The next big architecture

[LemonParty]

I consider that the time of x86 is coming to the end. x86 not wining nor in speed, nor in energy efficiency.
And instruction set of x86 is kind of bloated.

What will be the next "big" architecture? Will it be ARM or RISC-V?

[marcov]

I consider that the time of x86 is coming to the end. x86 not wining nor in speed, nor in energy efficiency.
And instruction set of x86 is kind of bloated.

Keep in mind that it is often general purpose x86(_64) software with  20 years of backwards compatibility compared with relative new, almost legacy free offerings from new applicants. Benchmarks are hard to compare also in other ways.

In some ways this reminds me of the mid nineties attempt to bring Power(PC) into the PC market. Lots of drama, but disappointing results and weak and postponed next generations, followed by  a quick exit, leaving early movers in a bind. The initial Qualcom laptop lineup took nearly a year to refresh, and got a lot more expensive in the process. Top performing parts by other vendors are promised but slow to emerge for client use.

And remember, one of the problems of all non x86 experiments seems to be that when a generation goes out of fashion, after a few years, support essentially falls of a cliff.

What will be the next "big" architecture? Will it be ARM or RISC-V?

Definitely ARM, though I would put  x86_64 remaining before either of them. I don't really see the benefit for vendors to chose RISCV.  While instruction licensing is cheaper, RISCV silicon reference designs are actually more expensive I heard. And that matters most for the small vendors that RISCV is supposed to be revolutionary for.

If x86 engineers find any kind of kludge to better parallellise the instruction decoder, that makes the whole discussion moot. As x86 might be suddenly on the advantage due to code density.

In summary, I don't see any big change, relative to a few years ago. Yes, some cloud providers might switch from x86 to arm, but for PC replacing clients? I don't see clear signs that a change is coming. Worse, I think the two failed attempts (the original Surface, and the Qualcom powered laptop generation from a few years ago) will make any new contender think twice.

If you really, really want non x86 for some reason, Apple seems to be the only (relatively) safe choice.

[LemonParty]

Yes, the first attempt of Qualcomm was fail. But the second one is close to solutions from Apple.
For reference top CPUs from Intel score 3000 in single core benchmark, when Apple has 4300. That is quite a big difference.

[marcov]

Yes, the first attempt of Qualcomm was fail. But the second one is close to solutions from Apple.
For reference top CPUs from Intel score 3000 in single core benchmark, when Apple has 4300. That is quite a big difference.

Which benchmark? Does it do anything that can be compared (e.g. a clearly equal amount of useful work?).  And even then how do you know it is the cores?  It might just be testing Apple's different approach to memory (soldered on, close to CPU) compared to regular DIMMs. Is there price parity ?

Cross architecture benchmarking is simply hard.

Yes, the remaining fact left over of the RISC philosophy is that instruction decoders parallelise easier. With Apple (and other top tier ARM vendors like NVidia) aggressively widening the CPU that is important. But it is a relative recent development, and it is not 100% sure that this can't be solved for more complex cases like x86 if you invest enough (CPU developer) time in it.

[LemonParty]

Geekbench.
That's the moment where compared CPUs https://youtu.be/Q77AzvY3FTE?t=373
The rest of the video is interesting too.

[marcov]

Geekbench.
That's the moment where compared CPUs https://youtu.be/Q77AzvY3FTE?t=373
The rest of the video is interesting too.

IMHO it doesn't really have much content. Random benchmarks on very strange chosen laptops with a vibe-y voice over. Eur 3700 and the processor isn't widely introduced yet ?!?!?   Also it is ODM vs all-integrated Apple.

And I repeat: How do you isolate the CPU anyway is such Apple to Oranges(Asus) comparison, specially since a "single core" benchmark probably also involves memory use  ?   And if you don't, how relevant is it?

That said, I find it really hard to find actual benchmarks that give me a feel for any system nowadays. If you don't strongly lean towards content creation (so Cinebench etc), and don't game heavy titles, what is relevant?  Office suite benchmarks are gone due to 365 (and worse, the termination of eternal Office/Mac licenses)

And why not simply compare a normal price range that private people actually buy, let's say in the Eur 800-900 price range. Or even entry level, say Eur 500 ? Why these extremes?

[Marc]

If you really, really want non x86 for some reason, Apple seems to be the only (relatively) safe choice.

For as long as it takes, seen their architectures from the past 68k, ppc, x64, arm....

[marcov]

If you really, really want non x86 for some reason, Apple seems to be the only (relatively) safe choice.

For as long as it takes, seen their architectures from the past 68k, ppc, x64, arm....

Within the usual Apple constraints, yes. Lifecycles both by Apple itself, and with 3rd party Apple specific software.

The lifecycle thing was my beef with apple since the powerpc deprecation, but with Microsoft now also deprecating fully functioning software, that becomes a more complex issue

[VisualLab]

It's probably too early to declare victory for the "divine" ARM architecture and proclaim the fall of "Great Babylon" in the form of a 64-bit version of the x86 architecture. As previous speakers have noted, there's no indication that this architecture is about to decline.

The fact that Apple currently uses ARM means nothing, because their hardware and software are more like "IT jewelry" or "IT fetish" than a commonly used tool. Yes, they have their own relatively stable niche (for many years), but there is no significant increase in sales of their equipment.

ARM has essentially dominated the smartphone space (more) and made some gains in the SBC space (less). However, this entire "ecosystem" is highly fragmented and diverse. And if it weren't for Linux and Android, few would be interested. Microsoft's moves with the ARM version of Windows are more of a precaution and a statement, something like: Hey nerds, we also have Windows for "modern architectures". For now, it's hard to consider ARM-based SoC "computers" as standard everyday tools.

RISC-V is still a novelty, even in the world of microcontrollers. It doesn't look like it will see practical use in the world of PCs or servers. Although I personally support RISC-V because ARM is "too corporate". RISC-V itself is almost exclusively of interest to Chinese semiconductor companies, which is not surprising, because thanks to this they are not dependent on ARM licenses. And yet, few of them currently deliver MCUs based on the RISC-V architecture. All other cases are just curiosities for now. Outside of highly experimental versions of Linux, there's no software for them.

[creaothceann]

I consider that the time of x86 is coming to the end. x86 not wining nor in speed, nor in energy efficiency.
And instruction set of x86 is kind of bloated.

What will be the next "big" architecture? Will it be ARM or RISC-V?

x86's big advantage is backwards-compatibility. (Just ask Intel.) The office and home user markets don't really need faster CPUs, AMD is even bringing old CPUs back. Improvements in price and thermals are of course welcome though. For users of mobile PCs, ARM seems promising due to it being more efficient.

Program code is usually decoded into micro-ops (μops) that are stored in the instruction cache. The size of the user-visible opcodes doesn't matter much, as long as that cache is big enough.

ARM has essentially dominated the smartphone space (more) and made some gains in the SBC space (less). However, this entire "ecosystem" is highly fragmented and diverse. And if it weren't for Linux and Android, few would be interested.

There are users though who would be fine with just a browser and the programs/tools bundled with the OS.

[VisualLab]

ARM has essentially dominated the smartphone space (more) and made some gains in the SBC space (less). However, this entire "ecosystem" is highly fragmented and diverse. And if it weren't for Linux and Android, few would be interested.

There are users though who would be fine with just a browser and the programs/tools bundled with the OS.

A most valid comment. I would add that these are mostly users who:

• have never owned a personal computer (desktop or laptop) – these are the older ones, and
• would never buy a personal computer (desktop or laptop) – these are the younger ones.

This is a broad group consisting of subgroups of people who:

• they were never interested in computers - for them, computers and software were (are) boring boxes for even more boring activities and they did not see any use in them for themselves,
• they were unable to learn how to use computers and software - i.e., they are overwhelmed or terrified by them, and they consider their use too complicated and excessively confusing.

You are absolutely right that for such people, smartphones or tablets with installed and configured software limited to the operating system and web browser are the best solution. They only use access to resources on the Internet as if it were an advanced TV with greater possibilities. Some of them like to use the TV as a large screen for their smartphone. They buy and treat a smartphone or tablet the same way they treat a washing machine, shoes, coat or food. Classic consumers without excessive demands. And perhaps to some extent they are right. In any case, there are most of them. The problem is that for such customers, producers (corporations) will use the cheapest methods of software production, which affects its quality. But which corporate CEO cares these days? After all, none of them knows anything about the structure and operation of computers and software. He has technical department employees for that. The days when company CEOs had any understanding of hardware and software are long gone (the 1960s, 1970s, and 1980s). And this probably didn't apply to all companies, for example Bill Gates or Steve Jobs, who were total amateurs, laymen with no education. They were just computer enthusiasts (and even then only for a while). Today, even that seems to be missing among the owners of computer and software corporations. They only have enthusiasm for "big and fast" money, because thanks to it they can organize some closed island, some Epsteins and pathological parties.

[creaothceann]

for example Bill Gates or Steve Jobs, who were total amateurs, laymen with no education

Well, there are three types:
- the Ray Kassar type who thinks that "anybody can do [software]"
- the Steve Jobs type who knows how to sell a product
- the Bill Gates type who is the exception to the rule.

[marcov]

There are users though who would be fine with just a browser and the programs/tools bundled with the OS.

But those are also the users who use their systems the longest. And their numbers have decreased because of migration to chromebooks, tablets and telephones.

The fact that Chromebooks in 5+ years doesn't seem to really grow a hold beyond the educational sector says enough IMHO.

[marcov]

Program code is usually decoded into micro-ops (μops) that are stored in the instruction cache. The size of the user-visible opcodes doesn't matter much, as long as that cache is big enough.

As far as I know, the instruction cache contains instructions (cache lines) that mirror memory. Only the uop cache contains decoded instructions (in the magnitude of a few thousand)

[mas steindorff]

I'm a small low/no power CPU man myself but this video opened my eyes on the current high powered CPU + DRAM combos and how the refresh cycle is slowing down stuff at the lowest level. 
https://www.youtube.com/watch?v=KKbgulTp3FE
Laurie's 2 core fix is an eye opener and will likely become part of the hardware someday.  This allows systems use to keep using big cheep DRAM sticks which, IMHO will drive the next "big" archtecture
MAS