New References to AMD Processors Discovered in macOS 10.15.4 Beta

[koyoot]

Obviously yes.
But the question is what marketing effort they have to investigate into that to make people actually buying them?

I will love to see a Mac with a much power efficient CPU. That means more performance in same form factor. But not able to run windows is already a problem for their retail team.

Everybody who talks about Mac on ARM hasn't got a clue what it actually means.

"More performance in the same form factor". Just this sentence means you have zero idea about the differences between ARM and x86.

 

[MikeZTM]

Apple ARM based CPUs are not faster than Intel CPUs because they have better cores. They don't. With each cycle, any 4 GHz x86 CPU will do 4-5 times more work than any ARM based CPU. There is a reason why ARM is RISC(REDUCED Instruction Set...) Architecture.

The reason why iPad processes video faster than Intel CPUs is because of integrated Image processor, not the cores, themselves. The cores are, and will always be slower, than any x86.

Guys, if you are not enough educated on the topic of CPU architectures, don't hype products up, because if Apple will switch to ARM on anything other than basic Services-access computers, competitors to Chromebooks, to put it plainly - you will make other customers hurt in their workflow.

ARM is not ready, and will never be ready to replace x86 for anything High-Performance. There is a very good reason why there does not exist a single SuperComputer that is based on ARM, all of them are based on x86. Apple is not going to change this.

Incorrect. And also you are not familiar with the term CISC and RISC. You need to be educated on topic of CPU architectures.

CISC and RISC are not performance related. They are philosophy related.
A x86 command still takes multi cycle to finish just like ARM command. All modern x86 CPU are pipelined so one clock tick only push the command to the next node in the pipeline.

CPU performance can be benchmarked using industry standard benchmarks like SPEC. These benchmarks are not like Geekbench which runs tiny code that can fit in the CPU cache. These are real world heave workload.
[automerge]1581112489[/automerge]

Everybody who talks about Mac on ARM hasn't got a clue what it actually means.

"More performance in the same form factor". Just this sentence means you have zero idea about the differences between ARM and x86.

For example:
Amazon AWS Graviton2 M6g instance.
32Core per CPU, 105W TDP.
Single core performance of that is faster than all x86 server CPUs(EPYC or Xeon) that has similar core count.

This is real world server/database workload. Not a remote connect client PC. The benchmark is done using nginx and PG SQL.

These new servers are amazing and also cheaper to rent. There's no reason apple can not build a laptop that use their own chip and not faster than Intel in the same power envelope.

 

[koyoot]

Incorrect. And also you are not familiar with the term CISC and RISC. You need to be educated on topic of CPU architectures.

CISC and RISC are not performance related. They are philosophy related.
A x86 command still takes multi cycle to finish just like ARM command. All modern x86 CPU are pipelined so one clock tick only push the command to the next node in the pipeline.

CPU performance can be benchmarked using industry standard benchmarks like SPEC. These benchmarks are not like Geekbench which runs tiny code that can fit in the CPU cache. These are real world heave workload.

I won't even budge into the first part, So I will just write this: Every industry validates ARM based on SPEC benchmarks.

I guess that is why every single SuperComputer is based on ARM architecture, and not on x86.

Oh wait, that is not right, is it?

 

[MikeZTM]

I won't even budge into the first part, So I will just write this: Every industry validates ARM based on SPEC benchmarks.

I guess that is why every single SuperComputer is based on ARM architecture, and not on x86.

Oh wait, that is not right, is it?

If you ever know something about CISC and RISC war back in 80s, you will never link them to performance.
CISC reduce command memory usage a lot compare to RISC and when you only have 640k ram that is a huge advantage. RISC was actually faster back then.

Today nobody care about command memory/binary size. We got data much larger than that.

BTW the problem of x86 is x86--the compatibility is killing the power efficiency of every x86 chip. Even if Intel build a CISC chip today without x86 legacy support if will be much faster. Rumor about Intel's roadmap even shows next micro architecture will drop full x86 compatibility(mostly dropping x87 fp ops) to gain more performance.

AWS Graviton2 already shows high power budget ARM chip can run faster than x64.

And for you: Intel officially submit SPEC score to SPEC database.
Sure they are not important right?

 

[Zdigital2015]

Yes, they are.

Zen 3 will 15-20% faster, per core, than Zen 2, again because of new architecture.

AMD is on a yearly schedule of innovation. And they will get double digit increases with each generation.

And you should get used to it already. AMD is not going to stop innovating.

Unless you are actually from the future, you cannot guarantee anything. You speaking these things as fact that don’t exist are to prop up your blind devotion. Your proselytizing and prognosticating and zealotry has no internal counterbalance.

EVERY company has highs and lows and AMD is no different in that regard. They have hit a high and I hope they can ride it long enough to put a serious dent in Intel’s warmed over hash CPU money machine. However, they WILL hit a wall at some point, every company does, AMD is not immune.

 

[koyoot]

AWS Graviton2 already shows high power budget ARM chip can run faster than x64.

No.

It can't. At least not in general purpose computing. In specialized, selected cirumstances - it can, by a smidgen. But not overall. ARM is not designed for high-performance computing. Period. Never will be.

 

[xnu]

That would be PCIe 4, not Thunderbolt 4, though, right?

Indeed, I was insane, sorry. Still would be great to have a choice for AMD .

 

[MikeZTM]

No.

It can't. At least not in general purpose computing. In specialized, selected cirumstances - it can, by a smidgen. But not overall. ARM is not designed for high-performance computing. Period. Never will be.

It is. I'm not arguing with you. This is a pure objective fact.

 

[koyoot]

Unless you are actually from the future, you cannot guarantee anything. You speaking these things as fact that don’t exist are to prop up your blind devotion. Your proselytizing and prognosticating and zealotry has no internal counterbalance.

EVERY company has highs and lows and AMD is no different in that regard. They have hit a high and I hope they can ride it long enough to put a serious dent in Intel’s warmed over hash CPU money machine. However, they WILL hit a wall at some point, every company does, AMD is not immune.

They had that low, for the past 10 years. Currently they are on a roll of technology.

It is. I'm not arguing with you. This is a pure objective fact.

Picture, or didn't happen .

And don't wash me with Amazon's marketing. There is a reason why those CPUs are used for basic stuff, and nothing meaningful. I have read about Graviton 2 and I am not eating the hype. You should also not eat it.

 

[mdriftmeyer]

No.

It can't. At least not in general purpose computing. In specialized, selected cirumstances - it can, by a smidgen. But not overall. ARM is not designed for high-performance computing. Period. Never will be.

More to the point ARM is an embedded architecture designed for a sandboxed OS that has dozens of processes running at a time, not several hundred to several thousand or more.

 

[PickUrPoison]

Does ECC memory really that matter? I mean for a desktop user?

Not for a typical desktop. But it’s a feature you want in a 30-50k workstation, especially with large memory configs.

The iMac Pro is debatable (to me), but it does currently have ECC.

 

[theluggage]

I guess that is why every single SuperComputer is based on ARM architecture, and not on x86.

Not "every single" one, no - but:

World's largest ARM supercomputer is headed to a nuclear security lab

Most supercomputers are focused on pure processing speed. Take the DOE's new Summit system, which is now the world's most powerful supercomputer, with 9,000 22-core IBM Power9 processors and over 27,000 NVIDIA Tesla V100 GPUs. But processing performance isn't everything. Last year, Hewlett...

www.engadget.com

Note: its not all about the raw speed of individual cores, its about how many cores you can cram on a chip and how much room (and power) is left for specialist accelerators, vector processors etc.:

(from link above)
it's considerably more power efficient and denser (meaning it can fit more hardware) than a comparable x86 system. Notably, that ARM chipset also offers 33 percent faster memory speeds than many x86 CPUs.

BTW the problem of x86 is x86--the compatibility is killing the power efficiency of every x86 chip. Even if Intel build a CISC chip today without x86 legacy support if will be much faster.

Thus. It's not a case of 'is RISC fundamentally faster than CISC' anymore - its about ARM's ability to cram more cores on the chip for the same power and the pick'n'mix licensing scheme that lets companies like Apple roll their own systems-on-a-chip.

 

[mdriftmeyer]

Thunderbolt 4?

Clearly he/she meant PCIE-4.0.

 

[Bandaman]

Obviously yes.
But the question is what marketing effort they have to investigate into that to make people actually buying them?

I will love to see a Mac with a much power efficient CPU. That means more performance in same form factor. But not able to run windows is already a problem for their retail team.

No.

 

[Nozuka]

Would love to see these in the Macbooks.

 

[koyoot]

More to the point ARM is an embedded architecture designed for a sandboxed OS that has dozens of processes running at a time, not several hundred to several thousand or more.

I would not go that far .

Yes, the general consensus of ARM community is to not have SMT/HT, and simply run the same amount of threads as there is cores, but... saying that ARM is not capable of Multithreading, because that is what it effectively means, is a stretch .

Not "every single" one, no - but:

Who, the **** cares about Power Efficiency in High Performance Computing sector?

What moron cares about it? Guys, Apple has just released a desktop computer that has 1.4 kW Power Supply!

The last thing High-Performance Computing cares about is power efficiency. This debate is beyond ridiculous.

 

[mdriftmeyer]

Intel’s fab issues aside...the $64,000 question is WHY would Apple maintain both in a shrinking PC industry? Why would they move from Intel to AMD given that the Mac sits at 8% of it sales? Because Intel is not doing great, but they aren’t so bad that Apple couldn’t ride it out a while longer.

Apple’s upgrade path for Mac Pro customers is to buy a new Mac Pro, they aren’t going to take your trade-in if they introduce an AMD-based Mac Pro.

IF Apple is willing to maintain both AMD AND Intel product lines in tandem, it means that a quantum shift in Apple’s thinking about the PC industry has occurred. People need to adjust their expectations accordingly, myself included (IF it happens).

The market is pivoting and starting to accelerate once more. The reason for this ``shrinking'' has mainly been due to longer shelf life of average systems purchased before refreshing to get new hardware.

As Operating Systems embed more Machine Learning and AI frameworks the heavy the demands on basic hardware requirements. And with that systems will be hitting a new stride as the large swath of 5 year systems will be recycled for modern systems to have current versions of macOS, Windows 10.x, Linux, etc.

When Smart Homes and the IoT starts kicking in systems will choke a lot more with all this demands on system resources.

Right now I've 387 Processes and 1,624 threads on my Macbook Pro and it's currently just keeping one instance of Logic Pro X running with iMessages, Safari, a few Terminal instances and the desktop running.

Five years from now the average system with have two or three times those numbers processing intercommunication demands.

 

[PickUrPoison]

Factuallyincorrect. Since Zen+ the ECC RAM limit per Zen+ processor is 2 TB. The 256 GB limit is an OEM motherboard manufacturer decision.

Apple has custom designed AMD motherboards in-house for years.
When they flip the switch it will be ready.

Nothing regarding AMD X86_64 architecture will be foreign and/or is foreign to Apple.

The amount of FUD about what it would take is sad.
AMD’s latest APUs going on sale early March already best Intel’s top of the line mobile iGPUs and was demonstrated last month.

Factually incorrect. Do your research, and stop spreading FUD.

AMD has segmented the desktop vs. workstation market by supporting ECC RDIMMs and LRDIMMs only on EPYC.

It is up to motherboard manufacturers whether or not they want to support ECC in the first place. But they can’t make AMD’s memory controller work with anything but UDIMM ECC.

 

[koyoot]

Factually incorrect. Do your research, and stop spreading FUD.

AMD has segmented the desktop vs. workstation market by supporting ECC RDIMMs and LRDIMMs only on EPYC.

It is up to motherboard manufacturers whether or not they want to support ECC in the first place. But they can’t make AMD’s memory controller work with anything but UDIMM ECC.

The memory controller on the CPU is the same it is for EPYC. So its only CPUs Microcode that creates this difference.

Who cares about this anyway? EPYC CPUs would end up in Mac Pro - anyway, and iMac Pro would get Threadrippers.

 

[konqerror]

I guess that is why every single SuperComputer is based on ARM architecture, and not on x86.

World's fastest supercomputers have not been x86 for a while. Look up Blue Gene. IBM put together a bunch of crappy PowerPC 440s to get scalability.

World's fastest supercomputers right now are IBM POWER, followed by a proprietary Chinese architecture.

On top of that, GPUs are definitely not x86 based; Intel tried that and failed miserably.

As was pointed out, there's an x86 tax. Intel spends more transistors paying the x86 tax than they do in the actual execution units. The frontend is about twice as big as the EUs.

This is the whole point of SIMD like SSE and AVX, to pay the x86 tax once. And why POWER and ARM don't need hugely wide 512 bit vector units.

 

[barnyard]

I still expect them to come out with their gaming computer with Ryzen and Nvidia this year.

 

[koyoot]

World's fastest supercomputers have not been x86 for a while. Look up Blue Gene. IBM put together a bunch of crappy PowerPC 440s to get scalability.

World's fastest supercomputers right now are IBM POWER, followed by a proprietary Chinese architecture.

On top of that, GPUs are definitely not x86 based; Intel tried that and failed miserably.

As was pointed out, there's an x86 tax. Intel spends more transistors paying the x86 tax than they do in the actual execution units. The frontend is about twice as big as the EUs.

This is the whole point of SIMD like SSE and AVX, to pay the x86 tax once. And why POWER and ARM don't need hugely wide 512 bit vector units.

World’s fastest supercomputer will be built by AMD and Cray for US government

Capable of 1.5 quintillion calculations per second

www.theverge.com

Is this ARM, POWER, or something else based?

Or they are 100% based x86 AMD CPU+ AMD GPUs?

I still expect them to come out with their gaming computer with Ryzen and Nvidia this year.

Considering that Nvidia stopped development of MacOS Drivers - you better move on, and get to reality that No more Nvidia in Apple computers for the foreseeable future.

 

[MikeZTM]

They had that low, for the past 10 years. Currently they are on a roll of technology.

Picture, or didn't happen .

And don't wash me with Amazon's marketing. There is a reason why those CPUs are used for basic stuff, and nothing meaningful. I have read about Graviton 2 and I am not eating the hype. You should also not eat it.

Coming Soon – Graviton2-Powered General Purpose, Compute-Optimized, & Memory-Optimized EC2 Instances | Amazon Web Services

We launched the first generation (A1) of Arm-based, Graviton-powered EC2 instances at re:Invent 2018. Since that launch, thousands of our customers have used them to run many different types of scale-out workloads including containerized microservices, web servers, and data/log processing. The...

aws.amazon.com

They can not lie about performance.

x264 and simulation is the key here and at least 20% faster than intel Xeon is impressive.

they are not deploying this for lite workloads.
This is for next gen general propose ec2 instance.

M5 C5 and R5 are all mainstream Xeon servers and replaced by M6g C6g R6g arm servers.

 

[dgdosen]

Now we're heading into a world where if Apple doesn't ship 8-core 4800U laptops, we'll be disappointed.

I wonder how soon they'll make it an XPS 13 Developer Edition?

 

[konqerror]

World’s fastest supercomputer will be built by AMD and Cray for US government

Capable of 1.5 quintillion calculations per second

www.theverge.com

Is this ARM, POWER, or something else based?

Or they are 100% based x86 AMD CPU+ AMD GPUs?

Considering that Nvidia stopped development of MacOS Drivers - you better move on, and get to reality that No more Nvidia in Apple computers for the foreseeable future.

AMD gave them the stuff for a very low cost so they could gain a foothold in HPC. Both the CPUs and GPUs. Otherwise, with the prevalence of CUDA, nobody would select that. Your blogger didn't understand the concept of "lowest bidder".

TOP500 List - November 2019 | TOP500 Supercomputer Sites

www.top500.org

And you helped my point. AMD GPUs do not run on x86.