The Secret Behind Apple's New Silicon: The Next Big Move is coming

[@AnastasiInTech]

Let me know what you think

[@danielbabu6010]

wait 2 become one(1) !

[@adamgibbons4262]

I wonder if apple will be able to make their own version of CUDA now

[@salmiakki5638]

i think this video title is wondefully topical given the season, but ultimately lacking evidence for such a statement made. video titles a bit more descirptive of the topics would be very welcomed

[@andreypopov6166]

@@adamgibbons4262 they have Metal/Performance Shader for general computing. But it is limited to their devices.

[@jonacacarr3839]

I watch your videos as soon as they come out. Did you increase the speed on this one, or was it the caffeine? Ha Ha

[@haseebasif100]

Im so glad Anastasi goes deep into technical details of Apple. Rather than just subjective User experiences that most youtubers focus on. There is alot going on in devices that meets the eye.

[@jameshobbs6092]

RIGHT!

[@legoplaytime1823]

Developing data centres with custom chips, instead of relying on existing cloud providers like AWS or even buying hardware from Nvidia, is perhaps Apple's biggest under-the-hood update for the new iPhones

[@petchlnwzaaa]

Yes, and this is very bearish for Nvidia. Not just Apple. Google and Amazon also designing their own chips so they can stop relying to nvidia

[@lindsaybruce1396]

The history of Apple silicon actually goes back even further to the 1990's when they partnered with IBM and Motorola over the PowerPC range of RISC processors. Apple has a long history of working with silicon partners to produce chips specific to their platforms. Since the very earliest days of Wozniak working in a garage, Apple have always had impressive software and hardware integration - something only really possible if you control both. This is why many in creative industries still favour Mac hardware, because it works so well with Apple's application software. See Final Cut Pro render times versus Adobe Premiere on comparable x86 hardware.

[@seeibe]

ngl I got an M2 mac mini recently, and for all the good things I had heard about "Apple's hardware/software integration", I was appalled by the performance of Visual Studio Code on a 4k screen. It legitimately felt laggy to type on. Now I get why Apple users created a whole new code editor just to reduce typing lag! Meanwhile I'm on linux with an nvidia GPU, a circumstance where the manufacturer of the hardware seems on a quest to make my life miserable, and everything runs smooth as butter.

[@lindsaybruce1396]

@@seeibe You wouldn't expect a Microsoft product to take advantage of the full range of hooks in an Apple OS - and historically Microsoft did a terrible job of coding for the Mac platform (there are some exceptions). The Mac native IDE is Xcode which has been around for over 20 years.

[@seeibe]

@@lindsaybruce1396 xcode is so irrelevant, Apple fanboys will rather create a new editor to replace VSC than use xcode

[@JohnSmith-pn2vl]

@@seeibe ROFL VSC

[@Deathrape-if4kl]

​@@seeibe Apple is just a 'protest' brand 4 phag itz. iOnly know about it because iMake $ investing in them now & then (through my mom who is a statistix expert) & iWoz gifted an iPhone XS. The whole 'ecosystem' is a $kam = like almost every 1 of their millions of apps is $pamware or fraudulently marketed in some way = they lie pretending it's free, or force ads into everything. Something as simple as an app 2 customize yor wallpaper is full of banner ads. The only way 2 use an iPhone is with NO data connection other than 2 download or update apps. When U R 'offline' U it blocks ads on most of the kraps from their krap$tore =) iAsked them about an app 2 automatically answer the phone & then hang up & add them 2 the block list, as iOnly call certain people iKnow, but they say there isn't 1. Somebody should make it = automatcally block 'everyone', with no sound interruptions =D iDOn't want them 're-calling', but 'blocked' =;-o iGet hundreds of 'junk box' mails a day & don't want the same with a ringer LOL

[@zzolaman]

Your analysis is an Oasis of crucial information. I have subscribed and I Don't Go Away. If you missed her analysis, Don't Look Back in Anger! Even though she's Half the World Away, you can Acquiesce, Listen Up, and get much smarter Little By Little. C'mon guys, She's Electric!

[@AnastasiInTech]

hahah We see things they'll never see! :))

[@zzolaman]

@@AnastasiInTech Yes we will, thanks to you! I remember your channel was less than 20K subs when I joined and now you're approaching 10x! It speaks to the quality of your content and your ability to deliver. Thanks so much!

[@skinthekat0530]

The CEO of PA Semi gave me an office tour when they were waiting on first silicon. Apple bought them a few months later - I think before their product launch. The chip wasn't ARM-based, but the design team was worth their weight in gold.

[@sierrasatan]

If Qualcomm charged 100$ for each modem chip to Apple, they would get 22.5B revenue from Apple for the 225M iphones sold each year. From Qualcomm annual report it looks like they are getting 9B from their biggest customer (they don't name the customers) and that includes licensing fees and chip revenues together. I don't think its anywhere close to 100$ for modem chip alone. But otherwise, very informative and detailed video as usual.

[@EnlightenedSavage]

So 40ish dollars a chip.

[@litical3719]

good good point

[@quantumbacon]

multiply by 5. For 5G

[@sierrasatan]

​@@EnlightenedSavage That includes licensing, which we don't know how much it is. Maybe 8 like in the lawsuit filed in 2017. Dylan Patel claimed it is 12+ but even if it's assumed to be low 8, I am not sure how much they will save building their own modem. From all the rumors so far, their modem will struggle to be area efficient, so they won't be able to integrate in their already large SOC that easily. Building their own modem is more of a licensing play I suppose and not motherboard efficiency.

[@imconsequetau5275]

So, around $40 per iPhone?

[@RuedigerDrischel]

Not an expert I like your explanations which make complex matters digestible in a way I can use them in my field of expertise

[@davidlanfair]

Whoa, so much info. Thank you!

[@Saderue]

Thank you for your thoughts, we tend to forgrt how innovative Apple is when we listen to standard 08/15 tech you tubers.

[@takeshikovacs02]

This is a great video Anastasi. With sites like Anandtech dying, I treasure the deep dives and industry perspective you bring to your content. In an age of shallow lists of affiliate links, people like you, Dr. Ian Cutress, Gamers Nexus and chipsandcheese are very special. Keep up the great work!

[@paulg123]

Great video. Great insight. A very informative video. Loved the Oasis references!

[@JoeBurnett]

Great analysis! Thank you!

[@pauls1ngh]

0:30 I may have briefly fallen in love 😍 That was too good 😂🤝

[@pauls1ngh]

And not with Liam u aholes 🤣

[@tranquilitybase6417]

Great synopsis! Finally I learn something new about Apple chips and their history!

[@matthewjfoster1]

another great video. loved the Oasis tribute :)

[@StevePaulSounds]

Really interesting. You’re such an insider!

[@TorukMaktoDracarys]

May I request a video update on Photonic Chips, since AI consumes so much power. Who is at the forefront of this technology to incorporate as quickly as possible. Thank you for ALL of your Awesome videos

[@RickBeacham]

I think there is a Chinese based company that are ahead of the curve on this?

[@peterclarke3020]

Very difficult to do AI with photonic chips, because AI requires lots of multiplications. And although multiplication can be done by looping addition, it’s complicated.

[@johnmajewski1065]

Thanks your Video insight and uniqueness is welcoming in the tech changes which will effectively effect everything! Keep up the wonderful work! 🎉

[@romen3]

Very interesting video, and well explained. Thanks for sharing!

[@kentbyron7608]

Anastasi long time subscriber here. Please make a video biography of your life story: your birthplace, your parents, your childhood, your education, your hopes, your dreams, your fears and your struggles. You could be a stellar engineer-science-math role model for little girls around the world. You are someone that can change people's lives - just by being you. Brilliant! Bravo! May you live a long, peaceful, healthy and happy life. 🌹

[@bruinscat]

Hi, @AnastasiInTech. You mentioned designed a new chip could cost $10M-$500M. I would love to see a video covering what that entails, like how do they start? What kind of researches , must they pursue. They must use software for designing, correct? So where does the cost and challenges come from? Thank you for your work & research! I love your videos!

[@Argelius1]

This is absolutely fascinating. Apple is a remarkable company. While profit is, obviously, a huge driver for them, they seem to always also want to continually make better products that preserve the privacy of their customers.

[@mjthethird1]

Wow! a4 45nm now we’re at 3nm that’s amazing!

[@jameshobbs6092]

RIGHT!

[@enviropediaxr6007]

This was really a treat to watch! I think the ARM-based Apple chips (SoC) could eventually outperform NVIDIA. Here's why: First of all, Apple doesn't need a PCI bus, NVIDIA does. Secondly, NVIDIA cannot access CPU memory, NVIDIA GPUs must have separated separate memory address spaces, and they require an external bus like PCI, which introduces significant latency. The best of everything is to have as many of your execution units on the same die, and in the same memory address space. I see this an an insurmountable problem for NVIDIA in the future.

[@ianstorey1521]

My hunch is that the time for the signal to travel on an external bus is becoming a limitation. Simple back of an envelope calculations, using the speed of light, indicate that external bus speeds are reaching a limit in the Gbps range.

[@michaelkhalsa]

​@ianstorey1521 Yes, but if I understand correctly, the bus latency does not affect the streaming speed, only the first bytes and handshakes. One would hope that small bits of data are retrieved mostly from on chip cache.

[@thereddog223]

It will be 15 years

[@kazedcat]

Graphics is limited by memory bus. So unless Apple starts using HBM in their microchip architecture they will not surpass Nvidia.

[@enviropediaxr6007]

@ehtasam2080 and none of that computing firepower can access the CPU without slowing it all down.

[@maartenb100]

I don’t think you’re right about Apple servers. I understand that they don’t have any data sources to train an LLM with, like meta, google and X has. They have also promised not to use iCloud user data or scrape other sources like KZbin illegally. This means that they probably are not going to train their own models but buy or lease pre-trained models from others, or outsource queries, like they do now with ChatGPT. This in turn means that they don’t need GPU-like chips but only chips suited for bandwidth and search, which the M2 chips are quite good at. This would explain why they’re not buying nvidea chips or boards, but increasing their M2 purchases.

[@bruxi78230]

Yes, what they've said is that the server software they built for AI (Private Cloud Compute) will only answer user requests, they won't be storing those requests on their servers. They will user third party companies to verify this fact. One enormous advantage this provides them with is ability to use their own data centers which run on renewable electricity. While every other provider in the industry must procure new sources of expensive electricity, Apple is able to generate their own supply of energy.

[@paulwharton1850]

Excellent ! A Class. Many thanks for making this video - I learnt a lot.

[@RickBeacham]

Apple started working on ARM back in the late 80s. And released a product named Newton that was a SOC in 1993.

[@dchdch8290]

Thank you for these insights. It all makes sense now. However, we all know what happens after such a large vertical integration…

[@viktoriyagrudyuk3154]

I’m not in the topic, but the information was so nicely presented that I was curious to hear everything till the end. thank you 🙏

[@alanbaker6727]

You mentioned 'based on a design', so I'll keep my comments as broad as I can. I have been retired for quite a few years now, and I know nothing of the inside workings of current electronic products! When I was employed, information passed on to me indicated that some manufacturers use designs developed in agreements with other companies. Sometimes this is obvious, open up the product and you can see the logos of the other companies involved. Sometimes this is kept confidential! So, only the companies involved in the agreements benefit here! The designs agreed upon by the outside companies will benefit from revenue, and if their product integrates well with the rest of the final product then there will be more work lined up for them in the future. If it turns out to be a runaway successful product, the manufacturing company will make a fortune and its reputation will grow. The staff within the company who designed that specific part of the final product cannot reveal the name of the manufacturer to the outside world, as a result of the confidential agreement. If it was a good design, then their staff possibly will be credited within that company, but sometimes not so, as it's just another project for the design engineers, their team leads, the authors, and managers!

[@wheelyflac]

Qualcomm is quality communication, modems is their specialty. Apple never gonna catch up haha

[@peterclarke3020]

Apple tried - it turns out that modems are a lot more complex than most people think - especially for widespread compatibility.

[@wheelyflac]

@@peterclarke3020 yeah there's so many carriers around the world

[@wavefuse]

Please go work for Apple , , would love to see you presenting at their events !

[@Steamrick]

I don't see Apple running AI Inference on their current silicon offerings... LLM inference is strongly GPU memory bandwidth bound and compared to the 3TB/s to 5TB/s current datacenter GPUs are offering, even the 400-800 GB/s that Apple's biggest M-Series chips are offering is peanuts. As far as Apple designing a modem / radio: They haven't succeeded making a competitive product in the last 10 years without infringing patents and I don't expect that to change.

[@HansSchulze]

Inference needs an industry-wide inversion to fix memory bandwidth requirements. Von-neuman doesn't scale. Apple cannot put 5G into SOC because the layout rules aren't compatible. Even soft radios (SDR) which is power inefficient barely fits in one silicon.

[@kazedcat]

Apple will build inference chips because it is cheaper than buying Nvidia chips. If Tesla can build their own AI chip then surely Apple with more resources can also do so. For modems that is a hard problem but standard changes so maybe 6G can work around the patents that are blocking 5G chip development.

[@maxcolvin9209]

I would love to see Apple redesign their motherboards so that you could swap out the SOC’S every year or two. It would give them a new revenue stream also.

[@OggVorbis69]

I like how the history of apple cpus for computers started from power pc completely omitting the Motorola era

[@null7936]

6502 ?

[@OggVorbis69]

@@null7936 true too. But 6502 was never replaced with other architecture in apple ][ and its variants. I was thinking about 68000 from first macs

[@bzahowell]

Or the fact that they co-founded ARM?

[@doodlebug1820]

@@null7936dman im old

[@jkjkok7754]

I didn’t think it was possible to be super model beautiful and scientist smart at the same time.. your breaking the laws of attraction!! 😍🥰

[@Greggozzo]

Your videos are trully fascinating. Learning so much from your materials.

[@DrinkingStar]

Thank you for this very interesting update.

[@Stu2be2]

I could literally listen to your voice for hours ❤

[@HappyPandaBear73]

Absolutely Fascinating & Thank You!👍🏼🙏🙂

[@rickkarrer8370]

I'd love it if Apple started selling servers again, ARM64 servers would be great.

[@kashyapchonekar5437]

Bluetooth 6 can do the same thing so uwb will be redundant i think

[@AnastasiInTech]

well, good point :) if channel sounding technology works as they envision it, let's say, there will be a competition between BLE and UWB. So far UWB is the only one that works for ranging application

[@StarNumbers]

Well said. Apple, and the US in general always thought that ALL other countries will be eager to sell stuff to the US. But guess what ...

[@BilichaGhebremuse]

Great

[@nofeerz]

such good information you give in 1 minute what takes me many many pages to find out and you do it best thank you

[@lilblackduc7312]

Delightful young lady, and smart, too! Thank you for explaining this, along with the slides & illustrations...🇺🇸 👍☕

[@Indrid__Cold]

2:39 I always wondered what "bionic" meant in terms of its iPhone chips.

[@rahulav4009]

Super amazing video ❤

[@lucysluckyday]

0:51. The density of those smt caps and resistors on those boards is eye-opening.

[@mshparber]

Super interesting!

[@mikewise992]

Thanks!

[@HeyNickJ]

Forgot about the R1 chip with Vision Pro and M2 chip in Vision Pro as well.

[@TheMetalMag]

Thx again for your very interesting video. Did you eat the apple ;)

[@Patrickckst]

I liked what I saw today in this video

[@jasonwidegren3211]

Love your content. Maybe I WILL get a new iPhone! 😊

[@wkrapek]

From what I understand the flops-per-watt are much better on the Apple silicon than Nvidia’s. Synergize that with then on-phone work - probably a lot more than on Samsung phones - and Apple’s servers can probably keep up. Or at least get close enough.

[@GerdCastan]

Apple not only utilizes bionic cores. The cpus also have dedicated AI areas, and obviously the graphics cores can also run AI code. As a developer, you can tell the operating system “run the model wherever you want”, or give preferences.

[@joseteixeira5900]

This a decent review. Other than this just only a kind of Anandtech review about sub architecture. To mention the M4 and A18/A18 pro are the first cpu with armv9 instructions set, so the 10 year armv8 ends now, since the A7 (iphone 5s).

[@darkstatehk]

I love this content, but man, Anastasi's voice is the ultimate ASMR, the hairs on my arms are standing up! Sorry, I know that's a bit weird, but it's the ultimate mix of great content and delivery, it's perfect!

[@EpicLib]

Ha ha, same here

[@jibcot8541]

Its great, cutting edge Chip/AI news and ASMR at the same time!

[@jaccurtis5789]

Been interesting seeing Apple’s moves into designing their own silicon they use now, and also interested to see what they do in terms of datacenter computing, seems (to me at least) a lot like what Tesla did a little while back starting with HW3 in cars then building the supercomputer chips & system for training

[@stargalacticfederation]

Chips and Transistors made with Graphene would be a big game changer for the Future of Electronics, it would use less power and Less heat.

[@Augustine-x5i]

You have a really good ear out for edge cutting news with an eye to see its depth. I appreciate your input on topics so far mentioned and yes...I think it's a data center optimization push with hopes for G6 speed & load. 🎼🤔🤨😎🎶

[@b.r.c9156]

Anastasia you are amazing, from where do you get all these informations!

[@karmatraining]

Datacentres running Apple Silicon is a huge step change, they are highly power-efficient and just overall extremely powerful chips. Even AWS is heavily investing in ARM, because they're both faster and cheaper to run, and the greatest cost for a datacentre is energy.

[@tahaatef8847]

very good ❤

[@joseph.cotter]

I could be wrong but I don't see them integrating 5g into the main processor due to heat issues. Anything that transmits a signal is going to generate a good amount of heat using current and foreseeable technology and I don't see wanting to put that into the same package as the main processor.

[@markmalonson7531]

Tell me why they don't make the handheld computers multitasking?

[@DaniilPotekhin]

Здравствуйте. Можете рассказать про разницу embedded-разработки в РФ и в Европе? Как у Вас произошёл этот переход?

[@scottwatschke4192]

Will Apple build their Fab facility and not rely on TSMC?

[@AnastasiInTech]

I don’t think this investment makes sense for them

[@bruxi78230]

TSMC is already building a new Fab facility near Phoenix. No way Apple would do that, they don't manufacture anything by themselves but always use partners. This policy was initiated in 1998 when Tim Cook came onboard.

[@shiccup]

TSMC makes it very difficult for new fabrication facilities to be profitable they are basically holding a monopoly on the technology but a low cost monopoly if they increase prices other fabricators can compete but Taiwan needs tsmc to be running at full capacity which is why they keep prices low. So i guess for now anybody that tries to make a fabricator will not be very profitable which is why Intel and others are moving semiconductors away. It's a long term play but it won't be profitable till the rest of the industry catches up. Or tsmc raises prices but if that happens Taiwan will be in a lot of trouble.

[@張榮華-z9o]

TSMC's US plant starts trial production of AppleA15chip With a yield rate of 70%, we can enter the full-scale mass production stage. Congratulations to the United States for being able to buy American-made chips.😊

[@shiccup]

@@張榮華-z9o lol American made chips that the profit goes to tsmc. I guess having the factory in the usa is still better as if tsmc falls we'll still have the fab

[@winstonsmith935]

And to think I met Sophie Wilson of Arm Computer back in 1983 in London, where I lived and worked. Used the first Arm Computer with Steve Furber. My first computer was the Acon Micro Computer In 1980, the first 32 bit computer .

[@springwoodcottage4248]

Interesting how many industries are moving to vertical integration for their business. Elon Musk made the idea cool & it has been a blessing for Tesla compared to the horizontal integration of many competitors. Apple’s biggest trouble seems to have been getting a substitute for Qualcomm without violating their patents, something that still escapes Apple, but if anyone can design a better chip than Qualcomm it is likely Apple. If they do it, it’s unlikely they will reduce the price if their phones, so the savings should help their earnings. Great video! Thank you for sharing!

[@100c0c]

Apple has always been like this. They just have the money now. And Elon didn't make vertical integration cool...

[@springwoodcottage4248]

Yes, Apple was always hardware & operating system, but added the capital or research base to make their own hardware, as you noted. Most, legacy auto (all ?), was fixated on horizontal integration, using as many suppliers as possible in the lowest wage country. When Tesla started, legacy laughed at Musk’s vertical integration & told him he would soon find out how to run a volume auto maker. VW in particular were scathing of his approach & now VW are closing factories & losing sales volume.

[@ws7001]

@@100c0cwhen was the last time Apple actually made a phone, iPad, laptop, computer station, accessories? Do they make anything?

[@張榮華-z9o]

Taiwan's MediaTek has already made mobile phone chips that surpass Qualcomm😊

[@張榮華-z9o]

It's also interesting that Intel, which is vertically integrated, is preparing to sell its fabs.😊

[@NGC-7635]

We all know you can sing better than 7:25, you just didn't want to distract us too much from the subject matter ;)

[@baileyestler1390]

Professor at top supply chain management university said his friend has had huge orders with Apple for data center batteries with strict deadlines and large quantity requirements

[@GetzAI]

I was hoping for more on the Server tech.

[@bobedwards8896]

any news on Teslas super computer? feel like i havnt heard about it in a while

[@emiel333]

Great video. Question: there are different GFLOPS and TFLOPS calculations. The CPU can be measured and the GPU. You said in the video that the Neural Engine of M2 Ultra adds additional TFLOPS performance. If we look at the Apple M4 SoC, how much is the combined performance of this chip? The GPU alone is not only faster compared to M2 (from previous iPad Pro), it also has hardware ray tracing, mesh shading, and the Neural Engine is capable of calculating 38 TOPS. TOPS are different compared to TFLOPS, though I hope you can answer this question. Thanks 🙏 in advance.

[@star1child109]

Honestly, beyond all this AI and chip improvements; what I’ve really been waiting for, that will be a milestone game changer for any and all Smartphones; is to have Solid State batteries and micro-LED Displays (not mini-LED which is night and day different). As a bonus for Apple, the staggering difference with Solid State batteries (which eliminates dendrites legacy) having a vastly superior overall capacity, power to rate ratio and lifespan longevity, means that they could stick Apple M# power chips in their iPhones rather than the power cautious H# chips. The micro-LED display upgrade can’t come soon enough as that will eliminate the elephant in the room of slow eventual screen burn (as well as vastly superior display quality attributes).

[@mostwanted2000]

The most beautiful presenter! 😉 It's hard to focus on the content 😅

[@davidbain8921]

Could Apple use Groq or Cerebras for Apple inference services?

[@krishuntzeroable]

Wait what did you just say… I was completely distracted by you as a person. 😍 But really though, I appreciate the videos as always. Wish I was surrounded by a bunch of people like yourself who follow tech.

[@johnosborn1793]

Anastasia what a great video, but why didn’t you mention the M4 chip for the iPad Pro

[@johnosborn1793]

Thank you very much.Praise the Lord ❤❤❤

[@seriessplayer62747]

It’s a very pointless chip

[@williamgidrewicz4775]

Maybe what they all develop next is some sort of quantum dots or structures chip.

[@xz2bzy804]

Anastasi how about you doing a review of the engineering part of The IPhone 16? Analyzing all its form and functions? 😊

[@ikust007]

0 trust. Honey pot of information more and more. Thank you very much for your work !

[@ScienceRaven1138-du1mw]

197 000 subscribers! Your channel is the most intelligent ever it's the best semiconductor channel actually. You would have 1 million if you were in cali... YT is biased to keeping funds in the USA and Moscovites. Garden robots are my thesis at the moment... CPU's can object ID about 40,000 photos every day on 2 cores (RPi5), the free object ID softwares are limited to 100 720p squares and cost 1 megabite of RAM... Still, this is enough for basic garden robots, so now I am researching mechanical engineering, because 25% of our planet is a monoculture with chemical food and I want to change that ASAP. research priority...logical? 1.5 billion humans are employed on farms, and it's hard work. Other side of Alps here, France :) see ya future intel boss!!! Neuromorphic chips from China can take over the world, it's random.

[@h069401]

You sing! More, more, more. Maybe?

[@kennethroberts2748]

Smaller cloud servers focused on commonly requested promps, using less electricity and placed near local community battery farms

[@KENMASHICK]

I really really want to know if this new iPhone along with Apple Intelligence is a leap forward or an incremental step. If it is a leap forward I want to buy it now and not wait until next year, and so will many others. This has implications for the stock which I am considering buying.

[@sparkpaul]

Same here

[@kazedcat]

It's incremental. As far as I know Apple has not changed the issue width of the core which is needed for big ILP improvement.

[@thescfiles1273]

Afaik apple is going to test their 5g modems with the se4 in March

[@spencergibson853]

I think the question of what chips are in their data centre, could be answered by asking why there were no M3 ultra chips. I know there woukd be a high loss rate but maybe that’s why they did not release an Apple Studio based on the M3 series

[@KaiseruSoze]

Apple charges 30% to list an app on their site. It's just listed. That is, you pay them to paint their fence.

[@scottwilson4078]

would love to know if A18 uses ArmV9 ISA

[@theMuritz]

The next step will be Apple running their own fab

[@aquavitae3824]

*Helium enters the chat* "Are you sure about that?"

[@fredrikbergquist5734]

Speaking of Steven Jobs legacy, the aquisition by Apple of PA Semi might be the last genious move by him. Wonder what they will do when other companies do the same. Fun fact: PA Semi designed an Amiga processor way back! These advantages are important in phones and watches but not so much in laptops. The new Rysen 9 AI has better benchmarks than M3 in everything except onboard graphics performance. Steamdecks has also good performance in many ways, who is making their silicon?

[@derekcorcoran8090]

With the chip design team that Apple has, I’m sure it’s not a huge leap to go from edge based NPUs to scaling the architecture for larger performance. Albeit, in the mean time partnering with the various AI providers will allow a smooth transition to happen until they are fully up and running on their own hardware. I don’t see the current architecture for M series chips being the right one but scale up the NPUs and GPU with some modifications like local memory for NPUs / GPU (because as far as I recall this is a major resource constraint for all AI chips). However, I’m curious to see what real software improvements we will see with AI. I am actually pretty excited about the future with the potential productivity improvements with data analysis with AI in the engineering world. I think there are many more opportunities in the typical office suite space as well as more targeted engineering applications that will really see the benefit. Just think of all those manual tasks we carry out in spreadsheets, presentations, documents that could be much more efficient. 🤔

[@nathanbanks2354]

Google & Amazon/AWS have their own TPU type chips which consumers can't buy (though we can rent them). It makes sense that Apple can afford this too.

[@RickBeacham]

Also AWS has ARM based chips and M series based servers? I think so..

[@nowsc]

… what is a “stack “; I see you using the term here in this video, I see it is in many TESLA videos, is this a new computer term which has acquired some distinct meaning?

[@scanlly]

I love you ! ❤