's not a question of ISA", then the only question becomes in house talent. AMD has a ton there, so they still get my vote.
And up to now, "cheaper for them" has simply not been anything but x86 yet. And I find no reason to think that the ARM world, who comes from smaller, lesser chips, is going to bridge the gap and leap above the x86 vendors.
Well, I say vendors, I mean AMD. Intel is dying due to its own problems.

Yeah, big question mark here in the medium term for R-V.

It'll be interesting to see where it grows or stumbles.

 

[SarahKerrigan]

Don't get semantical, you know I meant the ARM-based SoCs.

As I said in another discussion in this thread, that is an irrelevant debate. If you boil down to "it's not a question of ISA", then the only question becomes in house talent. AMD has a ton there, so they still get my vote.
And up to now, "cheaper for them" has simply not been anything but x86 yet. And I find no reason to think that the ARM world, who comes from smaller, lesser chips, is going to bridge the gap and leap above the x86 vendors.
Well, I say vendors, I mean AMD. Intel is dying due to its own problems.

ARM most assuredly has been cheaper for hyperscalers. Hence Graviton, Cobalt, and Axion. Google said outright that they've been running a bunch of workloads on ARM servers for a while - do you really think they'd be doing that if it weren't cheaper?

 

[Mahboi]

ARM most assuredly has been cheaper for hyperscalers. Hence Graviton, Cobalt, and Axion. Google said outright that they've been running a bunch of workloads on ARM servers for a while - do you really think they'd be doing that if it weren't cheaper?

Memes, all of them.
Graviton is the poster child for these chips: done in-house for the dozen or so services that AWS could either put on a Xeon/EPYC for full price, or design at home, buy at TSMC, and get 75% of an EPYC/Xeon's perf for 50% of the price. When you control the entire software/HW stack that the chip is meant to serve, it's just a smart call. None of those sell as high performance generic chips in the server market, except Ampere. The rest is all in-house-for-house, or a literal "we taped this awesome Grace CPU to your GPU and you can't buy the GPU without the CPU bruh".