Why is Apple's M1 SO FAST?
HTML-код
- Опубликовано: 16 сен 2024
- Apple's M1 System-on-Chip has demonstrated astonishing levels of performance. Is this a carefully orchestrated gimmick, or is Apple Silicon genuinely offering x86-beating performance?
In this video, I attempt to explain the M1's advantages and answer that question.
#AppleSilicon #AppleM1 #M1Macs
Our NEW podcast channel: / @theconstantgeekerypod...
AMAZON LINKS - As an Amazon Associate I earn from qualifying purchases
==============================================================
M1 Mac Mini: amzn.to/3q5jg0v (UK: amzn.to/2UWOyrR)
M1 MacBook Air: amzn.to/2UXKTu4 (UK: amzn.to/33dNW5K)
M1 MacBook Pro: amzn.to/2J4Uv3I (UK: amzn.to/3610a3l)
That’s the best analysis of the M1 chip that I’ve seen.
Thank you 😊
*pirata song noises*
@Using Intel as a heater Unnecessary to whom? Many of us enjoy the detail.
then you haven't seen a lot. Because he missed the key difference: ARM has fixed-size instructions and can decode shitton of instructions in parallel, while Intel can't know where next instruction is without decoding the previous one. It's fundamental flaw, that can't be fixed without changing the instruction set completely
The best geek channel on the internet. Thanks for educating us with such clarity. 🙏🏽
it's Cuckoooo!!!!
I remember laughing... Now I want the m1 mini.
Im running Windows on my Mac mini using 6 cores at a 1500/5600 Geekbench score using Windows for ARM w/x86 emulation.
Before buying a Mac Mini M1 I have seen ALL the reviews on youtube, in English and Spanish. You sir, are making the best videos by far. Technical but accessible to people not so expert as it is my case, without raising your voice too much. Top class. Thank you very much.
Glad you like them!
Watching this on my M1 MacBook Pro. This laptop is the best laptop I've owned. It runs everything quickly. Video editing is a breeze and I've pretty much quit using the desktop for that and can edit videos on the fly.
Are you using fcp or have you tried the unoptimized Adobe suite like premiere pro?
m1 made me divorce windows. They can have their 'think different' slogan back.
Agreed.
But you think intel and AMD didn't know about power of ARM architecture and apple M1 chips?
i, like you and many others... this time realy admire apple undoubtedly...
But IMO, intel and amd always have something to amaze us and at least not get behind .... specially AMD.
Anyway hope all these buzz and competitions benefit customers financially and reduce the prices in general.
The m1 is not in competition with Windows, it's much slower. You must just browse the internet, in which any tablet from the last 10 years would be suitable.
@@veneratedmortal4369 How is the M1 slower?
@@kaisinelmusic How is it not? It scores less than amd desktop CPUS. When you need graphics performance it is not even a consideration, the graphics performance is at best only good enough for running RUclips videos.
Don't get me wrong, it's quite an achievement in its efficiency but it's not as good as the hype. They are using a brand new manufacturing node, tmsc 5nm euv which is 1 generation ahead of amd and about 15 ahead of Intel. That means they can fit more transistors in a smaller area that are more efficient, also despite being a mobile CPU it's die size is 119mm^2 amd 5800x CPU has a die of 80mm^2 so it doesn't really surprise me that much.
@@veneratedmortal4369 Well that's wrong on multiple accounts, people have been seen video editing on apple M1 and doing it with decent performance. Not so good for gaming but still better than Intel Xe integrated GPUs, which everyone thought was great (for an integrated GPU) when it came out. They are also great for programming and the like, though that is does not normally require as much processing power as people make out
AMD also now has 5nm parts using the same process and has for while. While yes they do perform slightly better, the difference really is only slight and those are desktop parts that draw at least twice as much power if not 3 or 4 times more power.
Don't get me wrong the multicore performance could be improved but that is what the M1X chip or whatever they end up calling the chip that goes into the 16" MacBook pro and iMac is for. This was designed as a efficient lower power chip that can go into small laptops like the MacBook Air and smaller MacBook Pro.
The computers that ship with the new chip will also likely have other features like increased memory and hopefully better graphics, not just CPU cores.
Then there is the neural engine which is it's own special thing that I am really not qualified to talk about but I take it whatever it does it does well.
watching this in my M1 macbook air. this machine just incredible, out of this world. never would i imagine using my macbook on blanket, on my lap, doing work without worrying the heat.
Man I can’t wait for mine to arrive
While my thies are burning when using Acer nitro 5 man 😂😂
And best thing is thst it's completely silent!
i was thinking the same thing las night when using mine. dead silent and literally cool to the touch. this is my first ever mac. really impressive.
@@churchill378 The Intel based MacBooks (and the prior G4 PowerBooks) were very good at getting very hot and imitating jet engine noises with their fans. Building small and thin has a cost. The M1 MacBook Air is amazing.
M1 has an extremely wide out-of-order rearrangement buffer. I.e. it can look ahead *a lot* of instructions. This also may be part of why some people have seen good performance with less RAM than they would usually consider "needed". It can request memory loads from the SSD far in advance of needing the data, so while it's working on other data, the SSD will ready up the request in the mean time. - All desktop chips basically do this, but M1 can do it further in advance than most
x86´s ooo engine is significantly slower because it has to support 1 to 15 bit instructions randomly while arm only uses 4 bit instructions for everything . on arm the compiler does more and the ooo/ooe does less, on old ass x86 , the compiler does less and the ooo is so much more complicated. well x86 had a good run and it was pushed to hell and back but now its old and newer architectures like arm are the future
@@alexmehler6765 I was surprised he said this was out of reach for the average joe. It would literally be the only technical bit in the video. And otherwise it’s just hand waving.
@@Ryan-sr9zv i think the video was good , he did reach on the fact that this enables m1 to perform up to double as much instructions per clock ... whats really interesting to me is how the software rosetta x86 translator on mac is faster than the hardware ooo on x86 processors , because x86 progs already outperform x86 processors on m1 and rosetta .
another good point of the video was the usage and easy implementation of hardware video decoders and other everyday-task optimisations just like its done on phones.
another good point of the video was the unified memory asdress space which we already know from consoles which also comes to pc in some form soon (gpu can directly access cpu memory) .
As usual you do an excellent job of giving a simple explanation of a complex system
Greetings from Portland, Oregon! I am a Senior Citizen who is also a geek. My first computer was a DOS machine. I really appreciate the evolution of the technology. I love the 13 inch ultrathin and light form factor, whether MAC or PC. I have limited financial means, so whatever I purchase has to work well today and be future-proof. The machines I will be saving for are either the M1 MacBook M1X or the Lenovo X1 Nano. At this point, I am strongly leaning towards the MAC......Your explanation was BRILLIANT!
Thank you sir. Greetings back at you from Devon, UK. 👋🏻
This is so simple and well explained. No goofy faces, chop edits, multiple cutaways or hip hop music. Subscribed!
Nice Video (and btw, I do listen to your podcasts to the end 😁). I just edited a 4:44m H.264 4K video in FCP on my M1 MB Air. Title, Color grading, Ken Burns effects, stabilization. Exported to 1080p (to a Samsung T5) for youtube in less than real time (2:45m to be exact). Super happy!
Imagine trying to do that with the Intel MBA 😬 Thanks for your support 👍🏻
I can do the same (perhaps faster) on my iMac 2020 plus I have nearly 100% third party software support.
Yes, but it is not portable and is twice the cost....
@@applerunner1184 Yes, agreed. But for me software incompatibility is a dealbreaker. I can't justify this to my clients.
Many years ago when PCs used MSDOS the practice I was working for used Apricot computers which were faster. I also went at that time to a computer show where Acorn were demonstrating their new RISC chip. A chap on the next stand was demonstrating their new ‘super’ PCs but Acorn realised that by running the same software on the RICS at the same time (co-ordinating hitting the ‘start’) that even running Windows (it was new then) in emulation the RISC chip was noticeably faster :-) The semi-monopoly operated by Micro$oft masks a poor product in comparison to the British software in the mid 1980s. My wife was a columnist on Acorn User magazine and her Acorn RISC powered laptop still worked the last time we tried 3 or 4 years ago when we sold the working RISC desktop.
I’ve never really got on with Micro$oft’s hype ever since :-)
RISC OS was definitely ahead of its time! Just think how different the computing world would be today if the likes of Acorn and (dare I say) Amiga hadn’t tanked financially!
@@ConstantGeekery , I suspect they were deliberately put out of business by Micro$oft ‘business’ practices.
@@ConstantGeekery I loved RISCOS back then; but it had major flaws. Cooperative multitasking is just not good enough. Also, no virtual memory (swap) support. ADFS was, er, primitative ;-) But I think mainly, with it being wrote in ARM assembler it was very fast to execute but far too complex to maintain.
Excellent and balanced analysis. I've been using Macs since 1985 and even an old grizzled veteran like me is still marvelling at how damn fast my new M1 16GB mini is. What has really surprised me is how well and how seamlessly Rosetta 2 performs-truly an under-appreciated aspect of the new systems. This is a paradigm shift in technology, particularly as it has immense scalability. Fugaku, the world's most powerful supercomputer is the first ever to be built on ARM and its performance beats the No. 2 machine Summit, which runs on Intel, by a factor of three.
Expect to see this kind of architecture being widely deployed in data centres in the coming years if for nothing else than its outrageous performance at low power levels.
ARM has come of age and is set to revolutionise all aspects of the industry in ways that we can not yet imagine.
Excellent video as always and very well explained! Do you think this is the end to pcie graphics cards for apple? I guess adding a graphics card like we do today would undo what they've created by having the graphics chip on top of the cpu.
Thank you 😊 I don’t think it’s the end of PCIe GPUs. Even if Apple can increase the core count and performance, they would need something like a tenfold improvement to get to the level of the best GPU option on the current Mac Pro... and many professionals who buy the Mac Pro (e.g. audio pros), won’t be bothered about massively improved GPU. It makes more sense to have an onboard GPU doing the basics, and a PCIe GPU to take over when there’s some heavy lifting to do. I like the idea of having the best of both worlds!
how can you not subscribe to this guy. The video starts at 00:01, no long intro, no hype, no promises or baits to watch the full video. Good on you sir. Have my sub and bell, Thank you
Thank you 😊
Great explanation. But, I f i may sum up: Apple's SoC stuff is freaking awesome, and they've only just got started.
Nice and simple explaination..thanks
Thank you 🙏
I sense a subtle dish on Linus. Really nice video.👍
I just picked up the M1 MacBook Pro 16GB and I have already initiated the trade-in process for my 2013 4-core Mac Pro. It is faster in every way. It is really a bridge to hold me over until a more pro level M1 Mac comes out, but I needed something now because my Mac Pro has been struggling with video edits lately. It is crazy that this laptop is more powerful than my old tower haha. Great video!
There’s no competition. I’ve owned a mac for about 5 years. I worked at a place that gave me a surface pro. Great, but on the same level as my mac. I moved to another company that gave me a $3500 Mac Pro laptop. Wow. Fantastic. Loved it. I recently purchased a Mac mini with M1 chip. I also recently was offered a job with a new company. This company provided me with a $6000 Dell Precision 7750.
My new $1000 Mac mini M1 absolutely destroys it.
Extraordinary explanation. Keep up the good work. You deserve much more subscibers.
Superb content as always.
A video on how user developed apps can shift workloads from CPU cores to the neural engine or GPUs would be great.
The subscriber count is so misleading of the quality of the content..
I'm confident this however is only temporary ...
Thank you 😊
You said it much better than me. Ditto.
yes, but the likes to view ratio is simply one of the best on any video on youtube....
@@ConstantGeekery this channel has the best Likes/Dislikes ratio. My favorite RUclipsr.
best analysis, technical yet simple to understand, perfectly explained, Thank!
Gosh. I've been looking at this for some time but this is the first time that I've actually understood what the M1 architecture is
Very informative and easy to understand.
I remember MacBooks with RISC/ARM architecture back in 2004-2006. Everyone was hyped and then Apple decided to put it aside and use Intel. Some of my friends (and myself) were so disappointed. Now I see this and I'm happy. Hope this time lasts.
Love your stuff just subscribed!
Thank you 😊
WOW!! Dave, arguably the best video yet from both 'Constant Geekery' channels. You've shown how the person who knows explains with ease. New tech new to us all explained coherently from First Principles. Thank you! I'm proud to have subscribed to both channels from the earliest days. My decision proved correct. Liked, commented and subscribed. #NoHypeNoBluffing.
Thanks Brian, you are very kind 😊
The most important factor is the M1 is on 5nm, AMD is on 7nm, and Intel is on 10-14nm. Each nm jump is about 20-30% improvement on heat, and thus energy saving or more power.
The RISC vs CISC performance are very close; the major difference is the die size as x86 requiring a lot more area thus requiring more power. With the extra space saving, it leaves more room for cache. AMD 4800U has 4MB L2 cache while M1 has 16MB L2.
Good points - thank you. Just imagine what will happen if Apple get to 3nm with the next generation!
This is the best M1 explanation I've seen.
It was an easy and calm explanation to understand. Well done 👍
You very easily simplified a very geeky subject. Great going.
As an old guy, who used to test prototypes of main frame computers in the early seventies, I have to say: Excellent explanation. However I have some remarks:
- The remark about executing 2 x as much RISC instructions than the CISC instructions, is comparing apples and oranges. Stick to the remark that RISC is more efficient.
- I expect that some of the ideas you explained, will be implemented in the AMD64 chips too, probably complemented with two ideas of AMD as used in the RDNA2 architecture like:
- a fast cache for GPUs;
- full access of the CPU to the GPU memory.
- I use Ubuntu (Linux) booting and running from ZFS, I have some of the same advantages the Apple system has. ZFS storage uses by default lz4 compression and uses a large dynamically sized memory cache, that is also lz4 compressed. ZFS has some additional advantages compared to APFS.
- My first 2 GB of SWAP space is on my nvme-SSD (3200/2300 MB/s) and I have no noticeable effect, if using that swap space. The remainder of my SWAP space is on a good old HDD, but that space is never used and I constantly monitor the the throughput used by both SWAP spaces.
- Don't worry about the SSD, I run the Host OS and constantly 1 or 2 Desktop Virtual Machines on my system. Looking at my SSD usage in the past year and extrapolating it, I expect to use that SSD for the next 12 years.
"And that's why the M1 can convert video codecs like h265 without breaking a sweat whereas an x86 machine will have to work a lot harder"
I'm not sure where you got this idea. In my own benchmarks, the MacBook Air M1 was approximately 50% slower than a Core i7 8750h in Handbrake x265 encoding, and 20% slower in x264 encoding, both running Big Sur and the latest universal binary of Handbrake.
Handbrake doesn’t appear to be well optimised yet. In FCP, H.265 works well on the M1 in my testing.
I’ve gone from a dual Xeon MacPro with significantly upgrades GPU, RAM, internal drives, etc. to a 16Gb/1Tb M1. My 13” MBP runs rings around my poor old cheese grater for photo editing, video editing, music creation and almost anything else I throw at it. I really wanted to see how the newer gen Apple Silicone performed, but I’m not in any way disappointed with this laptop. Between the M1 chipset and Thunderbolt for my external storage, I’m just beyond delighted with its performance.
cheers. that was simple and wery clear.
Thank you for your clear explanation. This is what I want to know: how does Mac M1 RAM really work!
This reminds me of way back when the Amiga had a custom chipset architecture to allow it to perform some great graphic effects (for the era) while only using a 7Mhz Motorola processor. The only downside is when things change in the future is whether those custom cores, in this case, will continue to perform well. So I wonder if the video encoding part will become obsolete if new video formats become the editing norm? Interesting times and definitely a huge shift in computing that we’ve not seen for some time.
Amazing content and explanation. Thank you so much!!! Subscribed!!!
Thank you 🙏
Clear and easily understood. Thank you. I am an Apple ecosystem user and it’s exciting to see a step change in computing now under way. Subscribed.
Thank you 😊
They also dropped compatibility with 32 bit processes in favor of emulation, removing a lot of backwards compatibility issues.
The end of x86? Many consumer users don't know the difference, or care about M1, but look for price, and are not willing to jump to MacOS, I know as I work with many. Plus as long as IT staff want to keep themselves in business, they'll keep recommending Dell and HP, unless I'm mistaking x86 for Microsoft Windows, I'm assuming they're synonymous. great video as always, the information goes over my head, but the conclusions are always excellent.
Probably the best overview of the M1 processor i've seen yet.
I'm hoping there is a problem with Handbrake for Apple Silicon, because my M1 Mac Mini is slower, and uses more power, than my laptop at work with Ryzen 7 4750u. I literally bought the Mac Mini to transcode to H265 24/7, and to have it be slower and use more power from my UPS that a reletively cheap laptop is more than a bit upsetting.
I thought to myself "hey, even if Apple is streaching the truth here a little bit, its still a RISC processor, on a 50% more efficient node, there is no way it will be less efficient" For the time being, it looks like i was wrong.
Oh, and before you suggest i use "video toolbox" i did, and the results were much like Nvidia Nvenc or Intel Quick Sync, it was much faster, and used less power(power from the wall went up significantly, but it took 1/3rd the time, meaning it used just a little more than 1/3rd the power), The problem is the file sizes were much larger than when using CPU only transcoding so i quickly ruled that out as an option
Now, even though i have ordered an even smaller PC than the Mac Mini PC to use for handbrake, i am expecting to eventually be able to use the nural engine to do things like automatic volume leveling, commercial detection, show overrun detection, and well things that a Windows/Linux computer wouldnt be able to do without an add in PCIe card
On the subject of single core performance, everyone forgets that these 'single core' tests actually only run on a single thread, leaving a significant ammount of performance on the table. This isnt an issue with Apple, beucase each core has only one thread, but with X86 processors with their super wide execution engines compared to ther Apple medium wide engines, there are between 2 and 4 threads per core(no 4 thread processors exist on the consumer market yet, but there have been specialized cards in the past, and the IBM Power 9 is a 4 thread archetecture) Now, you can use Ryzen master to disable all but one core if you had a Ryzen 3 3300x, or a Ryzen 5 5000 series non APU, then run a multicore test of your favorite benchmark, and you will get the true single core score. Note this only works with those processors because Ryen Master is unable to cimpletely disable all cores on a CCX, so the CPU you are testing needs to only have one CCX, the Ryzen 3 3100X and Ryzen 5 3600 cannot be used because they both have two CCX and thus, will have a minimum of 2 active cores. The 3900x and 3950x will have a minimum of 4 active cores.
For an anecdotal comparrison, because i do not have a 3300x or 5000 seires processor at my disposal, my 3600 (nonX) clocked to 3Ghz has a 2 core score of 2750 points using only 25w(according to HWInfo) Or 3167 points at 4Ghz drawing ~42w Putting the single core score somewhere around 1350-1580 depending on wattage. Compared to the single thread score of only 1217 while using 38w thats a big performance drop for not that much of a reduction in power. I cant wait to see 5000 series U processors, as then we will finally be able to see a processor in the same power class as the M1, and will hopefully be able to disable all but one core in the BIOS
Super useful and explained it really well, though I am now regretting that I went for the 8GB ... ho hum, 16GB or more when the M2 (or whatever will be called) comes out
Very good summarized, well explained, and as always, funny to watch.
Sir this was useful and interesting. You’re basically doing this for free and I appreciate it. Thank you. I had concerns about yet another m1 video but they were unfounded this was worth it.
Glad it was helpful!
Fast is an interesting term considering that the clock speed of these ARM chips are around 1/2 that in the CISC chips they are compared too. Once they match the clock speed used by the CISC chips, there will be no contest in performance. Also, the ARM chips have 2x the CMD decoders than CISC and because all cmds are the same byte size, ARM can do out of order cmd execution (which CISC tries to compensates with threads which is harder to code).
The most underrated tech RUclipsr
Very useful information. I have a better understanding of the M1. Thank you for sharing
Explicit info thanks
Very well presented and understandable
Thank you 😊
Very useful video, lots of tech information in it. like to learn more inthe future.... Thx!
I love the video. It was was the clarity of such greate explanation for my
Thank you 😊
Very interesting and I understood all of it (some of it!) Thanks.Subscribed.
Thank you!
Really impressive and informative explanation. Thank you.
Thank you 😊
So the M1 is a modern day Commodore Amiga. 🤔
When Apple announced they were going to put toy processors in their grown-up computers (the way I saw it), my heart sank as I envisaged the problems of being forced off the Apple platform after 30 years. I am therefore over the moon about being SO wrong and I can't wait to get my hands on an M1/M1X/M2 Mac. Whee
You're right that all those companies are unlikely to collaborate that closely, but Qualcomm engineers are. That is a fairer comparison since both make heavily modified Arm based SOCs.
Great overview.
This was just what I was looking for. Clear, concise, and most of all...Competent!
Subbed! Subbed! Subbed! 👍👍👍
Thank you. 😊
Excellent video, clear and to the point. Kudos from Colombia!
Thanks from Devon, England 👍🏻😁
Thanks very much. Everyday I have my new MacBook Air I feel better because of videos like this :-)
I have done the simple thermal pad mod (
When I received my M1 Mini last year, I experienced a feeling I really hadn't had since I sat down for the first time at the keyboard of a SPARCstation 1 over 30 years before. A "oh s**t this thing is FAST!" feeling that makes you want to giggle and make whooping noises. Yeah, I've worked on some really fast servers, but it's so very rare to feel suck a kick in the pants coming from a chunk of hardware sitting on your desk. Going from 680x0-based Suns to the SPARCstation felt like that, and going from an Intel Mac to the M1 Mini gave me the same thrill. I've done other transitions, and while the PPC Macs were fast, I never got the same "whoopee" feeling. Nor did the transition to Intel give me the same visceral thrill. Faster, yes, but not enough to evoke that feeling. It's nice to know that after so many years, I'm still capable of that feeling of wonder. Thanks, Apple!
Never lose your sense of wonder
I've seen and heard a number of M1 videos, but this channel always providea better explanation and info. Keep it up sirs and hoping we could have new batch of Apple silicon chips this year to explore^
You are very kind, thank you. Looking forward to the next chips too... though my wallet isn’t! 😩
Lucid explanation mate. Waiting for the M2 Pro before I upgrade my 2016 MacBook Air in 2023.
01 January 2022
best explantion of the difference between ARM and x86 chips
Excellent breakdown. Before I just chalked up the M1’s insane performance as some kind of magic, but it’s really brilliant engineering and tight integration.
Thank you for the great explanation! Amazing video!
Very in depth presentation. Congratulations
Many thanks!
This channel is one of the rare gems left on RUclips. I’m not sure how guys don’t have 100k subs already but I’m very sure you will. Keep it up.
Thank you - much appreciated 😊
Very thorough video and very well explained too!
😊
Thanks for explaining the SoC in such detail! Very informative.
IMPRESSIVE as it is so well explained. I just laughed at that...ahahaha. I have not seen someone as good as YOU. Great VIDEO.... SIMPLE...CLEAR.....Well STRUCTURED ANd WeLL Expained..... N' LOVE the ACCENT. ;DDDD GREAT VID.... I guess you are the Best on RUclips and online ahahhaha ;DDDD . Excellent!!!!!
I just took delivery of the M1 MacBook Air with 16GB RAM and 512GB SSD (which includes getting all 8 GPU cores). One thing I noticed right away over my 16" i9 MBP is that the WindowServer process uses a heck of a lot less memory on the M1. It's measured in GB on the i9, MB on the M1. I've not touched swap yet on the M1, but I have had memory compressed. On the i9, I have 64GB and have hit swap in the past. Since my last cold start, I haven't touched swap. I do have a lot of memory usage though.
I remember the original Macintosh. I had the 512K version. It also had shared memory. A portion of the system memory was used for the display. Drawing to the display was done simply by writing to that memory space. Of course the early Macintosh System OS didn't support multi-tasking or anything like that. At that time, the Inside Macintosh book set was in only three volumes and very accessible if you knew the Pascal programming language.
Anyway, they say silence is golden, so I went with the gold MacBook Air. It feels very solid. I just hope I can get TensorFlow and Keras up and running on it.
I wonder if the I/O limitations are real? I mean, lack of ports spark doubts on actual I/O performance of entire SoC - not something most people would be limited by, but definitely something to question in future Apple Silicon iterations?
I’d like to know the answer to that. There is clearly something not quite right with it.
You explain things in a way that is engaging and easy to follow.
Thanks 😊
great video! thanks
Thank you 🙏
This blew my mind. I didn’t even know system CPU’s were a thing…
Thank you for explaining just how a SoC implicates efficiency with lower clock speeds. It’s more similar to Zen3 IPC upgrades than I realized and explains just why Intel’s clocks aren’t translating to real world performance outside of specific applications.
Thank you for the best explanation
Great presentation and explanation. I’m quite excited to see where Apple ends up with their silicon if the first iteration is this good! My only real concern is the amount of SSD swap file usage in these machines which does seem to be significantly above the Intel based Macs and x86 systems. Not sure it’s a problem for most people after viewing your excellent video on the subject, but time will tell!
your summary of the architecture was useful. It makes you realise that intel Are so focussed on letting you plug in other vendors hardware, that it makes general computing inefficient.
Thank you a lot for this!
Glad it was helpful!
One of the best analysis I’ve seen. Thanks!
Thank you 😊
Your videos are simple and very easy to understand. I love it.
Avantages of M1 with increasing complexity (more cores, more RAM) may become its weakness.
That’s very possible.
I actually had to subscribe. He really breaks down a lot.
Thank you.
This M1 SoC is making me really interested in software and CPU architecture
Great channel, thorough explanations, you earned yourself a subscriber :)
Keep up the good work!
Thank you 😊
I ran out of memory on my M1 Mac Mini today (8 GB RAM). Just sayin'.
For those curious: I was trying to play something in Premiere Pro while opening Adobe Media Encoder. It wasn't a particularly complex project either, the M1 just freaked out for a bit. I should also note that Premiere's render settings were set to "performance," not "memory," so it's really my fault anyway. Just wanted to share because I'm proud that I've slain the beast.
8gb is too low, 32gb recommended.
@@robmaas3356 lol, if only 32 GB existed.
@@Sk3tchPad yes, therefore wait.
@Marc Nij yeah for sure. The M1 optimized version is gonna be awesome
@@Sk3tchPad the current M1 Premirre is missing features so be careful.
Great explanation!!!! Thx!
Thank you 😊
I’m surprised how well games run through beta parallels desktop running ARM windows, emulating x86 code (also only technical preview).
GTAIV is playable at 4K high settings (low draw distance because the game is so buggy) on the M1 Mini running through 2 layers of emulation.
ARM is the future. I just wish Intel and AMD make the transition before it's too late so that us consumers benefit from the added competition but for now I got to hand it to Apple as they really have shook up the CPU/Chipset industry with the M1.
Good analysis. Thanks!
This is almost the modern day equivalent of the Intel 4004, where it shocked people with its performance and efficiency and caused competitors to scramble to catch up.
@Muhammad Ehtasam I respect intel's work in this category, though x86 will never beat RISC in efficiency. Evo is efficient *compared to current x86 cpus.* Don't compare it to a full fledged M1 and say its better because otherwise it would be all over tech news.
@Muhammad Ehtasam that's a false claim, ARM is dominantly efficient over x86 due to reduced instruction sets and efficient architecture. You only need to look at the performance per watt of each cpu type and you can easily see that a 15W Arm cpu can keep up with a 80W x86. That's not to say x86 still isn't faster overall atm once you take it out a laptop and give it beefy cooling.
@Muhammad Ehtasam where is that data from? What program were they each running? Was the M2 running native or a virtual machine? Is the intel chip in a full tower pc with water cooling?
Also the 45W you quote for the i9 is base clock, not boost clock, and the M2 Max is a 30W chip not 45W.
I bet that the i9 has full tower water cooling and the M2 Max is in an air-cooled laptop.
Power on an IC is proportional to C * V^2 * F, where V is the supply voltage and F is the frequency. Power goes up linearly with frequency, it does not go up exponentially as stated in the video.
I'm by no means an advanced mathematician, so I'm working more on received knowledge from other experts. That said, I do know that when increasing the clock speed (frequency) of a given model of CPU, the voltage also needs to increase. Normally the system will be dynamically increasing and reducing the voltage as the clock multiplier is adjusted (i.e. for "turbo" frequencies or overclocking). Ultimately, to run at the highest clock speeds possible for the CPU, both frequency and voltage will increase.
Per your formula, you multiply the switched load capacitance by voltage squared, and then by the frequency. I don't believe this will give a linear result, rather each step increase of frequency and voltage, will require an ever increasing jump in power. Indeed, I've run some numbers on an Excel sheet and it generates an exponential curve for fixed step increases of voltage and frequency.
And that's before we consider the additional heat, which will require the cooling system to ramp up and draw more power. Have I missed something?
@@ConstantGeekery
At a given process node there is a small range of voltage that one can tweak. The 5nm node will have a lower nominal voltage than the 7nm node, for example. When both Apple and AMD use the same TSMC 5nm process, they will use the same voltage. AMD CPU runs at a faster clock with the same 5nm node because they design the logic pipeline to run at faster clock, not because AMD provide the chip with higher voltage.
Back to C*V^2*F, that is the basic formula used by the IC industry. When the geometry is shrunk by 80%, voltage can be reduced by 80% and electric field will stay the same. Electrons will travel the same speed but because the channel length electrons need to cross from the source to the drain is only 80%, you get an effective 125% speed improvement and at the same time the power consumption will be only 64%.
What you said about the benefit of integration are all true. However, a key reason M1 runs so much faster in raw power than other CPUs actually comes from TSMC 5nm process.
I loved the way you explain a complex topic with such simplicity. Keep up the good work !
Thank you - glad you liked it!
Succinct and useful explanation. There’s a career for you at the Open University 😁
😁 always good to have a backup career plan!