Lots of people are saying that Apple chips are good only because they have the node advantage. Let's see what they can do within the same power envelope now.
Not aware of those "lots of people", but all I can say it's the node advantage it's very real, at least it has been because TSMC hasn't really made any major mistakes in the last few years.
But real performance improvements post M1 didn't really match expectations. IMO the expectations Apple stated/implied in their usual marketing incantations are partly to blame, but they were the ones pumping the hype, so it's completely fair to charge them for that.
Compute performance improvements of ~30% are OK and in line with everybody else, meaning they are dealing with the same reality as other companies, hence the "let down"
Ignoring the fact that while doing that, there also happens to be a near 4090-level GPU and NPU and integrated memory on the M series chips on a die of approximately the same size...
You're comparing a 600W system (450W TDP GPU + 150W TDP Intel Core i9-13900KS) with a 90W chip (CPU, GPU & NPU combined). You don't see anything remotely problematic with that comparison?
No problem with the comparison. I was not comparing power efficiency. We all know that Apple prioritize efficiency.
Since power consumption vs performance is non-linear it's possible that the Intel+Nvidia combo would have roughly similar efficiency as M2 at lower wattage, although it certainly won't beat the M2 at 90 watts.
yeah, basically I think perspective is important to point out, when people are promoting the narrative "AMD has caught up to M1" but ignoring the fact it's not an Apples to Apples comparison. I'm sure 4090 can play cyberpunk with more FPS. And sure, I think the M series CPUs can eke out more single core performance if they devoted all 140 m2 of the die to CPU. But that's not what the M series chips are.
> Lots of people are saying that Apple chips are good only because they have the node advantage. Let's see what they can do within the same power envelope now.
Depending on what you're doing, modern AMD chips are very close to M series chips under load (specifically efficiency). The biggest issue is idle power is still higher. My amd framework (7840u) takes between 5-10watts idle unless I use something like the xtu tuner.
> Single core performance is only useful for artificial benchmarks
That is nonsense that none of the CPU competitors would agree with. In most applications single core performance matters very much. Not every algorithm can be multi threaded and there is an unavoidable overhead with those that can be multi threaded. Only some parts of some applications can be multi threaded.
For example, a 20 core 500 MHz CPU is much less capable and responsive for real world usage than a 5 core 2 GHz CPU, despite having the same instruction count per cycle.
A 100 core 100 Mhz CPU would take forever to boot up and feel unusably slow.
I just showed you that Apple is equal or better in terms of single core performance. This thread is a bunch of childish fanboy nonsense, attaching egos to some brand of CPU manufacturer and ignoring actual benchmarks.
Personally I don't care about $20 price differences. On a developer salary who gives a shit about price? I own Apple, Intel, and AMD cpus. They're all good.
>I just showed you that Apple is equal or better in terms of single core performance. This thread is a bunch of childish fanboy nonsense, attaching egos to some brand of CPU manufacturer and ignoring actual benchmarks.
So, just because you used one metric, then I shouldnt look at the other metrics?
You said "perf and price". Benchmarks show they are not beating Apple hard at performance. For price there is no straightforward way to compare since you can't buy standalone Apple CPUs.
At that level its competing on efficiency and capped by power consumption. It can't reach 5 GHz with only 80 watts for all cores. Running at 3.5 GHz. The Intel and AMD CPUs need hundreds of watts and reach 5 GHz+. It's a tradeoff for efficiency. Different design decisions. Different tradeoffs. Not exactly competing in the same market segments.
Mac Pro is often used for video editing. The M2 Ultra has hardware acceleration for video encoding/decoding that would need a separate accelerator card on Intel or AMD to match: "M2 Ultra can support up to 22 streams of 8K ProRes 422 video playback"
Yeah that's a multi-core rating. Apple doesn't lead on that currently, but for some reason that list doesn't include the M2 Ultra which would have their highest multi-core rating.
Apple beats them on single-core ratings. You can see the single-core rating if you click on one of those results.
Price cannot be compared because we do not know the price of an Apple processor. In fact, Passmark does not include Apple processors in their "best value" listings [0]
>Price cannot be compared because we do not know the price of an Apple processor
It cannot be compared perfectly, but you can try to estimate its perf/$
I'm not saying this will be easy, but imagine if the whole laptop was e.g 10k usd instead of 4-5k, then you'd instantly feel that something is expensive
> It cannot be compared perfectly, but you can try to estimate its perf/$
Hardly.
I don't see how one would be able to identify and normalize all the required variables, e.g. median life expectancy, average performance across metrics per watt, average power usage, residual value, etc.
For instance, I can sell my 2017 MacBook Pro for roughly twice as much as my 2017 Thinkpad, which has better specs. How do you factor that?
I wouldn't because it makes no sense. Occurance of great deals aint relevant here, imo. Why would I care that customers do crazy stuff on 2nd hand market?
>I don't see how one would be able to identify and normalize all the required variables, e.g. median life expectancy, average performance across metrics per watt, average power usage, residual value, etc.
How about building system for similar price to Macbook and comparing their performance?
It, of course won't be ideal, fair, whatever, but ain't it what gamers do? They find PC configurations and check how games run on them.
Exactly. There are so many dimensions across which to evaluate it. What I care about the most is 1) ST thread (running my personal workload which is inherently single threaded), and 2) Rust compilation (MT compile/ST link).
For 1) my fastest iron is i9-13700KS and Apple M2. They are very close. My Zen 3 is great and is notably more power efficient, but I'll evaluate 14700KS-Zen 5-M3 when possible.
ADD: because of winter I'm loving my i9-13700KS (not kidding, my office would be freezing without it), but come summer I'll care about efficiency.
It's for the entire laptop, but it fluctuates rapidly and is realistically closer to 6-8w. Disclaimer, I have a USB charger with voltage/current readouts so these values are from observing that when fully charged, though AMD adrenalin reports very similar results.
I see similar with a Thinkpad P14s with 7840U. If you can, try powering off the screen and observing power draw. If I haven't confused myself, mine drops down to more like 2-3 watts in that state.
As far as I can tell, it is not enabling LCD panel self-refresh. This may be where the extra idle power is going with screen on? If you think about it, it's a pretty expensive behavior to constantly read framebuffer content out of system RAM at 60 Hz.
I replied a longer reply to the other reply to your comment, but in terms of battery life 13th gen and lower intels are poorer to zen4 (7x4x) u series cpus, however apparently the intel hs/hx versions are slightly more efficient than their amd counterparts.
I highly recommend going through some youtube battery life videos and looking up notebook check reviews for whatever you're planning to buy/compare.
Recent intel systems unfortunately often push higher power to "beat" amd on performance metrics. Single core intel perf is still higher and likely will be, but multicore and efficiency of zen 4 is generally similar if not better than intel 13th gen. 14th gen helps efficiency but is barely available. Oh, and the amd igpu is quite performant, much better than the 13th gen and lower intels.
Btw as someone with a skylake laptop that also used to sip power, I suspect there's been a mild across the board power increase especially as newer chips clock much higher. My ryzen 7 iirc goes till 5.1ghz and is noticeably faster (i'm at 392 tabs in edge right now) than my skylake. I suspect your older laptop wouldn't clock so high, and a 3ghz limited intel/amd would have great battery.
Colder and hotter are temperatures and not measures of power consumption. A soldering iron can put out 75w at 800F, a cpu can put out 200w and top out at 175F.
In the modern era, AMD chips are actually known for running hotter for quite a number of reasons (much thicker IHS on AM5, stacked v-cache on X3D, boost algorithm deliberately saturating thermals, etc), even though the intel chips pull more power.
Akktually, according to the second law of thermodynamics you cannot get hotter with lower consumption of energy (with equivalent heatsinking, in terms of thermal resistance, measured in K per Watt) , at idle, as idle is a state of thermal equilibrium (and at rather low sustained power of 3-5w, well within ability of heatsink to dissipate), where none of your reasons are applicable.
> according to the second law of thermodynamics you cannot get hotter with lower consumption of energy (with equivalent heatsinking, in terms of thermal resistance, measured in K per Watt
cpus are not an ideal thermal system and do have their own internal thermal resistance. a 7800X3D runs hotter than a 7700X at equivalent (limited) PPT, which runs hotter than a 13900K at the equivalent (limited) PPT, because the thermal resistance is higher. these are objectively measurable things!
Also, generally, surface area is a component of thermal intensity as well and if you take the same flux and spread it out over more surface area you will get a lower temperature too. A threadripper putting out 250W does so with less thermal intensity than a 7800X3D putting out 250W and will run at a lower temperature too.
like yes, you are correctly describing the measurements in which these cpus are not the same thing, but then making the incorrect leap that "because in a spherical-cow world they would be equal" that these cpus are equivalent in these metrics in real life, which they are not. different cpus have different thermal resistances, and AMD's is generally higher right now because of the decision to go with a thicker IHS (to maintain cooler compatibility) and the move towards stacking (more silicon in the way = more thermal resistance).
and again, don't pretend this is some absurd or unknown concept, we literally spent years and years with amd fans making memes about "intel toothpaste"... thermals and wattage dissipated are not the same thing. you can have a great, efficient product with terrible thermal resistance, there have been a number of them!
it's just that AMD isn't on the top this time, so everyone pretends not to get it... or volunteers a bunch of theoretical reasons it doesn't matter... or ...
just like "thermal watts aren't the same thing as electrical watts!" etc
What I've interpreted from the consensus is that Apple's M-series advantage primarily comes from the relatively large memory bandwidth due to it's unified memory architecture.
Of course, it also has a very low performance-per-watt, comparatively speaking.
That's WAY-oversimplifying it. There are so many compromises and design choices and not just "One Cool Trick". Generally speaking the M-series traded frequency for IPC by going for wider and much deeper reorder window. This is easier to do with Aarch64 than x86 and generally is more power efficient (but always the devil is in the details).
The higher memory bandwidth doesn't do anything for ST perf.
Can someone explain to me how Apple managed to come from nowhere with a chip(M Series) that's competitive with the best chips from two companies(AMD and Intel) that have operated for decades exclusively in this field?
Shouldn't AMD and Intel have an expertise that ideally would take decades and billions to be reached?
Apple had been building its own mobile chips for a long time before releasing the m series. Their semi products absolutely did not come out of nowhere.
And they've done so much hiring since the PASemi acquisition that probably every much CPU designed in the last 20 years is well represented on the Apple team.
It wasn't nowhere. M1 was based on A14. It's called A14 because it's the 14th generation Apple Silicon chip (although the early generations were designed by Samsung so maybe it's Apple's 10th generation chip design).
In addition to what sibling comments already mentioned, also be aware that Apple was one of founding companies of Advanced RISC Machines Ltd (the company that became Arm Holdings plc), together Acorn Computers (which was the original A in ARM) and VLSI Technology.
This was back in 1990.
Before they switched to Intel x86, they were also involved in PowerPC, together with IBM and Motorola.
ARM64 and AMD64 are contemporaneous redesigns of the underlying architecture that got rid of a lot of cruft, so being ARM does not have an advantage there. One could argue that the more relaxed memory model of ARM gives a slight advantage, but that's a mixed bag.
Even without that, all modern chips translate the ISA into internal ops, and that translation take such a miniscule fraction of chip area that ISA in 2024 has minimal effect on power and performance.
This is due to it essentially having 6 channel memory versus 2 channel memory for AMD/Intel?
I would definitely like to see if the x86 industry could figure out how to include more channels while also not requiring 6-8 DIMMS to take advantage of it, like Thread Ripper.
I imagine they'll get it the same way Apple has done it with M series and AMD has done with MI300; by putting the RAM in the same package as the CPU. Doing so will break AMD's promise of 3 years of compatibility for AM5, so the change will either happen with Zen6 or they'll release a new range of chips called something other than Ryzen and sell Zen5 versions of both Ryzen and this new range.
Intel & AMD are leaving too much performance and profit on the table not to do so eventually; IMO it's more a matter of when rather than if.
I have to hand it to Apple, it's bold that they have on-package RAM that's so wide. (And then doubling and quadrupling whole core complex with tiling is a stunning move; Pro and Max.) Apple shipped a huge range at all once: a very capable mobile chip to a very beastly workstation grade chip.
Intel's most notable attempt to me was Lakefield (2019), a Mobile Internet Device (MID) class (sub-laptop) chip that I quite liked. The 1+4 architecture wasnt very fast and it was a bit more power intense than the Snapdragons of the day that it had a modest-to-significsnt lead over. But I loved that it was like this tiny tiny package that you basically just had to add power to, and the on-package LPDDR4X-4266 was a very speedy offering for 2019.
But both AMD and Intel already have big on-package ram cores. Intel's been shipping a "Max" Sapphire Rapids with 64GB HBM2e for a year (https://www.tomshardware.com/news/intel-launches-sapphire-ra...). Like Apple's Ultra, it's a quad-tile with each CPU having a it's own HBMe stack. AMD's MI300A is basically a GPU where some of the tiles are instead CPUs but it too has 8 stacks of HBM3.
I keep asking myself how & when & where is on-package ram going to arrive in. But there's already significant HBM presence in big cores! It didn't seem to make an huge difference for Sapphire Rapids; some help but unless one uses the expensive accelerators well there s probably not enough core to use it. Meanwhile MI300 is only just happening & more API focused. We have yet for on package ram to really be meaningful & available & making a difference like it did for Apple.
But as your post says, it feels like an inevitability. Someone's gonna make a core that can do more or be better by having lower powered faster local ram.
Part of my suspicion is that the market is resisting de-segmentation. The real issue is that Apple used on package ram to add many channels. Not of slow wise HBM memory, but mamy channels of DDR ram. These companies don't actually want to compete on throughout; they want throughput to be associated with $10k exotic chips. They are lament to build higher bandwidth more-channel consumer cores.
That starts to change some next year with AMD's Strix Halo, a big APU with quad-channel ram. There's no on-package ram as far as I've heard, but once you start having that much board real estate & energy going to ram, it sure would be nice to get even more performance for less power & much less space.
Damn I love on package setups. 2024 doesn't seem likely to offer much new or exciting, but 2025 has some possible signs for hope.
> I would definitely like to see if the x86 industry could figure out how to include more channels while also not requiring 6-8 DIMMS to take advantage of it, like Thread Ripper.
you can't really do that, DDR5 has a concept called "pseudo-channels" where a normal 64b channel can be broken down into 2x32 smaller ones, which improves parallel efficiency somewhat (now you can have 2 requests in-flight at the same time). But mostly bandwidth is down to the number of pins and how much data you can physically push down them, chopping the same pins into smaller channels doesn't help, other than letting you eliminate some inefficiency/overhead.
however, this is essentially the goal behind strix point/strix halo - narrower memory buses (compared to apple) with cache to improve the effective bandwidth. Just like in GPUs, this allows you to use fewer channels to get to the same bandwidth, which means less actual data movement. The downside is, of course, less "raw" bandwidth, if your workload is not cacheable.
basically my rough expectation is that it's going to be more expensive than apple silicon, and still probably not actually beat on power, but will allow you to do workstation laptops with 512GB or 1TB of unified memory, which is also something the apple stuff cannot do. They are different products, apple is targeting people who want a powerful ultrabook, amd is targeting people who want a mobile threadripper for actual work tasks (metrology is one example).
Really curious to see the efficiency on the mobile CPUs from this gen.
Been looking to replace a Tiger Lake ultraportable which despite being a good laptop otherwise is terrible in terms of battery life and heat, but have been waiting to pull the trigger until an x86 laptop that can manage battery life numbers similar to that of a MacBook Air without also throttling the CPU to oblivion appears. Hopefully this gen of AMD mobile CPUs delivers.
What is your battery life expectation? Genuine question, did not use Macs at all.
I've recently bought new MSI with a Core Ultra 7 155h, just finished making Linux work on it, seems that without turbo boost (set it up with TLP) it manages at least 10h of life, and I didn't notice any degradation without turbo in everyday tasks. I've very little experience with it yet though.
I expect that it will use less baterry once Intel Thread Director usage finally gets merged to scheduler, right now on 6.8rc5 it is still using all different types of cores as if they are the same.
Also, was surprised that there are 4 types of cores now in a sense - P, E, two new LP cores, and actually two out of six P cores have slightly bigger maximum frequency, 4.8 instead of 4.6 GHz.
The 13” Macbook Air is advertised to get 15-18h and 14” MBP 15-22h, and real world numbers with light-to-midweight usage is pretty close to that. My work/personal 16” M1 Pro/Max MBPs can end a day of work in Xcode and/or Android studio with 40-50% battery left. They do this out of the box with zero tweaking and have zero performance difference between plugged in or unplugged (no throttling).
My ultraportable (ThinkPad X1 Nano Gen 1) gets 5-7h on a good day with the system set to “power saver” mode, with little difference between Windows and Fedora. If I’m juggling several tabs or doing anything slightly more intense it’s going to be lower.
The best x86 ultraportable option I’ve seen from current gen offerings is probably the HP Dragonfly G4, which if configured with the 1920x1280 screen and lowest-TDP CPU can manage around 14h which sounds ok, but efficiency still isn’t as good as with the MacBooks because it’s only capable of that with light usage because it has a 16Wh larger battery than a Macbook Air, has a lower resolution display that’s about half as bright, and has to be in power saver mode.
More specifically, CES in January is when laptop OEMs announce their new products, with ship dates staggered over the next several months (though oddball products can sometimes take until September or October to hit the shelves). Laptop chip announcements are usually timed at or slightly before CES, because there's no point announcing a laptop chip out of cycle when it'll be an uphill battle to get OEMs to use the chip. (The importance of annual laptop updates announced at CES also contributes to why laptop CPUs and GPUs almost always get new model numbers every year, even if it isn't new silicon. OEMs don't like having to market a new machine with last year's chip.)
Wow this is terrible timing. I was literally planning on building my first pc in 20 years and I had settled on the 7800x3d. This might convince me to hold off for a bit if the 8800x3d is right around the corner. I'm also waiting to confirm whether the 5090 will have a 512 or 384 bit bus as that informs whether it'll be a 24 or 32gb card (need the VRAM for AI / ML).
The first to third-gen i5s and i7s, for being about as fast as several generations down the line (it wouldn't be until 2017 when AMD would release the first Ryzen processors that Intel would start to clock their processors at their full potential out of the box rather than artificially limiting them). This was most visible in notebooks, where the ultrabook craze would ensure comparable laptops wouldn't reach performance parity until several die shrinks later; this was also true for desktop processors provided you bought the unlocked versions and bumped the multiplier up to the 4-4.5GHz they were all capable of on stock voltages (they'd hit a thermal wall at 4.6-4.7; bumping voltages would cause them to heat to unsustainable levels very quickly).
The same thing is true of the GTX 1080 Ti, a card that 7 years later still has comparable performance to new mainstream GPUs, the upcoming mid-cycle gaming console refreshes, and the Steam Deck.
(This fact isn't lost on nVidia, who would like to have you pay for that next decade of usefulness up front; the reason for identical market prices per quantum of performance of the 3090, 4090, and soon 5090 are that way partially for this reason.)
I upgraded my 10 year old PC to Ryzen 5000 a couple of years back, and have never really looked back. Sure, it wasn't that long until 7000 supplanted it, but my PC is still just fine, and will likely be for the next 7 years or so (the biggest wild card is Windows support).
The only thing I really regret is getting a B450 Motherboard that only supports 1 NVMe SSD (and only PCIE Gen 3 SSDs). I would focus on making sure your motherboard has enough RAM and SSD expansion room, and then buy a big enough power supply for whatever GPU you want to run. You might want to figure out whether Zen 5 will have a new chipset, and if so, will the IO be significantly better. Nothing else is going to make a huge difference.
It greatly frustrates me that many AMD desktops have broken sleep/suspend modes.
I moved & started actually paying my power bill, and I really wouldn't mind my desktop's 100w draw if I could effectively suspend it and wake-on-lan it as needed (for either remote gaming, or to hop back into an existing tmux session). But if I suspend it, it goes down maybe ok, but never wakes up; I literally have to unplug it to get it back.
The various threads about give me the sense that I am far from the only one here with these kinds of issues. I tried windows, I tried turning on every wakeup I could find in the bios, I tried turning on every wakeup I could find in Linux. There's two different sleep modes, tried that. I wouldn't mind having lost like 8 hours of time to this issue, except I feel like I'm nowhere; no suspend, no tools to see what is or isn't happening. And it seems very prevalent on AMD. Frustrating.
Microsoft and Intel has a very overt conspiracy to kill ACPI S3 suspend. They mostly succeeded, so suspend support on recent platforms is now no better than in 2005.
Keep in mind that the 7000 X3D processors lagged the non-X3D processors by ~6months so if that follows for the 8000 series you might still be quite a ways out from those.
Based on the release of the x3d chips for the 7000 series, you would be waiting until January of 2025 or so. If you keep waiting until the next chips come out, you'll never build.
By the 7800X3D now, worst case scenario the 9800X3D is so good that you'll have to sell the 7800X3D for half the price on eBay
regardless of 7800 vs 8800, the next-gen launch will also come with USB4 built into at least the higher-end chipsets, and that's worth waiting for if you are an enthusiast who cares about expansion.
that's why I think there's no race on even buying a system now with a placeholder 7800X3D... motherboards should be significantly better next year too (albeit I'm sure more expensive too).
it may make sense to jump on RAM this year though, because most forecasts have RAM prices increasing 50-100% over the course of this year (and they are already up by 10-20% from the bottom) due to large cuts in production and a large rebound in demand. It's gonna be 2017 all over again.
wait wait hold up, is this why AM5 can't seem to run more than 2 dimms without the perf crashing? I'm having to find 32gb sticks right now because 4x 16 would be way worse
Get this [1] and never look back. They are HYNIX A Die, with a good timings (CL30) and with good speed (6000 MT/s), as well as with a spacious room for overclocking.
This fall I bought a 7700X with an MSI motherboard through a Microcenter bundle. The board came out in late 2022 and the firmware is still atrocious. I found a single release from August last year that is actually stable, anything newer causes significant problems. Failure to POST level problems.
If Intel can return to competitiveness without just dumping 300 watts* into a CPU I'll be switching back next time.
I have a sample size of one, so I really couldn't say. Problems with long boot times, Expo causing BSOD, context restore not working etc appear very widespread on the AM5 platform by what I'm seeing online.
Are you on AM5? My kids' AM4 board is a Gigabyte, and my friend has two AM4 MSI boards and they're all working fine.
I often see sentiment that x86 architecture itself is fundamentally a bottleneck, compared to ARM/RISC, when it comes to efficiency and performance per watt. Does someone with the right expertise have insight to share on this? I'm curious what the factors are here. I would imagine that a factor is that most ARM implementations we see (like from Qualcomm, Nvidia, and Apple) are full-on SoCs which will naturally have efficiency benefits. But I'd love to learn more about this before "taking sides" and declaring that so-and-so architecture is the future, or whatever.
X86 and ARM are practically the same these days in performance.
Let me explain RISC vs CISC, RISC is Reduced Instruction Set Computer and CISC is Complex Instruction Set Computer. The base component instruction sets of x86 are CISC. The base component instruction sets of ARM are RISC.
When technology evolves and newer instruction sets are required to handle those tasks they can often become more complex and so today with the variety of instruction sets on both x86 and ARM they are closer to each other more than ever. Still different.
Now going to the differences between ARM and x86 where it matters. ARM has the lowest power draw to performance ratio but it's need for power skyrockets as it approaches more complicated tasks. x86 starts higher in power draw but its performance is pretty maintained under all workloads.
Neither are wrong, just it depends what you're planning to do. x86 is probably the best architecture for the general user, but ARM is great in a phone if everything stays relatively simple. Notice how some phones with 4000 mAh battery will be dead in 30mins while playing a game which the Steam Deck can handle for maybe 1.5 hours of gameplay? However if I wanted something to stay on all day to receive text notifications and pretty much be idle in my pocket, I'd want an ARM processor. I think the real reason Apple went ARM is that after studying the life habits of their customers they realized most of their laptop users fold sleep their devices like a phone without ever really charging them. At least the majority, when working in a Mac Shop, the term used for a mac only software development team, the constant amount of fold close open recharge would've been better on an ARM processor. Someone at work even asked if they could just code on their Samsung phone because it got better battery life than the old Intel Macs.
Isn't there a problem with the parallel decoding of x86 streams due to all the different instruction lengths? I've read that going much beyond 4 parallel decodes in x86 gets increasingly hard, requiring expensive combinatorial logic. Meanwhile ARM instruction decoding can trivially be parallelized as much as you want.
Other than this I am not familiar with any other fundamental limits to making x86 as efficient as ARM or ARM as fast as x86.
You don't need expensive combinatorial logic. You just use a predictor to decide where the instruction boundaries are. This is the same strategy CPUs have used for branch predictors. Now your performance merely depends on the accuracy of the branch predictors and the prevalence of difficult to predict instruction sequences. That is hardly a show stopper for the high end and you have to remember that compilers try to optimize their code, so there is no reason for them to produce slow code on purpose.
Sorta, this is actually a CPU cache thing, ARM can do it efficiently not needing a lot of CPU cache to handle parrellel decoding. x86 requires more cache to do so. However more cache has its benefits not just in this task. Cache is also getting cheaper.
That still implies both more logic and more "hot" silicon, so decoding is higher overhead.
I recall reading about creating a subset of x86_64 that would be faster to decode, but this would effectively be a different architecture so at that point you might as well go to ARM64 or RISC-V.
I do know that if the instruction set decodes efficiently and is compact (to reduce memory bandwidth) it really doesn't matter much beyond that.
For starters, the CISC vs RISC debate has been dead for decades now. Often considered RISC architectures like ARM, have had vector instructions and branch predictors for a long time now.
> Now going to the differences between ARM and x86 where it matters. ARM has the lowest power draw to performance ratio but it's need for power skyrockets as it approaches more complicated tasks. x86 starts higher in power draw but its performance is pretty maintained under all workloads.
> So you agree that CISC vs RISC is not a thing nowadays.
Correct.
> What's a "complicated" task? What's the power draw baseline? What are the examples?
Complicated tasks typically involve the use of numerous instruction sets working together to complete a task like with video games that have physics and AI. Exclude AI co-processors for this example. Or even burdening the system with tons of multi-tasking. ARM succombs.
AMD's x86 Ryzen chips rival M series processors, but under stress can do more. M series is the pinnacle of ARM, you won't find anything ARM near it in any way.
> Or even burdening the system with tons of multi-tasking. ARM succombs.
This is clearly untrue, and you can tell it to my laptop running ~500 processes right now.
> AMD's x86 Ryzen chips rival M series processors, but under stress can do more. M series is the pinnacle of ARM, you won't find anything ARM near it in any way.
"Can do more", of what? What are your metrics, other than what appears to be a gut feeling?
Really it would take you moments to do research. Instead of saying something is worse or better and pointing a finger. If you doubt what I am saying you should really provide that detail too.
ARM64 and AMD64 do not have any significant differences in complexity. In 2024 they are both massively CISC, and most significant idiosyncrasies of both were ironed out in the 64 bit transition. They both use translation layers that a miniscule fraction of the transistor budget. ARM uses a more relaxed memory model which can give it a slight advantage, but there are enough trade-offs there it's not a pure win either.
IOW, in 2024 there's no significant advantage in ISA choice.
There are dozens of factors that matter more than ISA choice -- process node, design team competence, transistor budget, design goals, memory bandwidth, et cetera.
Your comment would be much more accurate if you said "Zen4" and "M1" instead of x86 and ARM. (I chose those two because they're on the same process). Zen4 is better than M1 in some metrics and M1 is better in others. But that's mostly because they had different design goals.
> Notice how some phones with 4000 mAh battery will be dead in 30mins while playing a game which the Steam Deck can handle for maybe 1.5 hours of gameplay? However if I wanted something to stay on all day to receive text notifications and pretty much be idle in my pocket, I'd want an ARM processor.
Untrue, one can easily play GTA Trilogy or Genshin Impact on an iPhone for ~4 hours. But even if it was, a phone battery is rated, on average, between 12Wh and 20Wh, whereas the Steam Deck is 40Wh.
> I think the real reason Apple went ARM is that after studying the life habits of their customers they realized most of their laptop users fold sleep their devices like a phone without ever really charging them.
Nonsense, suspend to RAM has been a staple in laptops for decades now.
> Someone at work even asked if they could just code on their Samsung phone because it got better battery life than the old Intel Macs.
GTA Trilogy 4H on an iPhone. I really doubt even the Max variant at lowest brightness can do that.
Generational improvements has been mediocre and I know for a fact that if you get 2H runtime with a moderate game (like Risk) it's already pretty good.
This does not work and just adds noise. This is not Twitter, if you want him to see something, you should send an email. Also, submissions in non-English languages are strongly discouraged so the quality of your source does not matter. You can submit a better article in English, though, which should not be too hard in this case.
I sometimes wonder at what point will compute density in DataCenter outgrows our compute needs. Example If you have servers that is 5 - 7 years old. You could have swap them with Zen 5c EPYC and reduce your server count by 70%+.
Apple M3 chips are using the tsmc 3nm stock as well so you could look at that vs m2 chip. Basically a few more cores for each sku, and faster clock speeds for both performance and efficiency cores, iirc it was a 30% gain.
It is a roughly 20% improvement to speed and power per area. And about 20% density.
So you get more, faster cores/units, in the same footprint.
sram.is the exception as it scales poorly, so don't expect much. If anything, this is the largest and most important hardware issue facing all the chip gains, since so much of performance is memory dependent.
For example, it is rumored that AMD is using the extra transistor density to double the width of the FPUs from 256 bit to 512 bit. Probably all the core structures like the ROB and branch predictor will get a little bigger as well.
Because the article goes on to say:
> "Zen 5 will be using the 4nm node while Zen 5C will be made using the 3nm process"
> "It might be possible that UDN has confused the Zen 5C 'Prometheus' core with Zen 5 'Nirvana'."