MP All Models Is Apple ****ing themselves over the ThreadRipper 3 yet?

[Pro7913]

Why?

Apple controls the entire chipset. They can do anything with it they want or optimize for anything they want.

AVX512 seems better than NEON, but AMD doesn’t have that either... And Apple could add their own instructions.

It really does come down to cost. Is Apple willing to pay for that development work?

Do you really expect that to happen right now? No. That ARM is limited only for laptop, not for desktop-grade performance. What about software? How many software are there ready to support ARM-based Mac?

Also, there arent any data comparing between ARM CPU and Intel/AMD CPU with the same software to see how ARM CPU performs.

You are living in the illusion are you not?

 

[Unregistered 4U]

Not true. I have a Hackintosh running the new Ryzens CPUs and I can tell you it outperforms everything else that's out.

I stand corrected, but others have mentioned here, there are nebulous legal concerns that, while it’s an option, it’s not a serious option for companies that are EXTREMELY sensitive to playing it ”by the book”.

I’m fairly certain that Apple isn’t losing any sleep over AMD’s new processors.

 

[goMac]

Also, there arent any data comparing between ARM CPU and Intel/AMD CPU with the same software to see how ARM CPU performs.

Sure there is.

https://browser.geekbench.com/v5/cpu/711233 (With the last generation A12)
https://browser.geekbench.com/v5/cpu/712287

The last generation A series, on an iPad with no active cooling, is faster than a Mac Pro 5,1 in single thread, and nipping at it in multicore with half the cores. At 2/3rds of the clock rate.

Is it comparing against an old Intel CPU? Sure. But it also beats Threadripper in single thread. Without a fan.

https://browser.geekbench.com/v5/cpu/712148

It still manages to come in at 1/3rd of Threadrippers multicore, without active cooling and half the cores, and in a much smaller design.

(Edit, that's last gen threadripper, pulling latest benchmarks again)

Edit 2: Even on latest gen Threadripper, it's pretty close in single core performance, although Threadripper takes the edge. Again though, that's on an iPad with no active cooling:
https://browser.geekbench.com/v5/cpu/700747

A scaled up A series would be very competitive with Threadripper. Question is if Apple would want to put in the resources to do it.

Apple does extremely good work. It's not worth comparing the A series with other ARM CPUs because the A series is not a generic ARM CPU. Clock for clock, it is even more efficient than Threadripper.

They are most definitely not ****ing themselves over the ThreadRipper 3.

 

[Pro7913]

Sure there is.

https://browser.geekbench.com/v5/cpu/711233 (With the last generation A12)
https://browser.geekbench.com/v5/cpu/712287

The last generation A series, on an iPad with no active cooling, is faster than a Mac Pro 5,1 in single thread, and nipping at it in multicore with half the cores. At 2/3rds of the clock rate.

Is it comparing against an old Intel CPU? Sure. But it also beats Threadripper in single thread. Without a fan.

https://browser.geekbench.com/v5/cpu/712148

It still manages to come in at 1/3rd of Threadrippers multicore, without active cooling and half the cores, and in a much smaller design.

A scaled up A series would be very competitive with Threadripper. Question is if Apple would want to put in the resources to do it.

Apple does extremely good work. It's not worth comparing the A series with other ARM CPUs because the A series is not a generic ARM CPU. Clock for clock, it is even more efficient than Threadripper.

lol, Geekbench does not represent the real CPU performance and I already expect you to bring Geekbench scores to show how ARM performs. Guess what? Wrong.

https://www.forbes.com/sites/patric...ting-the-industry-and-consumers/#7d5db6fa2dc6

ARM and x86 are not directly comparable and you failed to provide information. Geekbench is not a reliable source and it is not able to compare between ARM and x86.

I told you to bring sources by testing the same application or software like Adobe Photoshop.

Also, there arent any software to compared because only a few developers made software for both ARM and x86 like Photoshop. If you gonna compare, bring something realistic.

 

[goMac]

ARM and x86 are not directly comparable and you failed to provide information.

I did. You just don't like it.

Geekbench is not a reliable source and it is not able to compare between ARM and x86.

It compares just fine between ARM and x86. You just don't like the results, so you've decided the results must be wrong.

Also, there arent any software to compared because only a few developers made software for both ARM and x86 like Photoshop.

Apple ARM is different than other vendors ARM. You can't just benchmark "ARM."

That's why the blanket generalizations that ARM is not faster is irrelevant. Which ARM implementation are you talking about specifically? It's like calling AMD and Intel the same performance because they are both x86, when clearly they are not the same.

 

[Pro7913]

I did. You just don't like it.

And I told you. Geekbench is not reliable. It does not represent the real CPU performance and yet you are not listening.

It compares just fine between ARM and x86. You just don't like the results, so you've decided the results must be wrong.

I provided an article. Did you even read it?
[automerge]1575331875[/automerge]

I did. You just don't like it.



It compares just fine between ARM and x86. You just don't like the results, so you've decided the results must be wrong.



Apple ARM is different than other vendors ARM. You can't just benchmark "ARM."

That's why the blanket generalizations that ARM is not faster is irrelevant. Which ARM implementation are you talking about specifically? It's like calling AMD and Intel the same performance because they are both x86, when clearly they are not the same.

https://www.reddit.com/r/hardware/comments/6hnm4a

 

[goMac]

And I told you. Geekbench is not reliable. It does not represent the real CPU performance and yet you are not listening.

And yet it also doesn't represent nothing. It's certainly representing some measure of performance. It's not just spitting out random numbers.

I provided an article. Did you even read it?

What's the relevant part here?

 

[koyoot]

Do you really expect that to happen right now? No. That ARM is limited only for laptop, not for desktop-grade performance. What about software? How many software are there ready to support ARM-based Mac?

Also, there arent any data comparing between ARM CPU and Intel/AMD CPU with the same software to see how ARM CPU performs.

You are living in the illusion are you not?

There is data. You have to search for it.

Ask people from Data Centers, Servers, and HPC space, why haven't they switched to ARM, if it is so powerful, and power efficient.

 

[goMac]

https://www.reddit.com/r/hardware/comments/6hnm4a

Your own link you provided said these benchmarks are fairly accurate and represent the progress Apple has made.

 

[Pro7913]

And yet it also doesn't represent nothing. It's certainly representing some measure of performance. It's not just spitting out random numbers.



What's the relevant part here?

I added a Reddit post about this topic. It just proves that you didnt even search google about Geekbench.
[automerge]1575332108[/automerge]

Your own link you provided said these benchmarks are fairly accurate and represent the progress Apple has made.

I guess you didnt even read.

 

[goMac]

I added a Reddit post about this topic. It just proves that you didnt even search google about Geekbench.
[automerge]1575332108[/automerge]


I guess you didnt even read.

Uhhhhh ok.

Anyway what I was going to say before we went on this tangent, is the one place Intel is still king is vector instructions. They're beating both Apple and AMD on that. The recent Threadripper benchmarks still seem to show that's the one advantage Intel has.

If Intel continues to lead in vector performance, that could give them the edge in pro scenarios. Seems to come down to the instruction sets. AMD isn't supporting the latest AVX, and ARM NEON hasn't moved the ball forward yet. Doesn't stop Apple from getting a custom design from either AMD or in house.

 

[Pro7913]

Your own link you provided said these benchmarks are fairly accurate and represent the progress Apple has made.

The main reason why you failed to convince is that you provided only the Geekbench score which is also not reliable for comparing between ARM and x86 directly. This is why you need to bring several sources.

 

[goMac]

Ask people from Data Centers, Servers, and HPC space, why haven't they switched to ARM, if it is so powerful, and power efficient.

One issue is that there is no "big" version of Apple's ARM chipset. And Apple's ARM chipset is the one to beat right now. Even the data center vendors don't seem to be coming close.

That leaves us to extrapolate from iPad performance, which is still very impressive. But benchmarks on third party ARM CPUs is still different than benchmarks on Apple's ARM CPUs.

It's also complicated by Apple being able to tune the A series however they want. Bigger A series CPUs could be tuned different, change big/little configurations, etc etc.
[automerge]1575332573[/automerge]

The main reason why you failed to convince is that you provided only the Geekbench score which is also not reliable for comparing between ARM and x86 directly. This is why you need to bring several sources.

Again, it's still relevant. Even if it's only 90% accurate, it's still showing a trend.

There is nothing special about ARM vs x86 in Geekbench. There is nothing that makes it any better or worse a tool than if you were comparing AMD vs Intel.

Also it's not comparing "ARM" vs "x86." There is no generic "ARM" CPU. It's Apple's A series CPU. The question is could an A series compete with Intel or AMD.

 

[Pro7913]

Again, it's still relevant. Even if it's only 90% accurate, it's still showing a trend.

There is nothing special about ARM vs x86 in Geekbench. There is nothing that makes it any better or worse a tool than if you were comparing AMD vs Intel.

I wouldn't say it's relevant. Do you really expect A13 can run Adobe After effect or Premier Pro smoothly? Do you have a test result about using same software with both ARM and X86? We are talking about real-life results. This is why the benchmark especially Geekbench does not represent the real performance. You need to realize what you are talking about.

 

[goMac]

Do you really expect A13 can run Adobe After effect or Premier Pro smoothly?

Yes. And I don't actually need to speculate on that in since Premiere runs on the iPad (along with a lot of competing editing apps.)

This is why the benchmark especially Geekbench does not represent the real performance. You need to realize what you are talking about.

You keep saying this but you don't provide any justification.

I know Geekbench isn't 100% accurate. But it also doesn't mean nothing.

 

[koyoot]

One issue is that there is no "big" version of Apple's ARM chipset. And Apple's ARM chipset is the one to beat right now. Even the data center vendors don't seem to be coming close.

That leaves us to extrapolate from iPad performance, which is still very impressive. But benchmarks on third party ARM CPUs is still different than benchmarks on Apple's ARM CPUs.

It's also complicated by Apple being able to tune the A series however they want. Bigger A series CPUs could be tuned different, change big/little configurations, etc etc.

Which already should raise quite a lot of eyebrows.

Apple has extremely advanced design from the Abundance of FF units on the silicon. ALUS and AGUS.

The problem is this. To equal the length of x86_64 instruction, you need 4(!) instructions on ARM.

In other words. You may have higher IPC on Apple CPUs, compared to even Intel, or latest AMD's CPUs. But you still may do less, with each cycle than you do with x86_64.

And that is just instruction level problem. Show me an ARM CPU that can get 40 amps of power, or more. People forget that Apple's designs are running at low voltages, under 1.0v, and they are 5-7W designs at best with very low clocks relative.

From what I remeber, what David Kanter said based on discussions with Apple engineers - there is no way they can provide more than 40 AMPs to their system on chips.
It was 2-3 years ago, but I believe there was nothing that changed on this front. Its ARM architecture, after all.

If Apple want performance degradatgion by going with ARM - sure, that will be great thing to see.

But Im not sure that it is what consumers want to see.

 

[goMac]

Which already should raise quite a lot of eyebrows.

Apple has extremely advanced design from the Abundance of FF units on the silicon. ALUS and AGUS.

The problem is this. To equal the length of x86_64 instruction, you need 4(!) instructions on ARM.

In other words. You may have higher IPC on Apple CPUs, compared to even Intel, or latest AMD's CPUs. But you still may do less, with each cycle than you do with x86_64.

Sure, but that's not really ARM vs x86. That's the RISC vs CISC argument we've been having since the beginning of time.

Maybe CISC is the way to go. But people a lot smarter than me are still treating RISC vs CISC as an open argument.

Apple has also been investing a lot in their pipelining, and their pipelining architecture seems far superior to the other ARM variants. They would help address the RISC instruction set issues compared to a lot of other "dumber" ARM implementations.

And that is just instruction level problem. Show me an ARM CPU that can get 40 amps of power, or more. People forget that Apple's designs are running at low voltages, under 1.0v, and they are 5-7W designs at best with very low clocks relative.

From what I remeber, what David Kanter said based on discussions with Apple engineers - there is no way they can provide more than 40 AMPs to their system on chips.
It was 2-3 years ago, but I believe there was nothing that changed on this front. Its ARM architecture, after all.

I think this is the actual problem. Apple can't just pump more power into their existing design, and it's very likely not even going to be as simple as "just make it bigger."

Their chips perform extremely well, but scaling them larger is where they don't have experience or possibly even expertise.

However - You have to think it's something Apple is looking at. They could probably throw an existing A13 into a MacBook Air no problem. But they'd need to scale up when they start looking at MacBook Pros, or even Mac minis and iMacs.

But it's an investment. And it might be easier for them to throw some money at AMD to make the Intel problem go away.

I'm not really concerned that "big ARM" is just impossible, it just hasn't been done by a company with the resources of Apple. Apple has shipped big RISC chips before, and they could do it again if they had the will.

 

[Pro7913]

Yes. And I don't actually need to speculate on that in since Premiere runs on the iPad (along with a lot of competing editing apps.)

PP full version? You have no idea what you are saying.

You keep saying this but you don't provide any justification.

I know Geekbench isn't 100% accurate. But it also doesn't mean nothing.

I provided justifications and you are the one who is not listening.

 

[goMac]

PP full version? You have no idea what you are saying.

Sure I do. It's the same Premiere Pro engine that runs on the desktop. Adobe doesn't have two Premiere engines.

Same with Photoshop for iPad. It doesn't have the same UI (or a complete UI), but it's the same engine. There aren't two Photoshop engines.

I provided justifications and you are the one who is not listening.

You haven't provided a justification how Geekbench would be different on ARM.

 

[Pro7913]

Sure I do. It's the same Premiere Pro engine that runs on the desktop. Adobe doesn't have two Premiere engines.

Same with Photoshop for iPad. It doesn't have the same UI (or a complete UI), but it's the same engine. There aren't two Photoshop engines.

Photoshop is the only official app supporting both ARM and x86.

You haven't provided a justification how Geekbench would be different on ARM.

Since you are not reading, I'll pass you cause you are wasting my time.

 

[koyoot]

Their chips perform extremely well, but scaling them larger is where they don't have experience or possibly even expertise.

It has nothing to do with expertise. Based on latest ARM servers benchmarks, 16 core CPU, without Hyper Threading is consuming up to 80W under load. While being slower than equally power efficient, but 8 Core CPU, at the same tasks. Call it lack of software optimization, call it hardware performance. Whatever.

ARM CPUs are hitting Diminishing Returns field when scaled up. Its not up to Apple to fix this.

Its up to ARM to fix this.

You haven't provided a justification how Geekbench would be different on ARM.

Instruction length. And you know that perfectly well. They may score brilliantly in Geekbench, which executes simplest instructions.

But when you will put the load on those CPUs, the picture may be completely different.

And you know about it perfectly well, goMac.

 

[goMac]

Photoshop is the only official app supporting both ARM and x86.

Again, the Premiere engine is running on iPad and ARM. Adobe uses it in shipping mobile version of Premiere.

Since you are not reading, I'll pass you cause you are wasting my time.

You've provided nothing to read on why Geekbench would be different on ARM vs Intel.

 

[Pro7913]

Again, the Premiere engine is running on iPad and ARM. Adobe uses it in shipping mobile version of Premiere.

PP Rush is not PP. You just have no idea what you are saying.

You've provided nothing to read on why Geekbench would be different on ARM vs Intel.

Links and comments. Also, Koyoot provided what I've been saying.

 

[thisisnotmyname]

How come 12nm Nvidia GPU outperforms 7nm AMD GPU? You clearly have no idea what you are talking about. We are talking about reality.

Intel is not able to compete with AMD especially with their security flaws even with 10nm CPU. Even they bring 7nm, there is no guarantee they will be able to compete with multicores.

AMD is bad, Nvidia 12nm process is better than their 7nm

(a few posts later)

Intel is bad, they're stuck on 10nm process, AMD is good and already on 7nm.



Although I quoted you this isn't directed at you, just general observations of a very vocal group around here.

People need to stop being hung up on die size and benchmarks. Die size can be good for reducing power draw but not the end all be all to silicon performance. Transistor density, lithography methods (e.g. self aligned quad patterning), and instruction sets (e.g. various AVX sets) all matter as well as die size. Benchmarks can be useful as a proxy to performance but processor X being 3% faster than processor Y per core in raw number crunching doesn't necessarily equate to better performance for specific tasks. Unless you have tied your identity to the manufacturer of a specific component in your computer it just shouldn't matter this much. The question should be will this do the job I need it to in a manner that's effective for me.

IF Apple some day moves to AMD processors, so be it, if they continue to meet my requirements and the requirements of my employees I'll still be buying them. If they continue on Intel or move to ARM, again, doesn't affect me so long as the systems perform as I need them to. I'm not losing sleep over the possibility that someone may build a system somewhere that can perform a render 10% faster than my team can, my business isn't going to suffer over that, my life isn't going to change over that. I don't suffer from benchmark envy and I'm not in one of the very rare businesses where microseconds matter (idk, maybe everyone else on these forums is an algorithmic high frequency trader). As long as my team is telling me that this machine will meet their needs and I see the return on investment I'll be buying them. Lots of anger around here, I'd suggest taking a step back and asking whether this really matters to you, be a little "zen" as it were. (bad pun intended ? )

 

[cube]

AMD then launched K10 architecture (Phenom), which never performed or scaled well. It also had a serious bug (TLB bug) that when patched caused a 5-20% performance hit. It was AMD's Netburst moment that, like Netburst, drug on for almost a decade.

We are still using Phenom X3 and Phenom II X6. It did not make much sense to "upgrade" to FX at the price.