Nvidia's Plan for ARM to Take X86 's Throne | Exclusive

member when a 2 TRILLION dollar company switched all their products to run on any of those above....

i'll wait
Proportionally, I suspect a couple of those might be close. Especially PowerPC; but I ain’t doing the math on it.
Company valuation aside, Apple is still a small part of the compute pie in the world. Lots of profit and revenue, but small part of the overall market.
 
I just don't see that happening right now though. Microsoft is slowly moving closer and closer to making desktop Windows free, and the mind share that would be required to get people move away from it is staggering. Apple has been trying since...before Windows. We've had OS/2 Warp, Mac OS, BeOS, a half dozen really good Linux distros, and most recently Chrome OS. We've even had Steam OS for gaming. Getting past that mind share is going to take a monumental shift in society itself right now. I can only think of two companies that stand a chance in pulling that market share and they are Google and Apple. At least for the US and European markets.
If anything is going to drive Linux forward on the desktop, then it'll be Microsoft. Not Microsoft adopting Linux but Microsoft screwing up Windows. If Windows does become free then you'll want to jump on Linux ASAP. What I expect Microsoft to do is start charging for certain parts of the OS, like DX12, DX Ray-Tracing, and etc. Things that won't prevent you from running Windows but are things that enthusiasts would want to get the most of their experience. Microsoft already does this with Dolby Atmos & DTS Sound Unbound, which they advertise when you right click the speaker icon and click on spacial sound. Sure these are Dolby properties but Microsoft is advertising it built into the OS.
I do think though that desktop computer may evolve to just become docking stations for phones and tablets. In that case, aside from artistic professional work, high security/scientific work, and games (for now), we might see Windows as a primary OS falter. Microsoft needed to really commit to the phone market, and they Zuned it.
That'll never happen for a number of reasons. Firstly because iOS and Android both suck. Too locked down for any serious work to be done, and doesn't receive updates as often as Windows, Linux, and Mac OSX. Also we're going to run into limitations in how small we can shrink the chip die, in which the only way to get faster chips is to feed them more power and have them get hotter. So battery power is also another factor.

Microsoft already tried to get into the phone market and failed. Unless Microsoft has any plans to go open source, they don't have a chance to break two very established mobile markets.
 
Fair enough. Still not sure China factors in at all.
Chinas banking regulators get the same say that the US banking regulators get to say in it. They also need approval from the EU and UK regulatory bodies as well.
 
DA059ED7-48B0-4C13-AB68-633DAAC73E0C.jpeg
 
If anything is going to drive Linux forward on the desktop, then it'll be Microsoft. Not Microsoft adopting Linux but Microsoft screwing up Windows. If Windows does become free then you'll want to jump on Linux ASAP. What I expect Microsoft to do is start charging for certain parts of the OS, like DX12, DX Ray-Tracing, and etc. Things that won't prevent you from running Windows but are things that enthusiasts would want to get the most of their experience. Microsoft already does this with Dolby Atmos & DTS Sound Unbound, which they advertise when you right click the speaker icon and click on spacial sound. Sure these are Dolby properties but Microsoft is advertising it built into the OS.

That'll never happen for a number of reasons. Firstly because iOS and Android both suck. Too locked down for any serious work to be done, and doesn't receive updates as often as Windows, Linux, and Mac OSX. Also we're going to run into limitations in how small we can shrink the chip die, in which the only way to get faster chips is to feed them more power and have them get hotter. So battery power is also another factor.

Microsoft already tried to get into the phone market and failed. Unless Microsoft has any plans to go open source, they don't have a chance to break two very established mobile markets.
With Microsoft’s gradual integration of more and more Linux commands inside of the windows Powershell I can actually see a time in the future where a full Linux kernel is running alongside windows.
 
I don't think Apple really leads the PC market, but I'll wait.
Apple is #4 in PC shipments. They may not be leading... however they grew market share for the last 2-3 years with crappy Intel hardware. MacOS is back somewhere in the 12-15% frame depending whos numbers you go by. No doubt they are going to grow it even further now that M1 is out... it may not be most [H] members cup of tea. However for the most part the mini and the air are getting rave reviews... and they seem to be selling extremely well so far. It will be interesting to see the next quarters sales numbers.
 
Apple was a founder company for ARM

As a part of selling away their shares (to softbank?), Apple made sure that they had a perpetual license to ARM

I was unaware. That's pretty cool. I want there to be more of a CPU market, and I had Macs before with PPC 601, 604, G3, and G4. I'm not against ARM.
 
With Microsoft’s gradual integration of more and more Linux commands inside of the windows Powershell I can actually see a time in the future where a full Linux kernel is running alongside windows.
If Microsoft integrates the Linux Kernel, it won't change how the OS functions. It'll be done to cut costs, as the Linux Kernel already takes care of a lot of hardware and security patches. It'll be modified heavily to fit Microsoft's needs as well. As a Windows user you probably won't even notice the change, but Microsoft would as they don't need as many people to maintain it.

Apple is #4 in PC shipments.
Yea, but that's not as impressive as it sounds. Gotta remember there's a lot of PC manufacturers, so their 4th place is only 6.7% market share. The other 93.3% are running x86 with Windows.
gartnerpcq2.png
They may not be leading... however they grew market share for the last 2-3 years with crappy Intel hardware.
There's a reason for that. Things like boot camp and Parallels has certainly made using MacOSX easier. The new ARM CPU's probably won't do Boot Camp, but I'm sure Parallels will update to be compatible.

MacOS is back somewhere in the 12-15% frame depending whos numbers you go by.
About 8.62%, while iOS is 14.36%. If you remove the mobile market then it's 17.65%. For the past 10 years OSX has certainly gained in popularity but that might change with the inability to play games on their ARM based Macs. Don't underestimate the power that gaming has over OS market.
No doubt they are going to grow it even further now that M1 is out...
Most likely not. If anything Apple is going through a transition period, which means they are going to have a hard time convincing people to buy them. Why you think Apple made anything based on the M1 cheaper than Intel products? They know there's going to be issues with people using products based on M1.
However for the most part the mini and the air are getting rave reviews... and they seem to be selling extremely well so far. It will be interesting to see the next quarters sales numbers.
Regardless of how well the M1 based Macs are received, it probably wouldn't matter. Black Friday was actually black for retailers. We are going through an economic situation here. Just very bad timing for Apple.
 
@ DukeNukemX

Would integrating the Linux kernel cause MS to have to go open source on other parts of the OS?
No Linux is a kernel... not a OS. All the other bits that make up a GUI ect can be open or closed or in between.
Nothing stopping MS from benefiting from the Linux communities work on the kernel... file systems ect. While running their own GUI or any other closed bits they like on top.
 
If Apple can make the M1 do x86 efficiently then it's obvious that AMD and Intel could do the same thing. It's still emulation and it is 2/3 the performance of native ARM apps, but it can obviously be done. Remember that modern x86 CPU's are just RISC with a CISC to RISC converter. Also every time someone says how bad and old x86, then suddenly we have new x86 CPU's that dominate in performance.

It is not obvious. This is not RISC vs CISC. This is ARM64 versus x64. The advantages of ARM64 aren't because it is RISC, but because it is a modern and clean ISA.

x86 lost the performance crown many time ago in servers. x86 is only the king in desktop, because this is a niche market and no ARM server player is releasing a 4GHz 32 core chip for desktops.

The problem is that Intel dominated the market for too long and they got lazy about increasing performance. There's a reason why 2500K's and 2600K's are still good CPU's to use because the amount of improvements we've seen since 2012 has been minor. Also why AMD's new Ryzen 5000 series are not good deals over h 3000 series. The improvements are just not that much.

x86 doesn't scale well. That is the reason why Intel tried to abandon it when switched to 64bits. It was a disaster, as all of us know.

What Apple did with the M1 was catch up to Intel in performance, but beat them in power efficiency hands down.

Apple has catched and surpassed AMD as well. AMD needs a Zen3 core and ~7 times more power to keep with a Firestorm core in single thread.

With Nvidia I expect them to destroy Intel in performance and power consumption.

Zeus cores already destroy any x86 core. I only expect Nvidia to speed up the ARM roadmap, and integrate with CUDA/Graphics. I expect Nvidia to sell complete servers and HPC clusters.
 
It is not obvious. This is not RISC vs CISC. This is ARM64 versus x64. The advantages of ARM64 aren't because it is RISC, but because it is a modern and clean ISA.
So was PowerPC, and yet Apple dumped the cleaner and modern PPC for x86. We've been through this before. Apple used to proclaim victory over Intel with benchmarks to show. Then Intel dumped their Netburst architecture for CoreDuo because AMD was riding their ass with the Athlon 64. This should sound familiar now with AMD yet again riding Intel's ass with Ryzen. Intel CPU's today are just Sandy Bridge with a lot of tweaks. Plus their manufacturing is shit now. Intel is due for a massive change up.
x86 lost the performance crown many time ago in servers. x86 is only the king in desktop, because this is a niche market and no ARM server player is releasing a 4GHz 32 core chip for desktops.
If x86 lost the server crown then why is x86 still #1 used for servers?
x86 doesn't scale well. That is the reason why Intel tried to abandon it when switched to 64bits. It was a disaster, as all of us know.
Itanium or IA-64 proved that just because you have a new architecture doesn't mean you'll have better performance. Also PowerPC. Also VLIW wasn't a great idea for a CPU. SIMD won in the end.
Apple has catched and surpassed AMD as well. AMD needs a Zen3 core and ~7 times more power to keep with a Firestorm core in single thread.
Clearly Apple has won in power consumption but not overall performance. Going by Linus Tech Tips benchmarks... no the M1 isn't that fast. They actually ran benchmarks that weren't synthetic. Well... not mostly. Rosetta performance is 2/3's to 3/4's the performance of native ARM apps. I do kinda expect Intel to sue Apple for integrating some x86 in the M1.

Zeus cores already destroy any x86 core.
Link benchmarks please.
I only expect Nvidia to speed up the ARM roadmap, and integrate with CUDA/Graphics. I expect Nvidia to sell complete servers and HPC clusters.
Speed up ARM yes, but integrate CUDA... not so much. Mali graphics is doing pretty well, but I doubt Nvidia would license out their tech the same way ARM does. Though if I were Apple I'd be afraid of what Nvidia does with ARM from now on.
 
It is not obvious. This is not RISC vs CISC. This is ARM64 versus x64. The advantages of ARM64 aren't because it is RISC, but because it is a modern and clean ISA.

x86 lost the performance crown many time ago in servers. x86 is only the king in desktop, because this is a niche market and no ARM server player is releasing a 4GHz 32 core chip for desktops.
The ARM ISA was developed in 1983. Sure, it's ~5 years newer than the original x86 processor, but we're not talking about something ~modern~ here - ARM has been around for a very long time, and has (at various points) taken sales from x86, and vice versa. It's back now because the x86 landscape has hit a wall they're not sure how to solve, which has happened several times in history (all things are cyclical) - eventually they'll break through. And eventually ARM will hit a bottleneck that they have to solve too - it happens for everyone. Hell, Intel has MADE ARM cpus in the past - as has AMD. There's a reason they abandoned those efforts at the time, and while it might again come back (who better to make an x86 compatible ARM cpu than the guys who own the x86 ISAs?), we just don't know - and all things are cyclical.

As for the server space - ARM has a few supercomputers, which are a completely different world (in 2005, we thought that a similar design using Power would be the king in the future - Blue Gene/L - it got beat by GPUs a few short years later). Things move absurdly fast in that space, and are often designed for very specific use cases. A full-fat interconnect like that has advantages and disadvantages against other HPC designs (GPU/hadoop style/disaggregated/etc), depending on your use case. They also exist in the cloud space for things designed for it - but the enterprise server world still belongs to x86, which means the consumer server and desktop world still belongs to x86, as does the majority of cloud servers (still). That may change, sure - but right now ARM hasn't "won" that space - the majority of revenue is still going to Intel and AMD.
x86 doesn't scale well. That is the reason why Intel tried to abandon it when switched to 64bits. It was a disaster, as all of us know.
Intel hit a wall with x86, yes. Itanium was an attempt to solve that wall - as was AMD64. One won, the other... did not. Much like ARM might win now. But I'll point out - ARM might win now, as AMD64 beat Itanium, but what happened after those first few Athlon64 processors?

Bulldozer.

Cyclical, all things are.
Apple has catched and surpassed AMD as well. AMD needs a Zen3 core and ~7 times more power to keep with a Firestorm core in single thread.



Zeus cores already destroy any x86 core. I only expect Nvidia to speed up the ARM roadmap, and integrate with CUDA/Graphics. I expect Nvidia to sell complete servers and HPC clusters.
Sure, for a single thread. Now show me a 16 core/32 thread ARM processor I can buy today for home/desktop/workstation use. With expansion slots. Gigabyte made a board for a bit - and abandoned it. HPE makes the Apollo systems, but they're really just HPC.

The thing ARM ran into with their attempts to break into server/high-end space 3-4 years ago was that when you add the same amount of cache/RAM/etc as a normal server, your power consumption isn't mostly from the CPU anymore - it's the accessories (powering 1.5TB+ of RAM takes a lot of juice), and your processor isn't saving THAT much over a comparable x86 one either (the nightmares of 14nm+++++++ aside). Now, they've sorta solved that from what I'm seeing - sorta - but I'm not getting any requests for ARM in the enterprise space yet, and the software vendors I work with are keeping an eye on it, but aren't really planning on building ARM-native yet either.

Yet.

It's coming - that's for sure - but all things are cyclical, and x86 certainly ain't done yet. It's an exciting time to be involved in the industry though!
 
member when MIPS was going to take the throne away from x86?

member when PowerPC was going to take the throne away from x86?

member when SPARC was going to take the throne away from x86?

member when Alpha was going to take the throne away from x86?

member when PA-RISC was going to take the throne away from x86?

member when RISC-V was going to take the throne away from x86?

7Mm31kE.jpg
 
So was PowerPC, and yet Apple dumped the cleaner and modern PPC for x86.

PowerPC is a bloated and inefficient ISA compared to ARM64.

We've been through this before. Apple used to proclaim victory over Intel with benchmarks to show. Then Intel dumped their Netburst architecture for CoreDuo because AMD was riding their ass with the Athlon 64. This should sound familiar now with AMD yet again riding Intel's ass with Ryzen. Intel CPU's today are just Sandy Bridge with a lot of tweaks. Plus their manufacturing is shit now. Intel is due for a massive change up.

Reasoning by analogy is dangerous. Both Netburst and Bulldozer families were speed-demon microarchitectures and both failed due to process nodes. It was only after Piledriver iteration and many years of maturing and tweaking the 32SOI node that AMD was barely able to hit the original 5GHz target (and it did at the expense of huge power consumption 220W and extreme binning). Intel was never able to hit the original 10GHz target for Netburst. So both microarchitectures were much slower and power hungry than expected. Intel revolution was a consequence of replacing a broken speed-demon microarchitecture by a smart brainiac microarchitecture. The same with AMD revolution. But now AMD and Intel are on well-designed brainiac microarchitecture and so you will not see again a Bulldozer-->Zen or a Netburst-->CoreDuo transition.

If x86 lost the server crown then why is x86 still #1 used for servers?

For a similar reason why Intel continues being #1 despite AMD has better server chips: "Inertia". It took a decade to x86 to conquer the server/HPC space.
 
The ARM ISA was developed in 1983. Sure, it's ~5 years newer than the original x86 processor, but we're not talking about something ~modern~ here - ARM has been around for a very long time

Note that I wrote ARM64. This is a new and separate ISA that was developed recently.

https://www.realworldtech.com/arm64/5/

Intel hit a wall with x86, yes. Itanium was an attempt to solve that wall - as was AMD64. One won, the other... did not. Much like ARM might win now. But I'll point out - ARM might win now, as AMD64 beat Itanium
Itanium was an attempt to solve the x86 wall by replacing superscalar CISC by VLIW+. AMD64 wasn't. It is just a naive 64bit extension to the original x86 ISA.
Sure, for a single thread. Now show me a 16 core/32 thread ARM processor I can buy today for home/desktop/workstation use. With expansion slots. Gigabyte made a board for a bit - and abandoned it. HPE makes the Apollo systems, but they're really just HPC.

Desktop is a niche market. So companies aren't interested in releasing ARM processors for the desktop. There are a pair of exceptions, but largely irrelevant. Moreover, we have the problem of the operative system. Whereas running linux isn't a problem for servers, it is for most desktop users.

The thing ARM ran into with their attempts to break into server/high-end space 3-4 years ago was that when you add the same amount of cache/RAM/etc as a normal server, your power consumption isn't mostly from the CPU anymore - it's the accessories (powering 1.5TB+ of RAM takes a lot of juice), and your processor isn't saving THAT much over a comparable x86 one either (the nightmares of 14nm+++++++ aside). Now, they've sorta solved that from what I'm seeing - sorta - but I'm not getting any requests for ARM in the enterprise space yet, and the software vendors I work with are keeping an eye on it, but aren't really planning on building ARM-native yet either.

And still again and again studies and reviews show that ARM-based servers have both efficiency and price advantages over equivalent x86 servers. And this is the reason why many companies are using ARM servers or reselling them to its customers (e.g. Amazon AWS).
 
jardows i don’t know that I would call it a “niche”, but while pc sales in genera have increased (including tablets and chrome books, big growth recently) desktop sales declined notably in q3 and have been on a general downward trend for years:

https://www.canalys.com/newsroom/worldwide-pc-market-Q3-2020

https://www.canalys.com/newsroom/canalys-worldwide-pc-market-q2-2020

Shipments of notebooks and mobile workstations grew 24% year on year, while shipments of desktops and desktop workstations fell by 26%.

1606829060627.jpeg


I think people shopping generally don’t want the hassle of a desktop PC. Monitor? Keyboard? Mouse? Tower? Webcam? Speakers? All these cables? I’ll just get a laptop.

(this is [H], but don’t forget the PC market is far larger than an enthusiast diy community).
 
Note that I wrote ARM64. This is a new and separate ISA that was developed recently.

https://www.realworldtech.com/arm64/5/
Um, no.
https://developer.arm.com/-/media/Files/pdf/graphics-and-multimedia/ARMv8_InstructionSetOverview.pdf
https://en.wikipedia.org/wiki/AArch64
Still part of ARM, just like AMD64 is still part of x86. It's an extension on the existing 32bit ARM architecture, which is an extension on the prior releases. Same jam, different berry.


Itanium was an attempt to solve the x86 wall by replacing superscalar CISC by VLIW+. AMD64 wasn't. It is just a naive 64bit extension to the original x86 ISA.
Both were attempts to solve a business problem: Scalability of the x86 ecosystem. Stop looking at the technical details and look at the outcome - outcomes beat technical every time, since outcomes have MONEY attached to them - unless you're somehow running FOSS software only on a FOSS hardware platform. Money talks.
Desktop is a niche market. So companies aren't interested in releasing ARM processors for the desktop. There are a pair of exceptions, but largely irrelevant. Moreover, we have the problem of the operative system. Whereas running linux isn't a problem for servers, it is for most desktop users.
Those processors were cancelled. https://www.servethehome.com/impact-of-marvell-thunderx3-general-purpose-skus-canceled/
Now, as for the rest - define niche market. Revenue? Profit? Cash flow? Investment amount? Shipments?

And it's a problem for servers too, depending on what perspective you take on the industry. Outside of Amazon and GCP, very few ARM servers are shipping - because software isn't written for them, support doesn't exist for them, and containers still have a year or two to go before they start to take away from traditional virtualization, which is still the de-facto method for deploying workloads that are ~not home grown~. (most DevOps shops are moving to a container architecture as they can, but enterprise software - EPIC/EMR/CRM/etc - generally is not, and is also not compiled for ARM). There's a HUGE long tail of software there that has to be supported, and as long as people are writing for the x86 platform, they'll be buying development systems for x86, and that will give Intel/AMD possibly the time to respond.

On top of that, not everything just "works" on ARM -
https://news.ycombinator.com/item?id=19224579

"Node and Ruby applications do fail on ARM though, when it comes to native libraries and extensions. And now your whole distro is different than your development machine, which adds complexity.
Do I really want to be debugging why node-gyp fails to compile scrypt on the ARM distro on the new Amazon A1 ARM instance (which it did in my case)? And if I solve that, what about the other 2451 dependencies? Let's pessimistically say there's a 1% failure rate, I'll be stuck doing that forever! Nah, I'll just go back to my comfy x86 instance, life's short and there's much code to write :)

I think I'll side with Linus on this one. I saw first-hand how non-existent the x86 Android market was, despite Intel pouring megabucks into the project. If the developers don't run the same platform, it's not going to happen, no matter how great the cross platform story is in theory. Even if it's as simple as checking a box during upload that "yes, this game can run on X86", a huge chunk of the developers will simply never do that."

and:

I'm suffering through this now -- I have a custom C++ Node exception that needs to run on both x64 and ARM. The ARM cpu is onboard a mobile robot, where I care about power draw.
The good news is that Clang can cross-compile fairly easily. Much better than gcc.

The bad news is that there are a surprising number of missing libraries on Ubuntu/ARM64. For example, Eigen3. And although the code is fairly compatible, there's some extra cognitive load in learning to debug on ARM. For example, calling a virtual function on a deleted object crashes differently.

I'm willing to put up with it for ARM's advantages in battery-powered applications, but I wouldn't just to save a few bucks on cloud servers.

There are plenty of other examples on there. Having been a developer in the past, writing on a different ISA than what you were running on (writing on PowerPC, testing on x86, deploying to SPARC) was... interesting, even IF you weren't writing low-level code. I happen to agree with one of the guys from that thread - either M1 will enable the ARM ecosystem development of desktop / prosumer / power apps, or it will reduce Apple's market share in the long run. We'll see.
And still again and again studies and reviews show that ARM-based servers have both efficiency and price advantages over equivalent x86 servers. And this is the reason why many companies are using ARM servers or reselling them to its customers (e.g. Amazon AWS).

My experience says that when you have configurations like those, the RAM (especially when you whack up over 1TB) takes the majority of the power, so CPU efficiency is pretty much a wash and ignored. On top of that, server grade ARM chips aren't notably any more efficient than x86:
https://www.anandtech.com/show/12694/assessing-cavium-thunderx2-arm-server-reality/2 - look at the TDP values there. Cache/cores/etc cost power, regardless of which architecture you built them on. Note that the conclusion is that it performs well, in general, is cheaper to buy, but NOT more power efficient (except node efficiencies on the Qualcomm chip - which never really entered the server market). Hyper-scalars like Amazon are drastically different than "companies buying ARM servers" - when you buy an instance, it's not the same as buying a server (if we're comparing instances, we need to also look at the market for BSD Jails/Solaris zones and the like, as well as the x86 instance market). I'd say comparing actual sold chips/servers is a more accurate market for this debate. Heck, you can't really BUY an ARM server right now - most of the general purpose chips were canned for specialized silicon (see my later point on this).

Also:
https://securityboulevard.com/2019/12/this-is-finally-the-year-of-the-arm-server/
"
None of this is true. There are plenty of RISC alternatives to Intel, like SPARC, POWER, and MIPS, and none of them ended up having a power efficiency advantage.

Mobile chips aren’t actually power efficient. Yes, they consume less power, but because they are slower. ARM’s mobile chips have roughly the same computations-per-watt as Intel chips. When you scale them up to the same amount of computations as Intel’s server chips, they end up consuming just as much power."

Now, he also makes a good point - as we move to cloud-based models, custom silicon will become more important, which is where ARM (as a licensed ISA) is going to make inroads - but we're also going to need to see usable systems for developers/etc to use for that, which expands to the home (as my original point about x86 was - that's part of why x86 won initially), and we'll see if the market is interested in that or not. I don't honestly know, but my gut says that this, like most cyclical things in the industry, will grow, then shrink, then grow, and so on.

It's not as simple as you're trying to make it. The business side is important (money!), the developer side is important (software), and the home user (mix of both) is important, as is the hyper-scalar and the enterprise / commercial computing space. This is a long road for changes to happen- things don't happen that quick.
 
jardows i don’t know that I would call it a “niche”, but while pc sales in genera have increased (including tablets and chrome books, big growth recently) desktop sales declined notably in q3 and have been on a general downward trend for years:

https://www.canalys.com/newsroom/worldwide-pc-market-Q3-2020

https://www.canalys.com/newsroom/canalys-worldwide-pc-market-q2-2020

Shipments of notebooks and mobile workstations grew 24% year on year, while shipments of desktops and desktop workstations fell by 26%.

View attachment 304620

I think people shopping generally don’t want the hassle of a desktop PC. Monitor? Keyboard? Mouse? Tower? Webcam? Speakers? All these cables? I’ll just get a laptop.

(this is [H], but don’t forget the PC market is far larger than an enthusiast diy community).

This year is a BIG HUGE Ass*terisk.... chrome book sales are up cause millions of kids are schooling at home. PC sales are up cause millions of people are working from home. This bump is temporary. However as you point out Desktop sales are still down. Kids and people working from home just don't need a ton of local power anymore... they need a good internet connection.
 
This year is a BIG HUGE Ass*terisk.... chrome book sales are up cause millions of kids are schooling at home. PC sales are up cause millions of people are working from home. This bump is temporary. However as you point out Desktop sales are still down. Kids and people working from home just don't need a ton of local power anymore... they need a good internet connection.
And a lot of the actual workload is offloaded to the "cloud" - which is heavily running on x86. It's cyclical - we're back to the "mainframe and dumb terminal" days again, which isn't necessarily a bad thing - it's just a thing. And a chromebook/low power laptop works fine, if you have your stuff in the cloud and don't need local data manipulation (or ultra-high quality control). VDI helps with a lot of it too - except for the ultra high quality again (h.264 real time encoding is great, but sacrifices quality for bandwidth and CPU utilization).
 
jardows i don’t know that I would call it a “niche”, but while pc sales in genera have increased (including tablets and chrome books, big growth recently) desktop sales declined notably in q3 and have been on a general downward trend for years:
A declining market is a far cry from "niche." Even with the decline of "big box" desktop shipments in favor of laptop computers, ARM still has a long way to go to prove itself as a viable x86-64 replacement. Laptops are replacing the traditional "desktop" computer, but being used for the same workloads. Anyone who has the opportunity to do work on a decent desktop compared to a laptop will quickly realize the advantages of a desktop computer, so the demand will not decline to niche status until we have a dramatic shift in technology (advances that make the form factor and cooling needs irrelevant) or in the way that computers are used in the workplace. For the context of the conversation, even if the Apple M1 proves in real-world workloads to be a decent competitor, it will remain an Apple exclusive, and there is nothing that says other ARM chip manufacturers will be able to mirror Apple's success. Nvidia has had great success with GPU's, but have lacked serious competition. Just like Intel has often failed when branching out beyond what they "do best" nVidia will have struggles (and based on their similar history will likely fail) breaking into a new market that will attempt to displace extremely entrenched technology.
 
ChadD jardows As I said, I wouldn't call it "niche". But is IS a shrinking segment of the market.

The data shows that desktop PC sales have been declining going back to at least Q2 2017 - I wasn't looking for anything further back than that. This year has definitely been different, but it has been a whole year, and I don't think desktop systems will suddenly take off again in a year or two. If people are finding they're happy with inexpensive easily portable Chromebooks now, for instance, I don't see them suddenly switching back to desktops.

I was just pointing out that it's not as important a segment of the market as it was, say, 10 years ago.

It's not really relevant to this conversation, either, I think, I just felt the need to point it out.
 
  • Like
Reactions: ChadD
like this
If people are finding they're happy with inexpensive easily portable Chromebooks now, for instance, I don't see them suddenly switching back to desktops.
In my field, people are finding quite the opposite. Lots of people bought Chromebooks and have had to also by desktops or real laptops as the Chromebooks have been unable to run the needed software or access cloud resources outside of Google (which we don't use). It made the Chromebook makers money twice (once on the Chromebook and then on the desktop/laptop to replace them) though which is good, I guess.
 
In my field, people are finding quite the opposite. Lots of people bought Chromebooks and have had to also by desktops or real laptops as the Chromebooks have been unable to run the needed software or access cloud resources outside of Google (which we don't use). It made the Chromebook makers money twice (once on the Chromebook and then on the desktop/laptop to replace them) though which is good, I guess.
The question going forward though is do mass market consumers (including the majority working at home) need a ton of horsepower... or do they just need "responsiveness". That is something ARM does better... or at least most products built on ARM seem to. That may come down more to software... as the short amount of time I spent with a Windows ARM machine I also found it very "responsive".... as long as it was running proper ARM software, and not trying to emulate. I can understand chromebooks probably don't work for every industry... the ones they work in though they tend to work well and be very little admin hassle. It will be interesting to see how Apple does this quarter with their first ARM machines... I believe (could be wrong) that they are selling very well and are probably going to be a big shock to many doubters. I wouldn't be shocked to see Apple hit 20% of the market by next quarter.

I just hope whatever Nvidia (or another ARM players like AMD/Samsung/Qcom) cook up for MS is worthy when compared to the second gen Apple chips next year... and of course that MS is capable of unfucking its Windows ARM version. If not I fear x86 will become niche pretty damn fast cause frankly for most people that still need to consider tech support options as part of their purchasing decisions those Apple laptops with ARM chips are looking damn near perfect. It isn't exactly easy to argue they should go windows/x86 anymore when Apple is building such compelling ARM stuff. I admit for a few min I considered buying a ARM Air.... ahhh. lol
 
ChadD I did too, heh. I'm an Ubuntu guy and I've been interested in ARM procs for general purpose laptop/desktop-style computing for a while. Every few years I try to downsize a bit further, but I found once I got too low power (like the Intel NUC7PJYH, 4-core Pentium J5005 system) I started to really feel the lack of "responsiveness" as you put it. I figured an ARM based system, with the right SoC, could provide better performance than that in a similar power envelope, and maybe if it had decent I/O (SATA SSDs instead of eMMC or SD cards like a lot of SBCs, for instance) it would be that perfectly silent system that doesn't feel slow.

All the options so far have been low power SBCs in a DIY setup, hindered by slow storage (rpi), a small community (everything else), or not-really-made-for-what-you're-trying-to-do, too expensive and requires a ton of effort to get it working properly (nvidia tegra boards). TBH I am surprised we haven't seen a tiny desktop system or more laptops using Snapdragon 2019 / 2020 processors. They may not be as fast as Apple's M1 but... I don't think anyone would really notice. I know there are at least a few laptops out there, some fairly big name (like the recent surface).

The idea of a system like the M1 Air or Mini is very appealing to me, and I'm on the fence as to whether I'd be happy just running Mac OS on it and calling it a day. But I think I'm really hoping it gets ARM enough mindshare that other companies decide to take a chance on it and make some premium products that I'll be able to run my Ubuntu on. :)
 
Last edited by a moderator:
  • Like
Reactions: ChadD
like this
Um, no.
https://developer.arm.com/-/media/Files/pdf/graphics-and-multimedia/ARMv8_InstructionSetOverview.pdf
https://en.wikipedia.org/wiki/AArch64
Still part of ARM, just like AMD64 is still part of x86. It's an extension on the existing 32bit ARM architecture, which is an extension on the prior releases. Same jam, different berry.

No. AMD64 is an extension to x86. ARM64 is not an extenstion to ARM32, but a new separate ISA . ARM engineers used the transition to 64 bits for designing a new ISA and they have eliminated all the weird stuff used in the former ISAs. Details in the link I posted above.


"Marvell ThunderX3 Gets Canceled for General Purpose Sockets". Marwell is pivoting to custom designs, because it cannot compete with the new generic ARM cores. Ampere saw this before, cancelled its custom designs and migrated to Neoverse.

Moreover, TX2 got momentum in the HPC space, but the TX3, despite being a huge improvement over the TX2, cannot compete with Fujitsu A64fx. Originally the A64fx was going to be used only in Japan, but Cray will be selling them in Europe and USA

https://www.nextplatform.com/2019/11/13/a64fx-arm-chip-gets-a-big-push-from-cray/

On top of that, not everything just "works" on ARM -
https://news.ycombinator.com/item?id=19224579

Sure. Not everything works still. STH has a graph with the status of the ARM ecosystem, from what works out of the box to what don't.

My experience says that when you have configurations like those, the RAM (especially when you whack up over 1TB) takes the majority of the power, so CPU efficiency is pretty much a wash and ignored. On top of that, server grade ARM chips aren't notably any more efficient than x86:
https://www.anandtech.com/show/12694/assessing-cavium-thunderx2-arm-server-reality/2 - look at the TDP values there. Cache/cores/etc cost power, regardless of which architecture you built them on. Note that the conclusion is that it performs well, in general, is cheaper to buy, but NOT more power efficient (except node efficiencies on the Qualcomm chip - which never really entered the server market).

Notice that TX2 used a less efficient 16nm process node.

Also:
https://securityboulevard.com/2019/12/this-is-finally-the-year-of-the-arm-server/
"
None of this is true. There are plenty of RISC alternatives to Intel, like SPARC, POWER, and MIPS, and none of them ended up having a power efficiency advantage.


Mobile chips aren’t actually power efficient. Yes, they consume less power, but because they are slower. ARM’s mobile chips have roughly the same computations-per-watt as Intel chips. When you scale them up to the same amount of computations as Intel’s server chips, they end up consuming just as much power."

No. First this is not a RISC vs CISC debate. He is confused. No one is saying that ARM64 is more efficient because is RISC; what we are saying is that ARM64 is more efficient it is a new complete ISA that has been designed to be more efficient than x86, POWER, ARM32,...

ARM mobile chips have an efficiency much higher than equivalent x86 mobile chips, and when you scale up to server level the ARM chips continue being more efficient.

“Intel designed its cores for use in [systems] from laptops and desktops all the way to servers. It’s not optimized for servers. We have no x86 legacy, like 32-bit support and things like that,” said Hegde. “We are able to optimize our code, and our core area is significantly smaller [as a result]. Just to give you an idea, in the previous generation, if you look at ThunderX2, compared to AMD or Skylake, for the same process node technology [we get] roughly 20% to 25% smaller die area. That translates into lower power. When we move to 7nm with ThunderX3, our core compared to AMD Rome’s 7nm is roughly 30% smaller.”
https://www.hpcwire.com/2020/03/17/marvell-talks-up-thunderx3-and-arm-server-roadmap/
 
Last edited:
The question going forward though is do mass market consumers (including the majority working at home) need a ton of horsepower... or do they just need "responsiveness". That is something ARM does better... or at least most products built on ARM seem to. That may come down more to software... as the short amount of time I spent with a Windows ARM machine I also found it very "responsive".... as long as it was running proper ARM software, and not trying to emulate. I can understand chromebooks probably don't work for every industry... the ones they work in though they tend to work well and be very little admin hassle. It will be interesting to see how Apple does this quarter with their first ARM machines... I believe (could be wrong) that they are selling very well and are probably going to be a big shock to many doubters. I wouldn't be shocked to see Apple hit 20% of the market by next quarter.

I just hope whatever Nvidia (or another ARM players like AMD/Samsung/Qcom) cook up for MS is worthy when compared to the second gen Apple chips next year... and of course that MS is capable of unfucking its Windows ARM version. If not I fear x86 will become niche pretty damn fast cause frankly for most people that still need to consider tech support options as part of their purchasing decisions those Apple laptops with ARM chips are looking damn near perfect. It isn't exactly easy to argue they should go windows/x86 anymore when Apple is building such compelling ARM stuff. I admit for a few min I considered buying a ARM Air.... ahhh. lol
I would. I don't think they have the manufacturing capacity to ramp 5.5% marketshare in a couple of months particularly as total number of units shipping is increasing. I also would be hesitant and wait to see the reception of their new products at institutional users as well. I worked the PowerPC to x86 transition and it took us almost a year to qualify the new systems and then another before we started purchasing them. To date, we have zero work begun on qualifying the new MacBook Pro's or Air's and with budgeting the way it is I don't think we will see any work on it until mid next year.
 
No. AMD64 is an extension to x86. ARM64 is a new separate ISA from ARM32. ARM engineers used the transition to 64 bits for desing a new ISA and eliminate all the weird stuff used in ARM32. Details in the link I posted above.
Announced in October 2011,[1] ARMv8-A (often called just ARMv8, although there is also a 32-bit ARMv8-R) represents a fundamental change to the ARM architecture. It adds an optional 64-bit architecture (e.g. Cortex-A32 is a 32-bit ARMv8-A CPU[2] while most ARMv8-A CPUs support 64-bit, unlike all ARMv8-R), named "AArch64", and the associated new "A64" instruction set. AArch64 provides user-space compatibility with ARMv7-A, the 32-bit architecture, therein referred to as "AArch32" and the old 32-bit instruction set, now named "A32".

Still has compatibility, still ahs all that - despite whatever the talking head from Marvell might say. My link was from ARM themselves.
"Marvell ThunderX3 Gets Canceled for General Purpose Sockets". Marwell is pivoting to custom designs, because it cannot compete with the new generic ARM cores. Ampere saw this before, cancelled its custom designs and migrated to Neoverse.

Moreover TX2 got momentum in the HPC space, but the TX3, despite being a huge improvement over the TX2, cannot compete with Fujitsu A64fx. Originally the A64fx was going to be used only in Japan, but Cray will be selling them in Europe and USA

https://www.nextplatform.com/2019/11/13/a64fx-arm-chip-gets-a-big-push-from-cray/

HPC space is not consumer or enterprise space, again - stuff changes REALLY fast there, and their use cases are entirely different from anything else in the world (and depends on your data set and application, all of which tends to be custom to the case in question). I used to work in that market. Find me a generic ARM server for sale, and sales figures for it. Right now, they're not selling, except to the hyper-scalars, and they're still buying x86 too.

Sure. Not everything works still. STH has a graph with the status of the ARM ecosystem, from what works out of the box to what don0t.



Notice that TX2 used a less efficient 16nm process node.
Yep. Still similar TDPs, and if we hooked up a kill-a-watt (or similar) to a full fledged server, you're going to see power consumption really close.
No. First this is not RISC vs CISC. He is confused. No one is saying that ARM64 is more efficient because is RISC, but it is more efficient because the new ISA has been designed to be more efficient than x86, POWER,...

ARM mobile chips have an efficiency much higher than equivalent x86 mobile chips, and when you scale up to server level the ARM chips continue being more efficient.


https://www.hpcwire.com/2020/03/17/marvell-talks-up-thunderx3-and-arm-server-roadmap/
He's not making a point about RISC vs CISC, he's making a point that a more efficient architecture doesn't always beat a less efficient one, and that loaded up, ARM isn't notably more efficient than any other platform out there. I claim your second line (scaled up they're more efficient) to be incorrect, as per the evidence I've posted. Show me some true server specs (not mobile) and power draw, and lets see if you can prove it. And not in the HPC space, which again, is whack and weird.

Outcomes are what matter - what business outcome can you give me with ARM that I can't get with x86 that justifies the cost and overhead right NOW? Things may change over the next 3-5 years, but things may change the other way over that time period too (breakthroughs in x86, or power, or anything else!). That's the point I'm making - right now ARM isn't making major inroads because of compatibility issues and lack of software for enterprise/commercial, which limits how much of it the hyper-scalars will deploy, and how much of it the enterprise space will be interested in, for ~now~.

Oh, and your link is 6 months older than mine - that line was cancelled for specialty hardware for the hyper-scalars, while general-purpose arm stuff really hasn't hit the market outside of the niche players.
 
ChadD I did too, heh. I'm an Ubuntu guy and I've been interested in ARM procs for general purpose laptop/desktop-style computing for a while. Every few years I try to downsize a bit further, but I found once I got too low power (like the Intel NUC7PJYH, 4-core Pentium J5005 system) I started to really feel the lack of "responsiveness" as you put it. I figured an ARM based system, with the right SoC, could provide better performance than that in a similar power envelope, and maybe if it had decent I/O (SATA SSDs instead of eMMC or SD cards like a lot of SBCs, for instance) it would be that perfectly silent system that doesn't feel slow.

All the options so far have been low power SBCs in a DIY setup, hindered by slow storage (rpi), a small community (everything else), or not-really-made-for-what-you're-trying-to-do, too expensive and requires a ton of effort to get it working properly (nvidia tegra boards). TBH I am surprised we haven't seen a tiny desktop system or more laptops using Snapdragon 2019 / 2020 processors. They may not be as fast as Apple's M1 but... I don't think anyone would really notice. I know there are at least a few laptops out there, some fairly big name (like the recent surface).

The idea of a system like the M1 Air or Mini is very appealing to me, and I'm on the fence as to whether I'd be happy just running Mac OS on it and calling it a day. But I think I'm really hoping it gets ARM enough mindshare that other companies decide to take a chance on it and make some premium products that I'll be able to run my Ubuntu on. :)
Other then being a Manjaro / Arch lover I'm right there with you. lol I dashed my Air thoughts pretty quick. I am hoping that the non Apple players will give us something worthwhile soonish. I don't hate MacOS... but ya I would rather have something that can run Linux native (with good open source driver support). So far all the non Apple ARM as you say has been half assed intended to hit a budget market frame. I'm not looking for a 3k laptop or something... but I would pay more then budget level for something interesting.
 
Other then being a Manjaro / Arch lover I'm right there with you. lol I dashed my Air thoughts pretty quick. I am hoping that the non Apple players will give us something worthwhile soonish. I don't hate MacOS... but ya I would rather have something that can run Linux native (with good open source driver support). So far all the non Apple ARM as you say has been half assed intended to hit a budget market frame. I'm not looking for a 3k laptop or something... but I would pay more then budget level for something interesting.
This is the gap ARM has - either it's budget/mobile kit, or it's sorta-attempts at high-end server gear (that never sells well yet - not enough demand, not enough software)... nothing in the middle for people to use to develop on to BUILD for the two other ends. That's part of what made x86 win originally - killer apps and a middle-of-the-road price for consumer kit (which was a different market back then).
 
I would. I don't think they have the manufacturing capacity to ramp 5.5% marketshare in a couple of months particularly as total number of units shipping is increasing. I also would be hesitant and wait to see the reception of their new products at institutional users as well. I worked the PowerPC to x86 transition and it took us almost a year to qualify the new systems and then another before we started purchasing them. To date, we have zero work begun on qualifying the new MacBook Pro's or Air's and with budgeting the way it is I don't think we will see any work on it until mid next year.

I can see for some organizations it will take some time to get everything signed off. I think it will probably be much faster for most however today.... when PowerPC->x86 happened there where was no real i stuff on the network. Many many today have iphones and ipads on the system already, could make things a bit smoother this time at least as far as that goes.

I agree a 5-6% jump may be a bit much for one quarter... I don't think I would be all that shocked though. Apple made a lot of M1 chips, demand is high in general for laptops right now.... and the lower end Apples are probably the sweet spot for sales right now. With lots of people looking for something better then budget as there working from home but not in the market for insane cost either. Apple may have timed things really well. But ya will be interesting to see those numbers in a few months. :)
 
member when a 2 TRILLION dollar company switched all their products to run on any of those above....

i'll wait

you are joking right? all apple products ran on PowerPC for many years after they switched away from the /\/\otorola 68k processors
 
I can see for some organizations it will take some time to get everything signed off. I think it will probably be much faster for most however today.... when PowerPC->x86 happened there where was no real i stuff on the network. Many many today have iphones and ipads on the system already, could make things a bit smoother this time at least as far as that goes.

I agree a 5-6% jump may be a bit much for one quarter... I don't think I would be all that shocked though. Apple made a lot of M1 chips, demand is high in general for laptops right now.... and the lower end Apples are probably the sweet spot for sales right now. With lots of people looking for something better then budget as there working from home but not in the market for insane cost either. Apple may have timed things really well. But ya will be interesting to see those numbers in a few months. :)
I'm just looking at it from an organizational level and the issues we encounter. The i devices being present does not exactly make things better, especially here as we no longer issue iPads and all iPhones are personal property and your issue to support beyond syncing email, encrypting, and basic network authorization. We do zero application support. That being said, I think the fact that the roll out is limited will make it worse here not better. With it just being MacBook Pros and Airs instead of the whole lineup going forward right now we have even less reason to start working on qualifying them in our workspace since that represents an even smaller number of users/use cases to spend the same amount of time and resources on. It is just not practical with budgets the way they are right now. Thinking about that, my estimate may be off. We may not start qualifying them unless everything switches over since we can still issue users desktops that are already supported and laptops that are as well. Once the last of the current MacBook Pro's come up on their replacement cycle we will have had ~4 years to qualify the new ones or just tell users here is a Dell, enjoy, if the whole lineup has not gone over and or we don't just eliminate the Apple support altogether. That would not surprise me. Prior to the switch over from PowerPC to x86 there was no official Mac support and the addition was a considerable cost and time burden. Eliminating a minority of units and/or going back to the old "we will put it on the network and secure it but that is it. Don't call us to fix it." might be the preferred cost savings move.
 
  • Like
Reactions: ChadD
like this
This is the gap ARM has - either it's budget/mobile kit, or it's sorta-attempts at high-end server gear (that never sells well yet - not enough demand, not enough software)... nothing in the middle for people to use to develop on to BUILD for the two other ends. That's part of what made x86 win originally - killer apps and a middle-of-the-road price for consumer kit (which was a different market back then).
Apple it is, then :D Well, Nvidia has also been diligently selling reasonably powerful arm SoCs in the 400-1000 range for the past few years (pseudo SBCs, really). They were first to market with PCIe 4.0 (that consumers could buy), iirc. They are even starting to implement UEFI compatible boot environments for them, now that Raspberry Pi has done it.

But in the end, people always find an excuse for why they don't want to spend the money. Let's face it. People just want ease of use, and x86 provides most of that. Apple might be able to change that, but that's more of their own thing.
 
Apple it is, then :D Well, Nvidia has also been diligently selling reasonably powerful arm SoCs in the 400-1000 range for the past few years (pseudo SBCs, really). They were first to market with PCIe 4.0 (that consumers could buy), iirc. They are even starting to implement UEFI compatible boot environments for them, now that Raspberry Pi has done it.

But in the end, people always find an excuse for why they don't want to spend the money. Let's face it. People just want ease of use, and x86 provides most of that. Apple might be able to change that, but that's more of their own thing.
That's why I said business outcomes matter.

Technical wins lose to business outcome 80+% of the time. "It just works" beats the new hotness so often...

And as for Nvidia - none of those are systems; they're black boxes for developing mobile apps. Or for doing custom implementations of something weird. Fascinating, but less useful than a Raspberry Pi for the average developer. May have PCIE4... but not a single slot to use :p

There's no white-box / generic "ARM system" out there.
 
Announced in October 2011,[1] ARMv8-A (often called just ARMv8, although there is also a 32-bit ARMv8-R) represents a fundamental change to the ARM architecture. It adds an optional 64-bit architecture (e.g. Cortex-A32 is a 32-bit ARMv8-A CPU[2] while most ARMv8-A CPUs support 64-bit, unlike all ARMv8-R), named "AArch64", and the associated new "A64" instruction set. AArch64 provides user-space compatibility with ARMv7-A, the 32-bit architecture, therein referred to as "AArch32" and the old 32-bit instruction set, now named "A32".

Still has compatibility, still ahs all that - despite whatever the talking head from Marvell might say. My link was from ARM themselves.

No. That Wikipedia article is wrong and re-linking to it is not going to change anything.

ARMv8 introduced two architectural states: Aarch64 and Aarch32. Aarch64 only can execute instructions from the A64 set (what I have called ARM64). Aarch32 only can execute instructions from the A32 and T32 sets (what I have collectively called ARM32).

The first armv8 mobile cores (e.g., Apple Cyclone and Cortex A57) implemented support for both ARM64 and ARM32 instructions sets, the first for the new 64bit code and the other for all the legacy code. Apple has dropped ARM32 in its last cores and ARM plans to do it by the year 2022

https://www.androidauthority.com/arm-64-bit-1165264/

The armv8 server cores didn't have to support tons of legacy code because ARM servers were just starting. So engineers implemented only ARM64 in the server chips. E.g. the TX2 core doesn't supports ARM32

https://en.wikichip.org/wiki/cavium/thunderx2#CN99xx

That is just what Marwell is saying you in the link that I posted before. Their TX2 and TX3 cores only support the new ARM64 ISA. The cores don't have to support the old ISAs and the legacy 32bit ARM code. That is the reason why the TX3 core uses about 30% less die area than an equivalent Zen3 core (both in 7nm):

“Intel designed its cores for use in [systems] from laptops and desktops all the way to servers. It’s not optimized for servers. We have no x86 legacy, like 32-bit support and things like that,” said Hegde. “We are able to optimize our code, and our core area is significantly smaller [as a result]. Just to give you an idea, in the previous generation, if you look at ThunderX2, compared to AMD or Skylake, for the same process node technology [we get] roughly 20% to 25% smaller die area. That translates into lower power. When we move to 7nm with ThunderX3, our core compared to AMD Rome’s 7nm is roughly 30% smaller.”

ARM engineers implemented 64bit as a new separate ISA instead as an extension to 32bit, because the long term plan was to offer the technical possibility to drop support for legacy stuff in the future. AMD engineers implemented 64bit as a cheap extension to 32bit and this is the reason why any modern 64bit x86 core from either AMD or Intel has to support legacy instructions sets back to 16 bits.

HPC space is not consumer or enterprise space, again - stuff changes REALLY fast there

Not so fast. I posted in this thread a figure with the evolution of architectures in the TOP500 list and x86 needed 10 years to dominate the list. ARM is expected to need a similar amount of time.

Yep. Still similar TDPs, and if we hooked up a kill-a-watt (or similar) to a full fledged server, you're going to see power consumption really close.

The efficiency gain depends of what is being compared, but the gap is real and the reason why companies are migrating to ARM servers. I have some old graphs from Paypal with the reduction in "Power Consumption Per Year" resulting from the migration to ARM servers.

He's not making a point about RISC vs CISC,

He is making it in the first part that you quoted: "None of this is true. There are plenty of RISC alternatives to Intel, like SPARC, POWER, and MIPS, and none of them ended up having a power efficiency advantage."

I repeat again, this is no the older RISC vs CISC debate. ARM64 has a power efficiency advantage because it has been designed to be clean and efficient.

he's making a point that a more efficient architecture doesn't always beat a less efficient one, and that loaded up, ARM isn't notably more efficient than any other platform out there. I claim your second line (scaled up they're more efficient) to be incorrect, as per the evidence I've posted. Show me some true server specs (not mobile) and power draw, and lets see if you can prove it. And not in the HPC space, which again, is whack and weird.

TX2 in a 16nm node is as efficient as Naples in 14nm. Using the same node TX2 had been more efficient than Naples. Something similar happens when comparing TX3 to Rome. The ARM core needs about 30% less transistors and this translates into less power consumption. Jim Keller gave a talk explaining why ARM64 was more efficient than x86 and why K12 core was going to be better than Zen. Anandtech has an early analysis of the new N1 cores and of their efficiency gap with x86:

Switching over to multi-threaded workloads represented by SPECrate2006, we have to note that this is a best-case scaling scenario for all platforms as there is no serialisation or inter-thread communication, as the test suite simply runs multiple processes in parallel. Even with this in mind, Arm’s projected results for a N1 64 core design are just outright impressive considering the fact that we’re talking about TDPs much smaller than any of AMD and Intel’s solutions, creating a performance and efficiency gap that I have a hard time seeing the x86 solutions being able to compete against.

We have to remember that we’re comparing a 64 core platform against AMD and Intel’s current 32/28 core platforms. A more fair comparison would be AMD’s upcoming Rome with 64 CPU cores, here even if AMD manages to outright double multi-threaded performance and match Arm’s projected MT numbers, I don’t see them be able to at the same time lower the TDP to match Arm’s estimated 105W target (The Epyc 7601 has a TDP of 180W, Rome details haven’t been announced yet).

I hope that they can soon test efficiency of some Neoverse chip.

Outcomes are what matter - what business outcome can you give me with ARM that I can't get with x86 that justifies the cost and overhead right NOW? Things may change over the next 3-5 years, but things may change the other way over that time period too (breakthroughs in x86, or power, or anything else!). That's the point I'm making - right now ARM isn't making major inroads because of compatibility issues and lack of software for enterprise/commercial, which limits how much of it the hyper-scalars will deploy, and how much of it the enterprise space will be interested in, for ~now~.

Companies have been migrating to ARM no matter what you think about it. Paypal uses ARM servers, Microsoft is using ARM servers...

Moreover, the ARM-powered AWS instances are growing very fast.

400759_b7bdb9be-d97b-483f-a643-160de6c683bb.svg


EDIT: The Wikipedia page comparing ISAs is better than the other page you cite.

https://en.wikipedia.org/wiki/Comparison_of_instruction_set_architectures

As you can check, there is a single entry for x86 in the table, because AMD64 is an extension to x86-32. However, there are three entries for ARM in the table because A32, T32, and A64 are three different ISAs.
 
Last edited:
Back
Top