The Apple M1 isn't magic, it's good design - M1 Deep Dive pt 2 | Upscaled

[@slash196]

Upscaled should be its own channel, it's so head and shoulders above anything else on here.

[@hieusaturated]

Agreed!

[@teejstroyer]

Probably a smart marketing/rebranding strategy too

[@Rafayel06]

Exactly

[@bsvenss2]

Totally agree.

[@tomoliveri8168]

Would be nice if they took 0.5s to just... breathe once in a while... this is an assault on my eardrums

[@sprucegoose6933]

Just wanted to take a moment to say how grateful I am for your incredible in-depth videos on tech.

[@widnmbodken]

@@epiceconomist3811 you a very smat scammer , what a galaxy brain /s

[@Nors2Ka]

In-depth? More like in-deep pile of bullshit and misconceptions.

[@itsmilan4069]

@@Nors2Ka oof looks like someone didn't like the truth

[@Nors2Ka]

@@itsmilan4069 The truth about how CPUs work? Sorry, but what's written in Intel's cpu architecture manual just doesn't correlate with this video and I'm leaning towards what's written by the creators themselves than some soy boy on the internet tries to explain stuff to normies.

[@lisuvirizwa]

Upscaled carrying Engadget on its back.

[@kevinbatts2804]

Halfway through this I realized this was Engadget. No exaggeration this is the best thing I've seen from Engadget ever. Wow

[@AbhimanyuKumar-bt2xf]

in a single video i got all my answers to the questions I was looking for in various quora answers and reviews

[@CarlosSpicyWiener111]

Ikr!

[@RareMade]

what was your answer?

[@AbhimanyuKumar-bt2xf]

@@RareMade i wasn't answering all those ques, instead I was looking for the answer.

[@baroya]

This guy is so serious, enthusiastic - makes one of the most boring tech topics interesting! Thank you!

[@Satialfactory]

Defo not boring imo

[@knechtnoobrecht-c137]

on what earth is this boring my dude

[@salty4]

he's the reason I have subscribed to this channel

[@RomeoYouMust]

@@salty4 Same! I really love Upscaled videos and watch every single one of them, not so much the rest

[@WarriorsPhoto]

Yes he does. Glad I am subbed.

[@JoeLeonardo]

This man taught me with his words and gave me a panic attack with his delivery.

[@yellowboxster06]

True for me as well. I give him a lot of credit for subject knowledge but his presentation style made my head hurt.

[@corvoattano9303]

Perfect Description

[@jonathanodude6660]

@@yellowboxster06 why?

[@yellowboxster06]

@@jonathanodude6660 Fair question. Technical knowledge unquestionable. Message delivery: First reaction - dial down your caffeine levels...seemed too hyper.

[@1990iRock]

Put the video on 0.75x speed and suddenly the video is alot better to watch.

[@thsithks]

Please, please lay off the insane compression on the dialogue track.

[@SiriusXification]

Yeah like wtf this sounds horrible

[@theparryotter]

i think hes just not using a lavalier mic or maybe he was and the track got fucked so he used the camera track? its too echoey.

[@JordanFreshour]

Thank you

[@lexacutable]

agreed, it's giving me a headache

[@v3xman]

Yeah it sounds so crushed and horrible I hate it

[@almostinfamous42]

That buff vs buffer was absolutely hilarious 😂

[@MishaAmashukeli]

03:19 this article is total BS. The reason single-core performance is important is that some tasks are just not parallelizable. When people say "single-core performance", they mean single logical core / single-thread performance. When you have a program that does not utilize multiple threads, it does not matter how many logical cores you got - you only use one, so SMT does not help here.

[@utubekullanicisi]

What he says in this video though is that the M1 feeds its body with data so well that it doesn’t need hyperthreading to fulfill its potential, and that SMT is more of a “band-aid” for situations where some parts of the chip are sitting at idle at a specific task.

[@SkepticalCaveman]

A OS should be designed to be preemptive multithreded from the ground up.

[@piotrd.4850]

@@SkepticalCaveman Windows 95 could do it for 32 bit apps - can't MacOS?

[@swarnavasamanta2628]

Yes i agree, but i think it would be also clear to say that the basic principles of OS and kernels is to actually get good multithreadinv done. You may think implementing multithreading in programs or processes is sometimes counter intuitive, but overall when you think of large scale applications they all have multithreading capabilities since they do multiple tasks simultaneously and it would be a waste not to.

[@SkepticalCaveman]

@@piotrd.4850 I was thinking about BeOS.

[@ktkps]

You are a very very good technical communicator who can digest and present technical topics in a very unbiased and clear way. Keep continuing your work!

[@pweddy1]

RISC V designers are like “You call that RISC?” to Arm chips.

[@MaddTheSane]

What do you think ARM stood for?

[@pweddy1]

Originally it stood for Acorn Risc Machines. People who are ignorant of their history will call them merely Advanced RISC Machines. People are ignorant of computer architecture will assume that the name defines their behavior today. The whole point of the risk five project was to go back to base principles. Because architectures like arm have a abandoned the base principles. Risc 5 has the hallmarks of a Risc architecture, a small clean instruction set that facilitates a efficient decode process. I've been a computer programmer since the original risc came out. This isn't history for me I lived it.

[@benygh911]

you're Totally RIGHT, and some Preliminary Tests Demonstrate how RISC-V TOTALLY KICKS ARM's A$$, Not only in Terms of SPEED, but also Efficency 😁✌

[@bartomiejkomarnicki7506]

@@benygh911 actually ONLY in terms of efficiency, so far it doesn’t scale at all

[@benygh911]

@@bartomiejkomarnicki7506 I'll Try to Remember *where* I saw the benchmarks and throw you a Link here, but I saw it even WON Easily in _Performance_ 😆✌

[@arumaakadalamudhan.t9900]

Upscaled is awesome it's like no other KZbin channels out there in explanation. This guy is excellent in explaining things. And special mention to video editors of Upscaled at Engadget, did excellent job. They nearly brought the half of apple wwdc and put them at correct places, that helped in understanding the concept that this cool guy tries to deliver for us. It's a combined effort. Well done. I'm seeing ur channel is going to get popular soon. LoL it's already famous!

[@ondrejtrhon]

Seriously this is so good, accessible yet rigorous, please keep up! Love it.

[@childnuk]

Well done! This is the first M1 video I have seen on KZbin that refer to RISC and OOO execution. There are too many videos with very talky hosts who are doing just that, talking, nothing else! It is refreshing to see a video that touches lightly but does reach the core advanced concepts within CPU designs.

[@Al-nv7zm]

Upscaled is like a whole damn course in current electronics

[@SamAlexanderYouTube]

this guy has quickly become my favorite from Engadget. keep it up my dude!

[@TheLegitFew]

3:24 “Oh no, my $899 netbook running at 15 watt TDP and without a fan is only as fast as an i9 9800HK?”…That doesn’t sound like a performance gain that’s been “evaporated away” 🙄

[@Yzyenthusiast]

lol thats what i was saying haha

[@MrAyybee2cold]

Lol that made me wanna get a 16 GB M1 Air for photo video editing lol. Things a beast.

[@MegaAbzzzzz]

About the Cinebench benchmark, Intel 8 core 16 threads, or course it wins because realistically M1 has a little big config. Yes it is 8 core but only 4 of its cores CPU are high performance based cores, and 4 low power cores. That said this is crazy impressive what it has been able to achieve with what its got.

[@evtimpapushev7137]

Single-core (i.e. single-thread) performance tests aim to evaluate the ability of the CPU to run a sequence of instructions with data dependencies between them, branch prediction / recovery from misprediction; they test the effectiveness and efficiency of decoders, pipelining, Out-of-Order execution, not just pure computations. Apple's M1 has 8 decoders, but that does not magically help with resolving dependencies in the instruction pipeline, OoO execution, etc. On the contrary, there are more instructions that need to be reordered for best utilization of the pipeline; more instructions will be scrapped in case of branch misprediction, etc. On x86/64 CPUs there are only 4 decoders per thread as instructions complexity does not allow for more without making the silicon way too complex (according to Intel and AMD engineers). So, to work around this limitation, they've introduced hyper-threading and added 4 more instruction decoders. But, they cannot feed all 8 decoders from a single thread, they must process two threads. While that seems to level the playing field, it does not. Instructions from different threads are inherently independent, so the amount of analysis, reordering and planning that the CPU has to do is reduced. In the multi-threaded tests the advantages of hyper-threading are clearly visible and that's fine, that's how it should be. But changing the testing methodology just to push M1 down is utter garbage and has absolutely no merit.

[@steveseidel9967]

I believe there is a general misconception (not with you) about hyper threading (SMT). Some think this is an advanced feature that all chip designs should have. It's not. This is effectively a hack that is used to regain lost efficiency from a poor decoder (like Intel's design). This video makes it look like multiple threads on a single core will work 2x as fast. In reality, Cinebench is literally a best case example of how Intel chips can get back some efficiency through SMT and even there, it only gains about 20% performance improvement. That also comes at a cost of security as we've seen in the Spectre attacks and many enterprises actually disable this functionality because of it.

[@ThePowerLover]

_"But changing the testing methodology just to push M1 down is utter garbage and has absolutely no merit."_ *^This!*

[@bizmonkey007]

What’s counts is the M1 looks good on paper and performs well in practice. I could see changing the testing methodology if the benchmarks didn’t line up with performance, but that’s not the case.

[@shashanks.k855]

This guy is awesome... as a physics person myself, he talks about it technically enough not to bore me to sleep as most channels tend to. Please keep them coming. Again AMSL episode was amazing.

[@danielwoods7325]

Adding my voice to the calls for Upscaled to be it’s own channel - let’s keep building momentum!

[@sayakdasgupta8905]

I wish they made longer productions like this.

[@raguilaru]

First time watching the show. I’m impressed by the production quality plus the detailed technical content in a digestible format. I’m a fan.

[@pweddy1]

A execution unit is usually broken up into Integer, Floating Point, Load/Store not add, sub, multiple or divide.

[@flippert0]

Yeah, the statements about threads etc. aren't very in-depth (as some commenters here might think) but actually a bit light on content. A comprehensive treatment of those topics would of course go beyond a 15 min video.

[@muri038]

Very well explained, animated and really easy to follow. Good job mate.

[@SHlNOBUU]

the best video series about apples M1 chips I've seen!

[@torpedospurs]

7:31 Q: Why is it that you can run two instructions simultaneously if the core is single-threaded?

[@AlejandroLZuvic]

I sincerely congratulate you on the quality of both presentation and content. You mastered at explaining how all this works. I liked how you explained out-of-order execution and the fact is nothing new and all CPUs do this but at the same time explaining exactly where the M1 is different than its competition.

[@MePeterNicholls]

Great content. One tip though, less compression on your voice would be great

[@vasanthtcs2009]

Amazing and informative, not the same boring ones where ppl keep talking about performance benchmarks all the time.

[@rahulshah336]

This series is amazing, you guys should spin it off into its own channel!

[@Sacchidanand]

More upscaled. PS - Take your time, thank you so much for such detailed video. 😊

[@lukedstaten]

Excellent video! Seriously. I never thought I'd say it but Upscaled is totally underrated according to the quality

[@rajatmann8897]

Best most researched video on m1. Period.

[@zhengjia9580]

Hey, just a clarification, 11:40. Cortex X1 is a 5wide decode, not 8

[@seasong7655]

5:50 Doesn't the CISC instruction also do more work itself than a RISC instruction?

[@MikesterF1]

As someone with almost no understanding of hardware architecture, these videos gave excellent explanations.

[@Sirikiller]

I have a question, apple implemented x86 'like' instructions in m1 to boost emulation performance. So will intel and amd be able to redesign traditional x86_64 architecture enough to not to break legacy instructions but making the architecture more in line with RISC.

[@ShredBird]

This is one of the best explanations of instruction level and thread level parallelism I've seen. Great content!

[@Blondul11]

My only worry is with the SSD durability. I bought the 8GB version because the 16GB one would take too long to arrive, and I use it for Android/iOS + AOSP compilation. While it performs well in all, I have days where it writes ~500 GB of data to the SSD. How long until the SSD gives and turns it into a brick?

[@Teluric2]

Whats the problem? Ssd dies and dump it and get a new one.

[@arumaakadalamudhan.t9900]

Complex things, but the way of explanation given or presented is great. Those animations where add mul load where awesome which remained me my OS classes at clg.. Lovin it.

[@isthisjish3794]

This is the kind of tech videos I’ve been waiting for. In depth but greatly easily explained

[@Rafayel06]

This channel is so good, I can't compare it to anything else! It should get more views!!

[@philsarcade]

As you mentioned in the video, I'm really interested in how they push that amount of graphic power in a non fan Air and still consume so little battery power. What is their secret sauce and is it scaleable?

[@groszak1]

Planned obsolescence.

[@brightboxstudio]

Part of the answer is that Apple spent 10 years figuring out how to create high graphics performance in their A-series SoCs for iPhone and iPad, which have very limited cooling and power. Once Apple mastered that, they were ready to move that chip design to desktop/laptop devices, where it can do the work on much less power, which results in much less cooling needed. And leaving a lot of power/thermal headroom to crank it up. In contrast, x86 designers spent the same years assuming everyone had lots of power and active cooling, so now it’s a challenge for them to design an SoC with high performance at the low wattage/heat of the M1.

[@steakman7662]

Wide and slow design and super dense process node (TSMC N5).

[@AlejandroLZuvic]

They’re using tile based deferred rendering, which in reality is and always was more intrinsically efficient than immediate rendering. Sadly, as he mentioned in the video, there are very few details on how the GPU hardware is actually designed and implemented. Let’s hope somebody could reverse engineer how it works.

[@philsarcade]

@@AlejandroLZuvic This is what I'm saying. There is very little out there but someone knows or can examine the chip die and have a general idea of the processes involved. When you look at the like of Invidia and the likes, they are very fast, but seriously power hungry. Will be interesting how Apple scales their design and if they can keep that thermal envelope low.

[@SpacedAug]

Great video! Audio is a little rough though D :

[@andrewdunbar828]

Umm so then what's hyperthreading?

[@tjhrulz]

Hyperthreading is Intel's branding for what is called Simultaneous Multi-Threading (SMT). This is what this video explains as multithreading, which is not quite the right term to use in this context. To truly understand SMT you must have a basic understanding of what a thread and also what a core is (Which thankfully this video does start to explain). In the beginning of computers a running program had a thread of instructions it would run, aka a thread. As OS's evolved and we were able to run multiple programs at once so each program had its own thread and the CPU/OS would switch back and forth between which of the running threads it was currently working on. Over the years as CPUs got faster so did your program, YAY! However we started to run in to the laws of physics, turns out the faster you make the CPU the hotter it gets (as well as other issues with laws of physics) this makes it difficult to dissipate that much heat so we had to come up with a way to increase performance in other ways. Enter: Multicore CPU's. Basically the idea is what if we duplicated pretty much the whole CPU and glued those two together (Its a little more complex than this but you get the idea) we now refer to each part of that MegaCPU as a Core. So now we have twice the amount of cores and thus twice the amount of performance, right? Well, kinda. See each program is still only a single thread of instructions which may rely on the previous one to be executed before the next can be executed (Think one instruction says add 2+2 then another instruction says add 2 to the previous answer, I need to figure out what 2+2 is before I can complete the next instruction) so I can not run a single thread across the two cores of our CPU because each core would always be waiting for the other one to complete something which gives us no performance gain. So the issue is that our program only has one self-dependent thread and thus our program gets no faster since it is best to only ever run on one core.... However if we have two programs running then they will get faster since they each have their own independent thread and thus can each run simultaneously on the two cores, YAY! As time goes on and CPU's start to add even more cores developers decide to try to start making their programs have multiple independent threads of instructions so each thread of their program can be running simultaneously. This is hard as to really get a real performance increase each thread needs to be as independent as possible and rely on info from another thread as little as often. Thus not every workload or program benefits from this change but for certain things, like say rendering a video where we can split the video up into multiple threads and give each chunk of the video to a different thread, or maybe say a video game where we can have one thread compute what the AI is going to do and maybe another handles getting all the textures of the game ready for the GPU to use (Again oversimplification but that gives you some examples) then we start to see a single program benefit from having more cores. One last tiny thing needed to really understand SMT is what is referred to as pipelining, basically every modern CPU will actually start the next instruction before the previous one has finished. The best example of pipelining I can give is actually doing a large amount laundry. So when you do the laundry you need to take your laundry to your washer, then you load it into your washer and run it, when it finishes you then load it into the dryer and run it, then you fold it (maybe) and you put it away. So if we time those four steps lets say it takes 5+30+60+15 minutes to run so it would take 110 minutes to finish. Well my momma did not raise no fool so I of course after loading my clothes into my washer if I have more to run get the next load ready to run and when the previous load is done load in the next, sure this means that the time it takes before my first load is done is longer (This is referred to as latency) since the dryer holds everything up (So instead of adding up the total time of the steps I now have to account for the the dryer take 60 minutes and is thus holding up each step so I just do everything every 60 minutes so 60+60+60+60 mean a single load takes 240 to be done), overall I get more laundry done (This is referred to as throughput) in a day since I get a load done every 60 minutes instead of every 110 minutes. This is basically what pipelining is, each instruction has multiple things that need to happen before it is finished, so if we are willing to trade of latency for throughput we can do more instructions per second. So now for SMT. As Mark the CTO of RISC-V (A group that makes processors where every instruction take the same amount of time to complete) likely explains in his interview SMT is basically a hack to try to squeeze extra performance out of X86 CPUs that are running specific highly multithreaded workloads by taking advantage of the fact that since X86 is both a CISC based architecture (Different instructions can take different time to complete, of which some can be quite a bit longer) and that modern CPUs use pipelining like explained above we end up with some parts of our core that are done quick but those parts are waiting on the other parts to finish before they can move on to the next instruction thus wasting a lot of time. So the idea is what if we made two of every part of the core that takes a while to complete and just alternate which one we use. To go back to our laundry example if I bought my mom a second dryer and she alternated between which she used then she could get twice as much laundry done in a day (P.S. Don't but your mom a second dryer) since she could unload each cycle from the washer (Which is done every 30 minutes) the moment it is done into one of the two dryers (Which take 60 minutes to run but now there are 2) thus doubling her throughput of laundry done per day. However while we will get twice as much done we run into a problem, just like adding more cores a thread can not be run across the duplicated components since we may need the data from the previous instruction that a component has not computed yet. So just like this adding more cores to the CPU this does not improve the speed of my thread but we could run two threads at once on a single core. We refer to this as: Simultaneous Multi-Threading or SMT for short since I am able to run two separate threads simultaneously on a single core. This means we basically increase our theoretical max performance by nearly 2X per core and our CPU will now perform like if we added twice the amount of cores to our CPU but didn't have SMT, but only for certain multithreaded workloads, and in the real world it is not quite that much (This is why the old i5 with 4 cores 4 threads was still faster than the i3 with 2 cores but with SMT so thus 4 threads). Also for single threaded performance it has no improvement (And can sometimes lead to performance decreases) this is why Mark talked about that for certain workloads turning off SMT could lead to performance improvements, although SMT has matured enough that this mindset is a bit antiquated although that did crop up again in the early Ryzen CPU days and I am sure there are still some High Performance Computing workloads that you still want to turn SMT off for. TL;DR: While this is a gross simplification a multi-core CPU is basically duplicating the whole CPU and gluing them together treating them as one CPU. A CPU where each core supports simultaneous multi-threading is basically where we duplicate the parts of the core that take the longest to run, run it twice as fast, and treat it as two separate cores. This is also what this video tries to say: Basically, while Apple's new M1 has great single threaded performance that mostly only applies for applications that only take advantage of one thread. For more power hungry programs that have been made to take advantage of multiple threads X86 is slightly still king..... For now. So if you are browsing the web, writing a word doc, or paying bills on the internet you should rest easy knowing that M1 Macs are probably gonna do that faster than their Intel brothers, but if you are exporting video from premier then maybe you should hold on to your Intel mac for a bit longer.

[@tetow2484]

Its so nice to have someone talk about processors that actually knows something about ISA's microarchitecture

[@youngwt1]

I always thought running multiple threads on one core was called HyperThreading (intel) or SMT (AMD)? For intel at least that was usually a value added feature for the i7 line

[@steakman7662]

SMT is the technical term, "hyperthreading" is intel's marketing term. Just like AMD calls their L2+3 cache "gamecache" when it's really just regular cache.

[@TechTechPotato]

SMT isn't a band-aid either. Cores are designed with SMT in mind, and you make very important design decisions based on that idea. To suggest it's a band aid would mean that you don't consider SMT at the start and add it later. It's very much NOT the case.

[@LunarLaker]

@@TechTechPotato Hey Ian! Are you also here because of Andrei?

[@noelsantos7850]

This was so incredibly in-depth.. I actually understood some of it

[@MaybeSomeoneful]

Very detailed overview, that's what I really like to see

[@Peizxcv]

You take Wccftech seriousely?

[@MihneaStoian]

Love the M1 Air so far. Got it as a stopgap while waiting for the M1X/2 MBP, but I have to say that the fanless experience is something else, and it's surprisingly capable. As far as the SD card, all on board.

[@Touchgrassplz]

i feel like he's shouting into my ear T_T

[@seshpenguin]

Probably cause of the echo in the room.

[@AaronHilton]

Yeah, feels like the audio is so insanely compressed that it’s a continual shout. Also, no hold beats to pace the content. Need to let it breath a little.

[@utubekullanicisi]

@@AaronHilton Those are valid complaints. Still though, really enjoyed the content.

[@Chris.Wiley.]

The content is good, but he really needs to dial back that delivery. Slow down and stop yelling!

[@sirkisboy]

Too much base

[@WesPerry]

Wow. This was so much better than I thought it would be.

[@solom465]

Great video! Bought an M1 Air a few weeks ago. Been impressed by its performance and battery life.

[@GFClocked]

This guy is so great I could watch him all day. Seriously, great job. I subscribe just for him.

[@exlomo]

would you like some more compression on your mic recorded in an echoy room sir

[@dhananjayraje4572]

Just wanted to stop a moment and just write this comment to appreciate you about your very in depth research and easy to get explanation. And I like the jokes in between too. Nice work. Keep it up.

[@TheBackwardKid]

Awesome video, my M1 mac mini came in earlier this week and my god I love it, it's a gigantic upgrade from my nearly a decade-old hand-me-down Macbook Air.

[@VincentJGoh]

I got DoS:2 running on my M1 Mac Mini ONCE, and now it won't start up anymore. :(

[@arumaakadalamudhan.t9900]

Clap for the man right here, his way of explanation was just awesome 👌🏻

[@Jalae]

ya he's such a great *checks notes* /writer...?/

[@whigmalwhim]

What about the M1s x86 decoder? I have a hunch that some of that work isn’t just software and is hardware accelerated

[@TheCJUN]

When will Apple present the desktop versions?

[@steve8547]

Sooo much love for this series. It really makes some techie deep dive information really accessible to the enthusiast.

[@mygunzy]

Now I like to know how u explain tesla Processor plz.

[@Deez-Master]

Out of curiousity is there any reason you refer to their chips as RISC based rather than ARM based? Isnt ARM a bit more specific and descriptive, even tho ARM itself is RISC based?

[@AndersonVenturini]

Wow! Awesome content, Chris! Congrats!

[@nokind90]

impressed with the work you/the team put in. Really detailed info with nice flow

[@lavisharma3210]

Hey Engadget folks, Is it possible to have an upscaled episode every week ???

[@RegitYouTuber]

That's an incredible video. So in-depth yet with a couple of re-watches of certain sections a complete tech dinosaur like me was able to cotton onto it

[@JensHeuschkel]

I think there is something mixed up. RISC processors, in general, are not designed to execute an instruction in one cycle, by any means. Typical ARM processors in our phones are multi stage processors, meaning they execute a stage in one cycle but a instruciton has to pass all the stages the processor has. This is mainly done to improve the clockspeed of the processor and is common practice for nearly every processor design. Anyhow ... compared to CISC instruction, a RISC instruction is way faster (which should be sufficient to support your point)!

[@teahousereloaded]

The high bandwidth between components in the chip seem to allow for crazy perks, like previewing 4k 120fps Sony footage effortless. I would love to hear how that works.

[@Teluric2]

I do it on my amd 16 core like butter and its a 2 year old pc.

[@dbtest117]

Mac's been on quite a SISC RISK jurney. Starting with 68k SISC, moving to PowerPC RISC, then to Intel SISC and now back to Arm RISC.

[@duran9664]

Why you have never mentioned the heat and how apple is dealing with it 🤔

[@DomDomMartin]

Doesn’t the single core/single thread comparison not matter because its supposed to emulate tasks that aren’t written to run threaded or parallel ? So a cpu having two threads on those task is a disadvantage and pretending that all the threads in a core on intel matter in this use case is disingenuous, but then also the m1 relatively keeps up in multithreaded tasks performance so apple wins in both use cases so the difference doesn’t really matter

[@MV_96]

Didn't expect such a thorough video. Thank you!

[@WhenDoesTheVideoActuallyStart]

SoC actually just means "It doesn't only have CPU cores". A couple of years before 2010 Intel was calling their CPU's "SoC's" because of the iGPU. Ryzen has also been called an SoC as it integrated an on-die chipset.

[@mahadevovnl]

I switched from a first-gen 16" (Touchbar) MBP that was fully upgraded to a 2021 version of the 16" MBP (not fully upgraded) and now also to the 2021 MacBook Air (8/8 16GB 512GB) with M1-chip. I tried the 8GB RAM version as well, and 16GB is noticeably better for the vast amount of things I do on the laptop. What I'm waiting for is the 16" MBP with an M1 or better chip. Hopefully also with their OLED-replacement (mini-LED) technology; hopefully also with a higher refresh rate (120Hz or higher), hopefully also WITHOUT the touch bar, hopefully with a better webcam, ideally also with Face ID next to Touch ID, hopefully also with black bezels instead of continuing the trend of those new iMacs which have white/grey-ish bezels. But since I usually keep the laptop closed and just plug it into my screen I'm just hoping for a nice looking (wired!) keyboard with Touch ID.

[@goofynigiri]

Great video man, good work! I really appreciate that you put all this effort to bring us a video this much rich with details!

[@AbdulelahAlJeffery]

super informative series ... much much needed.. Thanks Engadget

[@djtomoy]

Would be interesting to see who they think there users are and if they need all that power, is everyone using a mac doing intensive tasks with it? Be nice if there was a cheapo version for just web browsing and using excel

[@diskyariajetmiko]

"Pls bring back the SD card" Apple: should we remove the thunderbolt entirely and charge the laptop wirelessly?

[@sloanNYC]

What about RISC makes it single threaded vs CISC? I thought an advantage of RISC is that the instructions are all the same size and easier to graph and execute in parallel? The consistent size allows the ability to break everything up and do great out of order execution. So what in the instructions themselves prevent multi-threading? Or is this really the difference between a thread = a program and parallel execution being something else entirely?

[@SimonBuchanNz]

Near as I could tell, every time he was saying "multithreading" he meant "hyperthreading" - that is running two threads on the same core at the same time (The M1 chip has 8 total cores, after all.) This can increase utilization because CISC tends to have a lot of instruction dependency blocking the core from using out of order execution to fill up the rest of the execution units. Strictly RISC could do the same, it just seems that it doesn't provide enough benefit to be worth the extra silicon with the more flexible instructions.

[@gunayorbay]

3:20 guys, single threaded apps don't magically start using two threads. cinebench single core test is still more representative of single threaded applications.

[@ZeoWorks]

Honestly, I'm glad to see people moving away from the x86 architecture. While it's amazing to use, it was also invented decades ago. We need innovation.

[@phitc4242]

x64 🤡

[@ZeoWorks]

@@phitc4242 I don't think I need to explain myself to an adult with an anime picture. :)

[@phitc4242]

@@ZeoWorks no you don't you, biased people are not good teachers!

[@Seirin-Blu]

@@phitc4242 64 bit and x86 are not mutually exclusive 64x or 64bit refers to the instruction set where x86 refers to the chip architecture. There does exist x86-64 or x64 for 64 bit systems though. There still exist quite a lot of x86 32 bit systems

[@Diadras]

@@ZeoWorks HEY

[@kishores7034]

Excellent man This is the most awaited tech series on KZbin Very well put I just felt that you compared the M1 to INTEL chips Hopefully you would also included latest AMD chips too in future Zen 3 deepdive this way is expected from you

[@BAmalakas]

I wonder how much can be squeezed on an SoC without compromising yields

[@SynysterNick]

Great color grading throughout the episode! This series is the BEST! Keep them coming!!!

[@RyanAmparo-tl]

So did you upgrade to the new m1 pro/mac yet?

[@arumaakadalamudhan.t9900]

I alone wrote 4-5 comments for this single vid. Cause it's just 👌🏻 Fun fact: This is my first vid that I'm seeing from this channel. Great work guys.. gg

[@chengcao418]

The content and explanations are exceptionally clear and professional (coming from a technical background in micro architecture), unlike most other media where the core concepts are shoved around and more or less twisted. Great job!

[@YuFanLou]

This video is embarrassingly disappointing coming from Engadget. The cited WCCFTech article misleads its readers as well as the video author into misunderstanding of modern CPU architecture and performance benchmarking. We have always benchmarked single-thread and multi-thread performance; there is no basis whatsoever to measure “single-core” performance with 2 threads, since for software developers using 2 threads versus more takes a similar amount of additional effort and care not required by single-thread. Unfortunately from the comments, many more people are being misled by this video. This is the most dangerous form of misinformation: a bit of truth, like reordering, but presented with a false methodology. I’d rather watch more of the interview with the RISCV CTO for his experience and wisdom.

[@tjhrulz]

This. Also the snippets we got from Mark even alluded to some of the issues with this reporting. SMT is essentially a hardware hack that is used to try to squeeze extra performance out of certain multi-threaded workloads, although as Mark alluded to not all multi-threaded workloads gain from it (although I will say it has gotten better over the years). I'm not going to trash talk the video to badly, SMT in the early days could sometimes hurt a single-threaded workload (although toggling it off and on on a modern CPU doing a single threaded cinebench score the variance in your results would likely be within margin of error), also many modern x86 chips go after a highly threaded chip as opposed to squeezing an extra few megahertz so it is definitely fair to say that you need to compare both single-threaded and multi-threaded workload scores. It's just it clearly needed someone with an architecture background to actually help with the writing and fact check the video.

[@keithkamps77]

Great video. I haven't tried the M1 as of yet, I'm still happy with my current MBP's.

[@Medic_V]

So does this mean that the new MacBook pro won’t support windows?

[@e21big]

11:59 cat in the background reflect exactly the attitude of none-apple user

[@suntzu1409]

I am dying

[@godkingmonkey20]

One of the main reasons that the M1 can do things slightly faster is that it's a reduced instruction set computer processor. Even if a processor is multithreaded a single thread process is still only a single-threaded process.

[@acoolonthebeat]

This was the most in-depth video I’ve seen on the M1 macs and I loved it

[@JustinZymbaluk]

The thing to keep in mind here is that even when the advantage “evaporated” you’re comparing M1 which is the lowest end and worst apple silicon chip we will ever see to some of the highest end x86 processors available. The M1 still punches way above its weight class

[@sidbrun_]

Yeah this is pretty important. So far the M1 has just been put in their lowest tier laptops, so really people shouldn’t necessarily expect high performance because the laptops they put them in aren’t meant to be high performance anyway. The bar has been moved, the maximum has become the new minimum.

[@Teluric2]

What advantage? M1 is a 5nm chip vs 14 or 10nm intel chips.

[@ArchieMcGeoch]

But how many years could a chip like this last? 1, 2? After that it just gets slower right?