Do you understand the VLIW issue? The VLIW is additive. It may account for 1/2 of the M1 performance advantages over x86.
It's pretty simple if you're an EE or have a CS degree.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Why does Apple act like Intel is always holding them back?
- Thread starter dspdoc
- Start date
- Sort by reaction score
Do you understand the VLIW issue? The VLIW is additive. It may account for 1/2 of the M1 performance advantages over x86.
It's pretty simple if you're an EE or have a CS degree.
Time will tell but I have been around the block more than a few times and can usually see what's coming. Its not about any degree its about intuition and industry experience. A degree gets you no common sense whatsoever just education. Common sense got me to where I am today Senior Vice President of Operations in a world leading semiconductor company.
I worked in industry for ten years before getting my first degree. A couple of employers paid for my Bachelors. I was in design meetings and didn't understand things that were discussed in the meetings that they assumed that everyone else knew so I got a graduate degree and then I understood what they were talking about.
You don't have to get a degree - you can just buy textbooks and read them. There are lots of online textbooks that you can read for free.
I'm a bit surprised that you would prefer ignorance to learning something that should be fairly easy to learn.
There are a number of articles on the web about this that try to dumb it down but you can literally read two paragraphs and understand it with a bit of theory.
Well, then please, by all means, explain the VLIW issue so that everyone here understands one of the two reasons why M1 is so fast using very little power.
You just know it would take a mammoth response to do justification to this which is why you set the challenge. Let me condense it. Custom silicon comes with custom software, get it right and you have an stellar performing computer (Apple M1). Get it wrong in either the silicon OR the software and you are screwed. VLIW works in the same way, get the software to do the hard graft not the hardware. Lets see how successful it will be compared to Apple hardware and software. Big is not necessarily better, proved that back in the old days.
It's clear that you don't understand VLIW.
You would need to understand what a Decoder does to understand VLIW.
Lets convene a year from now and see how the computer world looks. One way or another one of us will be able to say 'I told you so'
No hard feelings take care and be safe.
I'll write up something on why VLIW is so important later this afternoon and why it will be impossible for Intel or AMD to do the same thing with x86.
I don’t understand this. How are Intel chips “all-encompassing” and Apple chips narrow-purpose? Intel has a series of different chips for different use cases - Tiger Lake for low-powered mobile, (soon) Rocket Lake for consumer/enthusiast desktop, Cascade Lake for workstation/server. So far, Apple has only released a chip that targets the ultra-low-power mobile. There is nothing in the design of their processors that would stop them from building a workstation-targeting behemoth - assuming they are interested in doing so.
What is so “custom” about Apple Silicon? Sure, it contained optimizations for some operations that Apple often uses in their code base. But those are still general-purpose operations. And Apple Silicon outperforms Intel and AMD in most general-purpose workloads. For example, AS has very good branch prediction, huge caches and four 128-bit FP units. You can’t brush those components away as “custom accelerators”. It’s just a more capable CPU.
Can you clarify what you mean with “VLIW”? The only term I know with this mnemonics is very long instruction word and I don’t see how that would be even remotely relevant to this discussion.
Can you elaborate on this point? I can take any established code, say a compiler, a scientific computation library, a browser, that uses only plain old C or C++, without calling into Apple frameworks, and it will run faster on M1 than on Intel. Is that custom software too?
Chip design is not rocket science, or actually it can be because it is a marriage or software and hardware. I have been doing this in the Bluetooth arena for 20 years and it is not easy. Designing a chip in 5nm and marrying it to custom software is hugely difficult. Not only separately in isolation but together as a true SOC. Do not underestimate what it took to get to where the M1 is today.
Sure, the company I worked for over the last 20 years had twice as many software engineers as hardware engineers and budgeted three times as much for software development compared to hardware development. I have no doubt about your low level software abilities in a particular focus but how about mass market multiple diverse applications. How many user cases could you support alone given multiple silicon revisions and ongoing software revisions, customer support and bug fixing?
I really don’t understand what you are trying to say. The M1 literally runs the same software as Intel. Maybe some low-level OS components are optimized specifically for M1, but most code out there was written long before Apple Silicon was a speculation on obscure tech forums.
The reason why M1 is fast is because it is designed for running code that exists in the real world. Large cache, best in class branch prediction, fast atomic operations, memory parallelism, wide execution backend, sophisticated prefetch - all those are thing that Apple implemented by carefully examining existing software and its needs. You can call this “custom silicon” but I’m not sure whether this nomination has any practical meaning then.
Agreed. We have 8 CPU cores, 8 GPU cores (in the higher version), and 16 ML cores. Intel/AMD currently do not have dedicated machine learning cores, which can definitely be used in a lot of software. It also contains the secure enclave, low powered video playback, image signal processor and more. Things that Apple and software can make use out of to improve performance. Some of these are something a generic processor cannot include.
I think its a wash to be honest. I have 20 years experience in the semiconductor industry watching how hard it is to marry software with the hardware. You need twice as many software engineers as hardware engineers to even get close to making a chip successful. I can't convince you, you would have had to live my life for the past 20 years!
I think that is what is being mentioned, and it does impact complexity of decoders. Specifically, AMD has commented on how they noticed there's an upper limit on the useful number of decoders on x86 due to the need to handle these kind of instructions, despite being rarely used. The A14/M1 architecture is already beyond that particular limit, and it does benefit from it.
This is probably the closest thing to "custom" in Apple's CPU design (EDIT: specifically CPU design, not SoC). But that's more to support modern languages like Swift (and Objective-C using ARC). But there's a lot of other multi-threaded scenarios that would benefit from this as well, so even that's not terribly custom, it's just examining existing usage of the CPU and adapting to it.
It’s possible for individuals to, sure. But Apple is a public company whose sole job it is to maximize profits for shareholders. Does Apple have some good values and policy? Sure, but it’s all in the interest of maximizing profits. Privacy is a good example (and a reason I choose Apple actually - and a reason lots of Apple users choose Apple too). This allows them to cater to a certain segment of the market. I would bet money that if no one cared about privacy and security Apple would abandon those principals because they are no longer profitable. These are companies trying to make money, no more, no less. Sometimes they make cool things while trying to make money and make good choices but these choices are made because of market pressures, not to do the right thing.