The End of Big Silicon - Apple, Intel, Qualcomm 

Love your videos especially talks like this one. I do have a bit of feedback, please introduce your guest/segment instead of jumping directly to a question to an off-screen, unknown-to-the-viewer guest.

Intel / AMD was stalled with x86_64 and backwards compatibility. Apple merely scaled the designs from the iPhone SoC designs which are more ARM clones than actual ARM licensed designs. Even if Intel / AMD considered ARM and doing what Apple's done. It would produce non-compatible processors. Apple's secret weapon since Apple bought NeXT was the NeXTStep / OpenStep that evolved into Mac OS X and the developer API's. Developers if they didn't do a lot of weird stuff in their code were able to port from macOS to iOS/iPadOS on ARM64 with very little effort. In some cases, just checking a box and making some GUI changes. It doesn't work like that with Microsoft. It takes a lot more effort. Linux has always been running on most every CPU architecture out there and there is already a Linux distro booting and running on Apple Silicon. Apple still has a big head start and you have to understand Apple has internal Apple Silicon designs perhaps 5 years into the future they are developing. Some of which are waiting for FABs to be ready. Because Apple controls the OS and the Developer API's they can deliver radical change far easier than anyone else. No doubt Intel, AMD, Qualcomm, etc. will be attempting to change rapidly. The next 5-10 years won't be boring that is for sure.

The reason AMD has adopted the chiplet architecture for their CPUs and are starting to adopt it for their GPUs is that newer process nodes don't shrink down the size of memory cells and analog I/O as much as logic cells anymore, while the cost for every mm2 of the die stays the same regardless of how much of their area budget chip designers spent on logic vs. memory vs. analog. So AMD chooses to use flagship process nodes for the blocks that do the actual computation like CPU/GPU compute units, AI accelerators, display engines, media engines, etc. and TSMC's more value oriented 6nm node for memory/SRAM, where memory cell density is not too far away from what it would be on 5nm, but the price is much less expensive. It's not magic, however. Contrary to popular belief your yield rate doesn't increase when you take a single flagship process node die and slice it up into 2 pieces, the price stays the same, and you now have to accomodate for the die space that the interconnects that connect the two dies together will occupy as well. The value of chiplets is being able to mix and match more expensive and cheaper process nodes for different blocks of the chip. Keep in mind, 5nm still *is* better for memory and analog even if slightly, and the slight difference is only in terms of transistor density, there are frequency uplifts as well and stronger arguments to be made for the power reduction of newer nodes. That said, the power reductions are only relative to the power draw of memory cells on 6nm, not relative to the entire chip, where the compute units are responsible for 80-90% of the power draw. The reason Apple also uses a chiplet architecture for the M1 Ultra is because they figured being able to sell exactly half of the M1 Ultra's hardware would be useful, and doing that not by binning down a die that is at the reticle limits of TSMC's 5nm node (a theoretical monolithic M1 Ultra, if that was even possible) but rather making the M1 Ultra from two smaller dies and selling only one of those dies in cheaper and smaller products would be more suitable and economial. And they were able to create another chip SKU from the M1 Max as well, the M1 Pro, which is just the M1 Max with the bottom half of its GPU portion lasered off. There's actually only 2 separate dies that create the whole M1 family, the M1 (codenamed Tonga) and M1 Max (Jade-C Die). In short, companies that have the luxury to afford it with stronger market positions like Nvidia and Apple could stick with traditional monolithic architectures for most of their lineup a while longer to get those diminishing performance and efficiency gains, and only using the chiplet architecture to create multiple chip SKUs out of a single die or in cases where they're at TSMC's reticle size limits, while AMD settles for a little bit less performance in exchange of much higher performance/$.

Isn’t Apple already doing chip-lets with the M1 Ultra?

Very interesting! Many things kinda block quick development in tech but I guess that every company finds their way through. That‘ll bring competition and to us new products.

Ben did make one error here. Qualcomm still uses standard ARM cores. They tried their own a few years ago, but it didn’t work out.

Hey René. Could you discuss at some point how the technology in IBM's light speed chip could fit (if at all) into this conversation?

Ten minutes in, great conversation! We do miss your videos, Rene! 🙌

I enjoyed the interview. Blessings on your day!

Love this Rene, thank you!!

Very much appreciated this video. Thanks!

Another Great Conversation!

Just saw you on Tech News (Twit). After 3 days it had 5.7K views. Your video has over 45k views in 3 days. How times have changed.

interesting comversation, i always wonder if there a theoretical limit on how large a single die can be, imagine a modern 4 nanometer chip the size of 10 486DX processors, yes big but can still fit inside a modern gaming PC and would have more power than any modern computer. Yes, it would probably cost alot to produce but its only a scaling issue not development as we already reached 4nm capabilities.

Im reaching a point of limiting my CPU & GPU performance by undervolting/capping GHZ so I can get more efficiency/stability, this allows me to have pervious gen top performance but taking advantage of a new node to gain more efficiency, it works wonders on laptops. So I would say, buying the latest to get top performance of a previous gen at half power, is what motivates me to ‘upgrade’

Also, companies like IBM are doing research on new technologies like Optical Computing and are expanding their Quantum Computing. Some technologies go away and are replaced by other, faster and lower cost technologies!

At some point, it will be harder and harder to make chips faster, instead we need to also make the rest of the system faster as well - so even if the CPU isn't 20% faster - the overall system will be. We are starting to see some of that with PCI 5-6 and the move from SATA to PCIe based NVMe drives.

Thanks for sharing. However, do you think that the tech has gone so advanced now but in real life and among consumers, the power of an iPhone X/XS should be enough for what they regularly need now? We have not figured out yet if there is a real economic benefits coming out of these wonderful innovations, as far as the regular consumers are concerned. IPad pros continue to be too advanced for how it is being used now.

Like the other Steve said, it's about the software efficiency.... not the hardware. That was said so many decades.

Love your video. Want to correct your host that 2 nanometer is not made in China but in Taiwan. The Taiwanese and US government will not allow TSMC to build their 2 nanometer in China.

Ahh I miss your videos, I hope the new focus is going well :-)

1:15 this increase in R&D spend is just 6%-10% inflation over 12 years (CPI is mostly fake bc they change the products they calculate the inflation on each year, so when you look at the inflation and calculate it using the old method, the last 10 years have been about 10% inflation per year)

8:12 Junket?

What will happen then? If Moores law stops or at least is slowing down and at the same time the compute tasks keep on increasing fast, the energy needed will increase really fast, that might be quite problematic in the coming years

Keep the great content coming Rene! So fantastic to hear from Ben again. His TechPinions site and podcast were stellar!

Is the economic issue here really about the production cost or is it more about how much profit shareholders want out of those companies at the eventual cost of refraining innovation? Companies understood that as soon you say “economic” or “economy” a large frange of the population will understand (wrongly) and not question that much the why they’re doing or not what they’re doing: take us all for fool!

I really liked what they talked about, but I have a question: why has nobody talked about the fact that Apple has been stagnant for 2 years in its high-performance core designs? In 2020, when Apple released the A14 Bioinic chip, I presented it with the Firestorm cores, which reached a frequency of up to 3.0 GHz. The following year, I presented the Apple A15 Bionic with the Avalanche cores, which reached a frequency of up to 3.24GHz, and this year. Apple presented the a16 bionic chip with the Everest cores with a frequency of up to 3.46 GHz after 2 years and if you take the difference in frequency that Apple increased between its processors to see the increase in ipc that they would have achieved in 2 years, you get an improvement very small which is probably achieved by the increase in the L2 cache that apple has doubled in the high performance cores from a14 to a16 while in those same 2 years arm has achieved an ipc increase of more than 20% of a core cortex x1 to a cortex x3 nucleus

Thanks!

I think they should focus next on bringing down the prices of smartphones. It's becoming more and more ludicrous

Him saying building a 2nm fab in china is easy is a sign of not a lot of clue of the current situation it seems or did he include the RC meaning Taiwan the country.

Note AMD and Intel are laggards in performance per watt vs Apple. How can a specialist get this so wrong? Unfortunately it's the software that's lagging and is the driver of demand.

Rene, why the lack of an introduction? I found it hard to get into this conversation because it feels like I missed the beginning of it.

I see you with that WAN show hoodie, respect!

The funny thing is huawei is leading the way on the “stacking” technology…

A real question is how much speed do you require to do what you want to do. Most iPhones rarely if ever run into their maximum performance even playing games.

Rene, I am a big fan of your videos. However, this time I must admit, I had a hard time absorbing it, and it wasn’t because the topic was uninteresting, or that it was too confusing, or that you or you’re interviewee was not compelling. I found myself distracted throughout the video wondering “who the heck is this guy that Rene is interviewing?” I was so distracted, and I don’t think I got as much out of it as I normally would because my brain was thinking “did I miss the intro?” “ is this guy so big that I should just know who he is?”FYI, I am blind and listen to your videos. was there a graphic that told us who this guy is? Perhaps a short verbal intro would have been appreciated BTW, scanning through some other comments, I think the sighted people in your audience had some of the same questions that I did.

A. Your talking about proprietary technology amd has been developing and designing its chips around for years. Even if other companies adopt that approach it will still take a gen or 2 to catch up. B. It doesnt really make significant difference in processing npu/gpu both are matrix multipliers. Only one tech greatly benefits out of it which is memory - doesnt scale well C. Each company including Intel has been foreseeing this for a while. Intel for instance just released its roadmap for maintaining increasement throw architecture and transistor tech. They also have their tech for something they call tiles..

or you can sell same hardware with different name and slowing it down over time via close source software

Why don't we just refine our current coding structure to make applications and operating systems leaner and super efficient to "fly" even with a 10nm chip? I mean, I don't see the need for smaller and smaller nm numbers if we can still get the same performance for older chips with some ingenious coding tweaks. I remember the days Windows XP could fit in a 700mb CD and now, it's like a 4GB mamoth with endless services that just dont really benefit us that much. Back in the early 2010s when everyday smartphone applications never crossed 20mbs and now....we talking of GBs! I think the conversation should work in three directions. 1) Coding techniques 2) Efficiency per nm and 3) Battery technology.

The Limit for processors was hit years ago when they couldn't brake the 4GHz limit for a single core X86 processor, since then makers have had to move sideways in dev by adding more cores but lets the issue is still present. They have yet to make a +4GHz processor core. Edit: wow I feel old, was it really 2008?

I will be frank, I think we are about to hit a wall in the consumer demand for performance. We are about to have processors that perform like a 5950x on laptops, 3080/90 graphics performance on screens so tiny that you can barely distinguish 4k from 1440p. And I think we already reached that point with cellphones. I mean, how many people now buy cellphones because their previous one was too weak? On desktop, the best sellers right now as far as I am aware are ryzen zen 3, with the 5800x3d being particularly praised despite both intel and AMD just bringing their new cpus to the market. So, it makes me think, do your average consumer will really care for anything bellow 2nm? I mean, is improved battery life enough for a selling point? Perhaps I am wrong and there will always be a demand for performance by the general public but when I look to my phone... I only wish for something that would run just a little bit cooler. That’s not something that will make me spend hundreds of dollars on a new device.

I think the largest issue is heat dissipation and cooling that would make computers go to the next level

I remember in the 90s processor speeds doubled every 18 months. Once they hit 1 GHz the gains seems to have slowed. You get to a point where it is harder and harder to make more gains. Once they hit a wall on processor speed, they started adding more cores.

We need dilitheum crystals captain

The atomic radius of silicon is about 0.11 manometers, so we are reaching the limit of this element. Something will have to replace it, eventually, or we'll have to replace the transistors with something else.

If the CPU advancement stales completely, it will be fine by me. Why do we need more cpu power actually? People walk around with powerful computers in their pocket and do what with it? lol Even in science & engineering the cpu power is the least concern...so indeed, even if the cpus stop evolving - nothing bad would happen.

nice

should a 'D' be present in the 'AND' here?

And the point of buying a faster computer than the one you bought last year is?

He forgot the inflation is his argument about R&D.

Where is Risc V when you need it the most

Wow, what a cool video

The big shift is the move to AI-designed systems. NVIDIA has more AI-designed components on new GPUs than human-designed components. Chiplets, SOCs, etc. will be chosen by optimizing algorithms, not design teams. Moreover, AI design will democratize computer development.

I enjoyed this but didn't appreciate the way the guest repeatedly conflated Taiwan and China. There are no leading edge chips being fabbed in China, only in Taiwan.

The software needs todo the change now. It's years now since there was anything major new made for better performance and less memory and cpu/gpu usage. By hardware we can only scale if we put tons of accelerators in which is still no fix for jajf and the others..

Richie Ajja , namaskar

Why does the guest say build a 2 nanometer fab in “China”?! Does he know something about China invading Taiwan [Republic of China (ROC)]? The higher density fabs are NOT built in mainland China and that is not by mistake. The United States has expended substantial effort to make it illegal to build the newest technology fabs on the mainland. This isn’t just a little detail as mainland China getting control of Taiwan would have huge impacts on chip production, much more than what interconnection technology will be used. Besides, the gating technology for higher density chips is controlled by ASML Holding N.V. (commonly shortened to ASML, originally standing for "Advanced Semiconductor Materials Lithography") is a Dutch multinational corporation founded in 1984. ASML specializes in the development and manufacturing of photolithography machines which are used to produce the highest density computer chips. The other part of the equation is financial resources and execution of the application of this resources (which Intel seems to have continuous problems with the execution part). The guest even gets corrected about Taiwan but still doubles down on saying China. Very strange.

Goodbye Moore's-law. 👋

I think the battery technology needs to improve. Apple hardware design is not Friendly to DIY user. Everyone can write code Back to the 80th. Now you need to pay 99 dollars to use xcode

The future Power requirements won’t be able to support future Intel chips which are power hungry and heat generators.

It’s like the three-body problem is happening.

At a consumer level, how much more powerful machines do we need? I bought an M1P MBP and it's going to last well over 5 years before I even need to think about MAYBE upgrading it. I get people like you Rene, who need to process massive files and workloads to run video channels, but the average consumer is reaching a point in tech where anything they buy is probably much more powerful than they need, smartphone, laptop, tablet, etc. Even for you, I can only see maybe a few years before it's all capable of handling your workload completely with no wait or as little as possible.

They have been saying this for 40 years.

that's not true, it will matter for watches and smart phones as well. Everyone wants a phone and watch that runs faster and uses less battery power.

I hope that MacBook Pro computers soon also come with possible to have eSIM card so people how travel much can use MacBook Pro without connecting it to iPhone then we are I places without WiFi!

power draw does not keep increasing in fact it decreases the smaller nanometer chips you get

Apple gives you a 10% incremental increase and increase the price by 25%. How much faster do you really need to type a letter, how many dedicated Video Editors are their? Engineering 3D models seem to work perfectly at present, how much faster does an Artist need to paint? Still see lots of people hang out in the Coffee Room, so what’s their hurry for faster computers. The present Human Brain can’t work as fast as the present fastest chips, so exactly what is all the hype. ITS ALL IN THE MIND, 95% have computers, iPads, iPhones that are faster than they can think.

Stop making it smaller for now. And instead increase efficiency of the materials used instead

is anyone looking at risc v? right now arm is about to rule the world..we had an x86 monoculture and are now faincgin an arm monoculture.

Second... Also, I think it's going to transition into offboarding with cloud computing. A small highly efficient chip that can communicate instructions at a high rate and low latency with processors on the cloud. With advancements in wireless connectivity, eventually, wireless links are going to be fast enough to handle it...

I really feel bad when you ignore AMD and Mediatek

Not the end, The begining for qualcolm The snapdragon 8 Gen 2 is already on par with the A16 in everything but single core CPU performance It actually beats it comfortably in GPU and Ram performance Is it the end for Apple silicon ? Apple can advertise its single core performance all it wants by humiliating Qualcolm, equating it with A13 The fact is, the A16 is hardly 20% faster than the A13 and less than that with the 8 gen 2 The difference between the A12 and the snapdragon 845-855 was far more than its now The 8 gen 2 is far faster than chips 3 years older and Qualcolm is showing no signs of stopping Maybe Apple is reaping the results of its chip team talent loss But Hope not

I think desktop should move to arm architecture or something similar. X86 is starting to stagnate in terms of performance efficiency/perf gains with new process nodes as well.

Apple is playing the hype and profit game.

Quantum computing 🤔🤔

RISCV

Apple is already doing this.

Furst

And then IBM creates the photon CPU - millions of times faster than the fastest imagined silicon, millions of times less power.

Solve global warming with a Pentium

Menudo cuñado diciendo tonterías, no sabe ni de qué habla el pobre... Como se aburren los cuñados con dinero jajajajaj

His sound is poor ?

Do you think Apple will ever put an M1 chip in an iPhone?

Obviously chip future is ARM based.