Everyone has been using SSE for FP operations for a long while. In modern software, x87 is basically non existent, but it continues to be supported - and it’s a mess. Correct implementation requires x87 to use higher internal register precision, which has real hardware design cost. All fir supporting an essentially deprecated functionality that should have been dropped with x86-64... and then we wonder why x86 is having trouble.
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.
Will Intel and the others catch up to match M1
- Thread starter Bearbrickster
- Start date
- Sort by reaction score
Everyone has been using SSE for FP operations for a long while. In modern software, x87 is basically non existent, but it continues to be supported - and it’s a mess. Correct implementation requires x87 to use higher internal register precision, which has real hardware design cost. All fir supporting an essentially deprecated functionality that should have been dropped with x86-64... and then we wonder why x86 is having trouble.
It would be interesting to see how a fully maxed Cortex-X based processor on 5nm would compare.
I don’t want to speak for Vlad, but when I think about an ASIC, I can’t help but think about crypto-currency mining. Now, that may not be a market that Apple is eager to explore, but AMD just announced their fastest GPU, with no video outputs, designed for miners.
Great thread this, really informative, thanks to all the posters.
I'd suggest Intel are more concerned about AMD taking an ever increasing share of the enterprise market over the coming years instead of consumer grade products. EPYC is a real threat to them. Historically as other posters have stated we'd normally expect Intel to take a step back and redefine their approach (Core Duo, Intel Core). I still think they peaked with Sandy Bridge with the current architectural family. Intel have a history of sitting on their hands and in many ways holding back progress when they lack competition, hopefully both Apple and AMD will spur a response?
What I'm fascinated to see is how Apple scale this? Will they continue with an SoC approach or at some point break out to a more traditional component based architecture? Can they continue the former when they attempt to scale out to a high end MBP 16" or Mac Pro?
I'd suggest Intel are more concerned about AMD taking an ever increasing share of the enterprise market over the coming years instead of consumer grade products. EPYC is a real threat to them. Historically as other posters have stated we'd normally expect Intel to take a step back and redefine their approach (Core Duo, Intel Core). I still think they peaked with Sandy Bridge with the current architectural family. Intel have a history of sitting on their hands and in many ways holding back progress when they lack competition, hopefully both Apple and AMD will spur a response?
What I'm fascinated to see is how Apple scale this? Will they continue with an SoC approach or at some point break out to a more traditional component based architecture? Can they continue the former when they attempt to scale out to a high end MBP 16" or Mac Pro?
I saw a Postgres performance chart today on MacBook Pro models for the past ten years and the M1 smoked the older models. This M1 and macOS might make the foundation for decent servers.
A lot of the posts here about Intel still miss two very big elephants in the room, one they've always had a process node lead, and two they didn't have much squabbling in their ranks.
Unless Apple starts making Apple Silicon chips for the server market and gaming market then Intel and AMD have absolutely nothing to worry about. Outside of the US with regards to educational establishments, Intel and AMD are king because their machines are a lot cheaper than Mac machines. As for businesses, efficiency and productivity tests will be run by many IT departments to see if the new M1 machines can run certain programs faster than a windows machine can because the faster a computer can do a job, the more money a business can make.
The ONLY people who are going to buy the new M1 machines right now are Apple fan's just because it's Apple. Businesses and educational establishments are not going to spend thousands upon thousands of (type in currency here) just because Apple has released M1 machines. Both sectors are going to make sure they get value for money and they will do their own relevant tests to see if M1 machines give them value for money.
This forum is a good indicator of who are the ones purchasing the new M1's and it just so happens to be owners of Intel Macs who are upgrading to the M1's, no one else, which was expected.
The ONLY people who are going to buy the new M1 machines right now are Apple fan's just because it's Apple. Businesses and educational establishments are not going to spend thousands upon thousands of (type in currency here) just because Apple has released M1 machines. Both sectors are going to make sure they get value for money and they will do their own relevant tests to see if M1 machines give them value for money.
This forum is a good indicator of who are the ones purchasing the new M1's and it just so happens to be owners of Intel Macs who are upgrading to the M1's, no one else, which was expected.
My employer is making more and more of our daily tools available through mobile devices, such as the iPad. There is no technical reason why these also wouldn’t work on the M1 Mac. Inertia is probably the biggest factor keeping us a Windows shop.
Some small research labs put together their own servers from PCs. Then mix them with big iron that's in the IT budget.
Not from Apple, but...
ARM Processor - AWS Graviton Processor - AWS
AWS Graviton processors deliver the best price performance for your cloud workloads, optimized for a range of general-purpose, compute, memory, and storage-intensive workloads.
Ampere Altra is the first 80-core ARM-based server processor
Ampere has unveiled the industry's first 80-core ARM-based 64-bit server processor today in a bid to outdo Intel and AMD in datacenter chips.
...etc. If you're running a huge data centre, powering all those hungry x86 CPUs and pumping out the waste heat is a big deal, so if ARM even comes close to Intel performance while using less energy, it is attractive.
Apple's M1 isn't a server chip and a lot of its advantages over "regular" ARM are aimed at things like media playback and editing... although the Neural Engine might be interesting server-side... However, a high-profile success with the M1 (and Apple is very high-profile) won't harm ARM's "mindshare" at all and may change some of the "nobody ever lost their job for buying Intel" attitude.
Meanwhile, Linux is nibbling away at the Windows server market - a lot of modern web-service technology is built on open-source software rather than Windows/IIS/SQL Server that runs happily on Linux (although hell froze over a few years ago when Microsoft released SQL Server for Linux - they've also added a Linux subsystem to Windows and are pushing cross-platform development tools)... and while most Linux servers are currently x86, Linux has run on ARM since forever ago and ARM64 Linux is already well supported by most of the big open source projects. The whole Linux/Unix ethos is very much about source-level compatibility rather than Windows' obsession with running 25 year old binaries - so once Windows is out of the picture, shifting processor architecture (whether it is ARM, RISC-V or something new) is much less of an obstacle.
So, yeah, "Wintel" is too big to disappear anytime soon, but it's being eaten by ARM and others at both ends and has probably passed its glory days. (Hooray!)
I think the direction that AMD and Intel, as well as makers of Windows laptops, go in is going to depend on what happens next.
If Apple gains a significant market share over the next few years, I think they are going to make very significant changes and we may see a greater equivalence of 'matched' performance. If Apple's Mac market share remains mostly the same, then I don't think we are going to see such major changes across the industry as it effectively means 'good enough' and tradition will continue to be sufficient to sell.
If Apple gains a significant market share over the next few years, I think they are going to make very significant changes and we may see a greater equivalence of 'matched' performance. If Apple's Mac market share remains mostly the same, then I don't think we are going to see such major changes across the industry as it effectively means 'good enough' and tradition will continue to be sufficient to sell.
EPYC isn’t an actual threat right now in Enterprise. Not yet at least. I haven’t seen anyone ask any of my customers ask for it. It wouldn’t surprise me if they are getting a bigger boost in AWS and Azure. In Enterprise it’s hard because you can’t cleanly vMotion from Intel to AMD without a noticeable performance hit. Which means you’d have to create a special cluster just for AMD.
I’m more than happy to be wrong, but in all of my time as a Sales Engineer I don’t think AMD has brought been brought up ever.
ARM will ONLY take away server business from Intel and AMD if and I mean IF the ARM server can beat the Intel and AMD servers in efficiency, reliability, performance and running cost otherwise NO server business is going to transition to ARM because it would be too expensive to do so.Not from Apple, but...
ARM Processor - AWS Graviton Processor - AWS
AWS Graviton processors deliver the best price performance for your cloud workloads, optimized for a range of general-purpose, compute, memory, and storage-intensive workloads.aws.amazon.com
Ampere Altra is the first 80-core ARM-based server processor
Ampere has unveiled the industry's first 80-core ARM-based 64-bit server processor today in a bid to outdo Intel and AMD in datacenter chips.venturebeat.com
...etc. If you're running a huge data centre, powering all those hungry x86 CPUs and pumping out the waste heat is a big deal, so if ARM even comes close to Intel performance while using less energy, it is attractive.
Apple's M1 isn't a server chip and a lot of its advantages over "regular" ARM are aimed at things like media playback and editing... although the Neural Engine might be interesting server-side... However, a high-profile success with the M1 (and Apple is very high-profile) won't harm ARM's "mindshare" at all and may change some of the "nobody ever lost their job for buying Intel" attitude.
Meanwhile, Linux is nibbling away at the Windows server market - a lot of modern web-service technology is built on open-source software rather than Windows/IIS/SQL Server that runs happily on Linux (although hell froze over a few years ago when Microsoft released SQL Server for Linux - they've also added a Linux subsystem to Windows and are pushing cross-platform development tools)... and while most Linux servers are currently x86, Linux has run on ARM since forever ago and ARM64 Linux is already well supported by most of the big open source projects. The whole Linux/Unix ethos is very much about source-level compatibility rather than Windows' obsession with running 25 year old binaries - so once Windows is out of the picture, shifting processor architecture (whether it is ARM, RISC-V or something new) is much less of an obstacle.
So, yeah, "Wintel" is too big to disappear anytime soon, but it's being eaten by ARM and others at both ends and has probably passed its glory days. (Hooray!)
You also have to factor in serve maintenance, upgrades both hardware and software, repair and trouble shooting. Nearly every server business will have people working there who are experts on Windows server software and thus can get a faulty server up and running within a few minutes. A downed server can cost a company thousands by the minute, sometimes millions and as a result, getting the server up ASAP is of paramount importance. A server business will weight up this factor in wanting to transition to either Linux or ARM because it would mean having to retrain technicians in a software they are unfamiliar with which could means delays in getting a downed server up and running. ARM servers could provide financial benefits in SOME areas but having the fastest and most powerful server does not mean it's going to save the company money in the long term.
People have a nasty habit of thinking like that, it's the fastest, it the most powerful therefore it's going to be the best. It does not work like that in the business world because there many other factors to think of.
Therefore as I said, Intel have nothing to be afraid of with these M1 machines, neither servers or gaming machines will be affected, which is a big core of their market.
Apple's M1 is a complete package, not a CPU. Intel/AMD will not be able to match it at all simply because they have to tailor for the general public while Apple can customize its SoC solely for its software. Apple can customize every element of its SoC; CPU, GPU, Secure Enclave, Neural Engine (ML accelerators), ISP, media en/decoders, and so many more to work exclusively with its software stack and optimize the crap out of it.
We haven't seen what Apple can do with this just yet (remember, this is just wave one with just reusing what they have now, they'll be redesigning the entire ecosystem over the next several years to solely focus on their silicon).
Intel and others can still catch up and beat Apple at certain tasks like general tasks (AMD Zen3 is on par with M1 for multi-core IMO) but they will not be able to beat Apple for a complete package or its ecosystem. After all, for a lot of people, they're buying Apple for the complete experience such integration with Apple Watch, AirPods, Apple TV, iOS devices, etc.
This is why Intel is now trying to do more with its Evo ecosystem and making a custom 15" NUC laptop kit for others to use (https://www.theverge.com/2020/11/19/21573577/intel-nuc-m15-laptop-news-specs-features-price), they want to try to provide a complete ecosystem too. It's also why they also are investing so heavily in Xe Graphics and entering the dGPU market.
We haven't seen what Apple can do with this just yet (remember, this is just wave one with just reusing what they have now, they'll be redesigning the entire ecosystem over the next several years to solely focus on their silicon).
Intel and others can still catch up and beat Apple at certain tasks like general tasks (AMD Zen3 is on par with M1 for multi-core IMO) but they will not be able to beat Apple for a complete package or its ecosystem. After all, for a lot of people, they're buying Apple for the complete experience such integration with Apple Watch, AirPods, Apple TV, iOS devices, etc.
This is why Intel is now trying to do more with its Evo ecosystem and making a custom 15" NUC laptop kit for others to use (https://www.theverge.com/2020/11/19/21573577/intel-nuc-m15-laptop-news-specs-features-price), they want to try to provide a complete ecosystem too. It's also why they also are investing so heavily in Xe Graphics and entering the dGPU market.
Scalability is the key. That is a real test for Apple M1 and its successors. And for its competitors.
Last edited:
Zen is not really on par, far from it. I guess the issue is, that you compare M1 4+4-core implementation vs. Zen3 8 core. You would need to compare it vs the upcoming 4+8 configuration of AS.
Comparing the low-end M1 to a high-end Zen3 isn't really fare. Let's compare the low-end M1 to a low-end Zen3 chip.
The other thing is RAM. What do the numbers on Zen3 and for applications in general look like with 8 GB of RAM?
Intel Macs still use them for "long double" types in C (80-bit), but only a very small number of applications use this type.