look at the PPA not peaks you are comparing a 1.1mm2 core vs 3. something mm2 core
like i said take core sizes and caches into consideration ~3X the core size btw for M3P
Covfefe
Junior Member
- Jul 23, 2025
- 14
- 27
- 46
The front-end for an x86 CPU is going to be a lot more complex just because of the ISA itself. An ARM CPU (in general as there are some exceptions) just needs to deal with a smaller number of 32-bit instructions. With x86 you've got hundreds of instructions ranging in length from 8 bits all the way up to 120 bits.
Having to deal with that requires more transistors and there's not much that can be done to get around that.
That's really a minor thing. It used to be a much bigger deal when your transistors numbered in the low millions like in the Pentium days, that's why RISC workstations beat x86 handily despite a significant process disadvantage (since Intel had a near monopoly on Windows PCs giving them an endless money spigot to pour back into their fabs)
With transistor counts numbering well into the billions it is pretty much a non factor now. Yes it adds complexity, as does supporting all the stupid modes back to the original 8086, so there's probably a little extra validation workload, and I suppose it might make the pipeline a step longer, but won't impact cycle time or width which are the most important factors for overall performance.
The main reason Apple/Qualcomm are beating Intel/AMD is they have more money. Same reason Intel beat the RISC vendors a generation ago. Maybe someday something comes along which is a bigger consumer market than smartphones and ARM/Apple/Qualcomm are displaced in favor of something else for the same financial reasons.
I'm inclined to agree with this. But maybe we should look at the cumulative R&D spend. I'm not sure how reliable this data is but we have such a graph:
Stock Comparison Tool | MacroTrends
I presume as portion of spending AMD likely has the highest percentage of their total spend on CPU development. But since they're also the smallest total spend, especially cumulatively, does the graph suggest that they will be catching up to Qualcomm? (Though, in my personal weighing they're also already ahead e.g. HX370 vs. X1E)
Last edited:
1t perf/w is still quite a bit better on SDXE iirc, it's just that it has not much else going its way. (also you need a unicorn in 80/84100)I'm inclined to agree with this. But maybe we should look at the cumulative R&D spend. I'm not sure how reliable this data is but we have such a graph:
Stock Comparison Tool | MacroTrends
www.macrotrends.net
I presume as portion of spending AMD likely has the highest percentage of their total spend on CPU development. But since they're also the smallest total spend, especially cumulatively, does the graph suggest that they will be catching up to Qualcomm? (Though, in my personal weighing they're also already ahead e.g. HX370 vs. X1E)
Indeed. But I noted my personal preference for a reason. I do not wank to efficiency charts. The reason I like M1 and M4 is because they avoid fan noise while browsing even bloated websites. And the X1E-80 Samsung I tried also failed that test and so, in my estimation, it is effectively equal to the HX370 in the "real world benefit" to this category and behind due to all its other defeats.
I'm inclined to agree with this. But maybe we should look at the cumulative R&D spend. I'm not sure how reliable this data is but we have such a graph:
Stock Comparison Tool | MacroTrends
What the IRS defines as "R&D" and what we would consider R&D are two different things. For instance everything Microsoft spends on developing Windows is "R&D", but I at least would consider that to be maintenance of a mature product. Apple spent $10 billion over some years researching a car but ultimately ended the project. That's R&D spending but it was mostly wasted (I say "mostly" because they probably got something out of it applicable to other products either current or future, but nowhere near $10 billion worth)
So question: is ARM architecture created from the ground up to minimize resource idling hence SMT isn't necessary?
Given the degree to which AI and crypto are eating power, perf/watt is going to be increasingly important to lower margin datacenters (relative to AI which apparently has zero cost considerations). Maybe if AI doesn't turn profitable that'll become less of an issue as all of that generation looks for a place to go.
No. The fundamental issue is that single core perf got hard to scale and multicore was a lot easier so we went that route. But a lot of problems can't be multithreaded, like at all. Because a lot of server work is a lot of parallel jobs, more slower cores may be a better trade than fewer faster ones, depending on how sensitive you are to how quickly each job gets done. For instance web servers don't really need a lot of single thread perf. Each request can get it's own thread so is parallellizes nicely, and you tend to be bottlenecked behind database access so CPUs tend to be a bit idle. HPC usually goes the other way - there's a cost to threading it so fewer/faster threads are better. Basic desktop tasks tends to be more single core oriented, depending on what you're doing, but with the benefit that you might have a lot of disparate processes all going on you can spread across those cores.
What ARM brought to the table were asymmetric cores, and in the case of Apple at least, a lot of work on the scheduler to clear the P cores of time insensitive tasks and move them to the E cores. Think of linux/unix with a gazillion little processes doing things related to I/O and so on - shove all that garbage onto a couple of E cores. ARM designs tended to make that trade-out quite heavily and could make do with a fewer number of P cores that could be kept fed more frequently. And within a given power budget, the E cores free up power to put more powerful P cores in than you would have had with symmetric cores. So devices like iPhone get to be surprisingly powerful while also being responsive because you aren't running additional P cores at 20% utilization when you can use E cores at 80% instead and do so in a smaller power budget.
Er, yes and no. I don't think the performance of the M4 is a product of money. I think it's a product of much better design decisions, owing more to them not having to fit Apple Silicon to a diverse set of customers all with different needs. It's extremely tailored. I think in terms of their ability to put one of the highest volume architectures on the leading node, that's money. What would M4 look like if it was still on 5nm? Not as good. But on the same process AS seems to still beat Intel/AMD on single core, perf/watt, etc.
A better answer would be what would M4 look like if it had to make the tradeoffs Intel/AMD make
No, a better answer would be what x86 would look like if it wasn't saddled with an unbundled market that requires those tradeoffs. It's not like they're either a technological or economic limitation - just a business choice, and therefore arbitrary.
Ampere® Altra® Arm-based Workstation
Specification may change without notice. Product may differ from images shown. Single Ampere® Altra® or Ampere® Altra Max Processor RAM: 8 DIMMS, 1DPC, 2933/3200MHz, ECC, max 2048GB. Storage: 2x M.2 NVMe PCIe x4 SSD 2280. Storage option: up to 8x SAS/SATA SSD/HDD, RAID option, from add-on...
store.avantek.co.uk
Covfefe
Junior Member
- Jul 23, 2025
- 14
- 27
- 46
If we're being pedantic, ARM Neoverse is still derived from ARM's X3 smartphone cores. Whether that qualifies as designed for smartphones vs designed for servers is not straightforward to answer in my opinion. Though I don't particularly care either way. What's your point? I don't think that detracts from my overall point that ARM cores target lower power use-cases than x86 cores.Ampere® Altra® Arm-based Workstation
Specification may change without notice. Product may differ from images shown. Single Ampere® Altra® or Ampere® Altra Max Processor RAM: 8 DIMMS, 1DPC, 2933/3200MHz, ECC, max 2048GB. Storage: 2x M.2 NVMe PCIe x4 SSD 2280. Storage option: up to 8x SAS/SATA SSD/HDD, RAID option, from add-on...
As an aside, I guess this why say to avoid absolutes in writing. If you don't, people will come out of the woodwork with the well, ackchyually's.
I think ARM was created to solve some problems that x86 was very poorly architected to solve
.AMD's version of SMT costs only 5% of the core die space and provides 1.3-1.4x MT performance uplift. It's really hard to imagine a better ROI for that 5% die space.... at least for MT tasks.
For ST, some of the things that make higher ST performance are also left sitting around. More super scalar capabilities raises ST, but also leaves resources laying around lots of the time. It's another good reason for SMT.
ARM, having targeted a market where MT is largely useless, was primarily concerned with performance per watt on ST. As it turns out, some of the tricks from this are also a good idea on x86.
I think that SMT doesn't make sense in tablets and phones as much. The added complexity just doesn't pay off like it does in DC and HPC.
This is also why I think ARM struggles in DC.
My suspicion is that each architecture is optimized for it's own market ..... and that should either decide to move into the other market, they would find themselves utilizing many of the same design decisions the other was using as well.
I don't believe that either ARM or x86 are "better" designs from the ground up.
Seems relevant, but keep in mind they admit AMD is sponsoring the piece in some way https://www.servethehome.com/deploying-amd-instead-of-arm-in-our-infrastructure-2025-here-is-why/.
Schmide
Diamond Member
- Mar 7, 2002
- 5,726
- 1,015
- 126
I think we need to be a bit clearer on what we mean here. x86 exists to serve one, potentially two markets - consumer PCs/servers. That's it. It's garbage for everything else.I think ARM was created to solve some problems that x86 was very poorly architected to solve
ARM was created for embedded/mobile. Specifically for the Newton - the proto-proto-iPhone, which is why Apple cofounded it. It then expanded into pretty much all other non-PC/server markets. M0 can be installed in a cable. Hell, an AirTag is an M4. Like ARM is almost literally everything else. The whole march by Apple up through the A series to the M series was met with absolute skepticism that ARM could even do that. The outrage that A7 could even need to be 64 bit got executives fired. This was not an architecture that had any business even being discussed in the same paragraph as x86. And then when Apple announced they were switching off of x86 to Apple Silicon there were scoffs from all corners that Apple could compete with x86 on desktop, and when it came out, it kind of smoked x86. Not everywhere, but when you looked at perf/watt and things like that, AS had a huge gap - more than could be explained by any kind of process advantage.
So which ARM are we talking here? The ISA that powered the DS and Nokia phones which has only been iterated on but not replaced? The standard core designs that power Raspberry Pis and mid-tier Android phones? Or Apple's custom work that is knocking heads with the best x86 or Qualcomms work that is doing so to a lesser degree? Or Neoverse that is competing with that server market?
Those are all the same ISA. They are mostly the same core design philosophy until you get to Apple's work where it peels off into this whole space that x86 can't even enter because they get a 'greater than sum of its parts' benefit by co-designing against Swift and LLVM and in-house asymmetric cores and memory architectures and bespoke schedulers and a zillion other things that Intel/AMD often have zero control over. So the question of 'what was ARM created to solve' is very, very different from 'what was Apple Silicon created to solve'.
I would argue that ARM was created to solve the need for an easy to program high performance/watt architecture that was flexible enough to reach everywhere that Intel couldn't. Because power was part of the core mission, it cares a great deal more about idle resources.
x86 didn't have power as a core mission really until desktop stopped being suitable for server. On desktop and to some degree even on laptop their limit was 'will it melt?' and so it was balls-to-the-wall. Apple Silicon is kind of what forced them to get serious about that. I guess we can talk about Atom but I think that sort of explains itself.
Apple Silicon was created to maximize the benefits of all of the other technologies Apple controls for the benefit of device performance that no OEM can possibly match because nobody has control of the stack. Apple gets to shove compute out of the P core and onto AMX or onto ANE because they write the APIs, they write the support libraries, they write the compiler, they design the language and they get to decide from this enormous tool shop which set of tools to employ to solve a given problem. They don't need to rely on SMT. x86 has a lot less latitude there.
Finally a thread where I can openly call Apple absolute cowards. Because if they aren't, why don't they support at least one Linux distro officially so we can have a proper comparison with other architectures? They are sitting on a ton of cash. Surely they have enthusiast engineers who wouldn't mind doing open source in their free time.
Apple Silicon is a carefully orchestrated deception/delusion of hardware+software and it makes financial sense only for a very small subset of people with particular computing needs to do things the Apple way. If anyone thinks that Apple computers will become as popular as iPhones, keep dreaming. Maybe a decade if M$ keeps up their charade to force people off Windows with their stupid shenanigans. But I see it more likely that Windows on ARM will take the place of x86 Windows. The MacOS UI, the design of the typical application and so many other annoying little things will prevent most seasoned Windows users from switching.
Here's one reason why I think Apple Silicon is nothing special. I use VLC player. Sometimes, the battery lasts about an hour on my M1 MBA. Sometimes two hours. The battery life depends on the type of codec, is what I'm guessing. How is this superior silicon that needs properly coded software to work properly? On Windows hardware, you run anything you want and you can mostly expect predictable battery life.
Apple Silicon is a carefully orchestrated deception/delusion of hardware+software and it makes financial sense only for a very small subset of people with particular computing needs to do things the Apple way. If anyone thinks that Apple computers will become as popular as iPhones, keep dreaming. Maybe a decade if M$ keeps up their charade to force people off Windows with their stupid shenanigans. But I see it more likely that Windows on ARM will take the place of x86 Windows. The MacOS UI, the design of the typical application and so many other annoying little things will prevent most seasoned Windows users from switching.
Here's one reason why I think Apple Silicon is nothing special. I use VLC player. Sometimes, the battery lasts about an hour on my M1 MBA. Sometimes two hours. The battery life depends on the type of codec, is what I'm guessing. How is this superior silicon that needs properly coded software to work properly? On Windows hardware, you run anything you want and you can mostly expect predictable battery life.
Good points and this is exactly why comparing x86 to Apple Silicon is a fool's errand. It's like comparing public transportation to private transportation. The fact that x86 is as good as it is (thanks mostly to AMD) is a miracle.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter