Except that we're really running AMD's x64 ISA, not Intel's, so this moat concept doesn't really fit with the reality. Intel clearly thought x86s days were over when they started their 64 bit Itanium effort, and then quickly (and luckily) managed to get back into having a competitive x64 implementation when they realised they had backed the looser. I'm actually quite impressed with how they pivoted as lesser companies would have had trouble ditching that sort of investment.
That can't be the case, as the Zhaoxin CPUs have the same kinds of instructions as similar Intel chips, including those cross-licensed from AMD, like the 64bit instruction set.
> In short, there’s no meaningful difference between RISC/ARM and CISC/x86 as far as performance is concerned.
Yet that usually doesn't seem to be the case in my experience, unless perhaps you don't care about power consumption. Then again, maybe the author doesn't consider frequent interrupts to be an issue of performance.
The article never gets to the punch line, which is behind a paywall.
It also fails to mention how thoroughly Intel wiped the floor with competitors in the "Workstation" space who had their own RISC flavors and mostly went out of business with the exception of IBM's PowerPC.
It takes massive investment to make top performing chips which makes it a lot unlike, say cars, where a company like Ferrari can successfully make high performance cars in small numbers. Intel won HPC because it won the Desktop, not the other way around -- one reason why I am still irked when analysts talk about Intel being driven by HPC and hyperscalers while most of their sales still go to the client division. (If there's a reason to pander to hyperscalers it is that they're in a better position to move to another architectures.)
The first force to challenge Intel has been ARM and it is not because ARM is fundamentally a better ISA but because ARM parts for phones are produced in such huge numbers, ARM achieved economies of scale that Intel never did.
If I was going to point to a real weakness in Intel it is Microsoft, that is, the very slow evolution of the Windows platform such that Windows only started requiring the 2008 POPCOUNT instruction this year. It's not about POPCOUNT, it is about all of the new instructions added since then, which are a major part of claimed performance improvements. It's about a SIMD programming model that makes you rewrite your apps every 2 years, a model that delivers value to the national labs but not to the ordinary user. Apple retires older computers after a certain time in the sense of not providing new OS versions which means that newer OSes can be written to support 5 year computers, with Microsoft it is more like 15.
> "It also fails to mention how thoroughly Intel wiped the floor with competitors in the "Workstation" space who had their own RISC flavors and mostly went out of business with the exception of IBM's PowerPC."
Intel's own product in this space was Itanium which also went out of business.
Itanium though had fundamental problems with how it was built and couldn't have been saved with any amount of investment; explicit parallelism just doesn't work for ordinary kinds of code because the compiler never knows which data is in the cache except for special applications like DSP.
Itanium certainly seduced a number of those workstation vendors to give up on architectures that could have had a future and invest in one that didn't.
I think we are just now starting to have the software maturity to handle something like Itanium. We have toolchains with data flow analysis, languages other than C which could hint the instruction scheduler, we have profiling, we can gather profiling from not just a developers workbench but with instrumentation and we have Machine Learning to eke out patterns. But back then, Intel vastly overestimated the capabilities of the software world to take their tech and run with it.
The problem is, this comment gets straight to the point. You’ll never get people to sign up for your Substack if you just give it all away in a couple straightforward paragraphs!
> ARM achieved economies of scale that Intel never did
I'm not sure how does that work. ARM doesn't produce their own chips they are licensing their code designs to 3rd parties which funds their R&D. Supposedly their royalty is around 1-2% per chip which is peanuts compared to what Intel is making (most ARM chips are very low cost and even the high-end ones like Snapdragon 8 Gen 3 are only ~$200). Apple and now Qualcomm and Ampere are developing their own chips these days but that's not really benefiting ARM directly in any way...
There's a huge gap between different ARM chip makers. Apple makes chips for laptops and desktops which are really competitive, Qualcomm doesn't, despite trying for a decade.
The phone market is so big, however, that it can support multiple manufacturers which have their own IP.
Snapdragon X Elite is supposed to be very competitive (and possibly or par with Apple's chips) of course we'll see.
> The phone market is so big
Perhaps, but I don't see how this changed anything. Most chips are low cost, low margin and ARM is barely getting anything from then anyway. IIRC ARM is the only company that has competitive mobile cores besides Apple and they can't come even remotely close to Intel's R&D budget.
> This has always not always been impossible, as AMD, NEC, Via, and others, have done so in the past
Was VIA ever competitive performance-wise with Intel or AMD? They’ve always been a weird bit of trivia, for my whole life… and I’m not so old but I’m not very young.
IIRC there was a old computer history blog talking about the various processors, and there were specific times when the Cyrix and VIA were competitive (including price/performance) - if you were talking about processors that were actually available to buy and use.
AMD/Intel were always winning on paper but there were times when a CPU would be launched and simply unavailable.
If Intel's moat made them fall behind, it was by lulling the company into a false sense of security allowing the bean counters to take over and under-invest in keeping up with everyone else in the all-important foundry space.
Maybe that's the thesis, but if so it misattributes the cause. The cause is the same MBA-brain thinking that destroyed General Electric and Boeing.
The root error in this thinking is to get behind a spreadsheet and completely abstract away all human and physical realities. There's now just units, COGS, gross margin, etc., and so you just optimize those levers to maximize the KPIs the stock market cares about with zero regard for the effect it has on the actual machine or its long term trajectory in the space it operates within.
It's really the same set of fallacies that lie behind command economics at the national scale. MBA-brain doesn't work for the same reason centralized authoritarian communism doesn't work. It's the same abstraction away of details, the same optimization without regard for reality, the same hollow experts in "leadership" that aren't really experts in anything. Apparatchiks with no understanding of what's actually happening on the ground can't run things.
29 comments
[ 0.18 ms ] story [ 75.0 ms ] threadYet that usually doesn't seem to be the case in my experience, unless perhaps you don't care about power consumption. Then again, maybe the author doesn't consider frequent interrupts to be an issue of performance.
It also fails to mention how thoroughly Intel wiped the floor with competitors in the "Workstation" space who had their own RISC flavors and mostly went out of business with the exception of IBM's PowerPC.
It takes massive investment to make top performing chips which makes it a lot unlike, say cars, where a company like Ferrari can successfully make high performance cars in small numbers. Intel won HPC because it won the Desktop, not the other way around -- one reason why I am still irked when analysts talk about Intel being driven by HPC and hyperscalers while most of their sales still go to the client division. (If there's a reason to pander to hyperscalers it is that they're in a better position to move to another architectures.)
The first force to challenge Intel has been ARM and it is not because ARM is fundamentally a better ISA but because ARM parts for phones are produced in such huge numbers, ARM achieved economies of scale that Intel never did.
If I was going to point to a real weakness in Intel it is Microsoft, that is, the very slow evolution of the Windows platform such that Windows only started requiring the 2008 POPCOUNT instruction this year. It's not about POPCOUNT, it is about all of the new instructions added since then, which are a major part of claimed performance improvements. It's about a SIMD programming model that makes you rewrite your apps every 2 years, a model that delivers value to the national labs but not to the ordinary user. Apple retires older computers after a certain time in the sense of not providing new OS versions which means that newer OSes can be written to support 5 year computers, with Microsoft it is more like 15.
Intel's own product in this space was Itanium which also went out of business.
In the 1980s they thought the https://en.wikipedia.org/wiki/Intel_iAPX_432, https://en.wikipedia.org/wiki/Intel_i960 and https://en.wikipedia.org/wiki/Intel_i860 were the future. This chip, https://en.wikipedia.org/wiki/Intel_80286 which could have been vastly more useful than it was with tiny changes, was a big hit instead.
Itanium though had fundamental problems with how it was built and couldn't have been saved with any amount of investment; explicit parallelism just doesn't work for ordinary kinds of code because the compiler never knows which data is in the cache except for special applications like DSP.
Itanium certainly seduced a number of those workstation vendors to give up on architectures that could have had a future and invest in one that didn't.
I'm not sure how does that work. ARM doesn't produce their own chips they are licensing their code designs to 3rd parties which funds their R&D. Supposedly their royalty is around 1-2% per chip which is peanuts compared to what Intel is making (most ARM chips are very low cost and even the high-end ones like Snapdragon 8 Gen 3 are only ~$200). Apple and now Qualcomm and Ampere are developing their own chips these days but that's not really benefiting ARM directly in any way...
The phone market is so big, however, that it can support multiple manufacturers which have their own IP.
Snapdragon X Elite is supposed to be very competitive (and possibly or par with Apple's chips) of course we'll see.
> The phone market is so big
Perhaps, but I don't see how this changed anything. Most chips are low cost, low margin and ARM is barely getting anything from then anyway. IIRC ARM is the only company that has competitive mobile cores besides Apple and they can't come even remotely close to Intel's R&D budget.
Was VIA ever competitive performance-wise with Intel or AMD? They’ve always been a weird bit of trivia, for my whole life… and I’m not so old but I’m not very young.
AMD/Intel were always winning on paper but there were times when a CPU would be launched and simply unavailable.
VIA (via Centuar, the CPU design firm), focused on low-power, lower-performance chips. They only used Cyrix for the name, and not for very long.
Maybe that's the thesis, but if so it misattributes the cause. The cause is the same MBA-brain thinking that destroyed General Electric and Boeing.
The root error in this thinking is to get behind a spreadsheet and completely abstract away all human and physical realities. There's now just units, COGS, gross margin, etc., and so you just optimize those levers to maximize the KPIs the stock market cares about with zero regard for the effect it has on the actual machine or its long term trajectory in the space it operates within.
It's really the same set of fallacies that lie behind command economics at the national scale. MBA-brain doesn't work for the same reason centralized authoritarian communism doesn't work. It's the same abstraction away of details, the same optimization without regard for reality, the same hollow experts in "leadership" that aren't really experts in anything. Apparatchiks with no understanding of what's actually happening on the ground can't run things.