Apple dumps Intel and shows Final Cut Pro on new Mac with Apple CPU !
- Thread starter joe1946
- Start date
JAMedia
Well-known member
I know I was being humorous.
Apple supporters get very protective and its as bad as arguing politics or region with them
Last edited:
https://www.tomsguide.com/news/macbook-arm-leak-just-revealed-bad-news
I'll add what I posted earlier in this thread. Microsoft Windows 10X, that was supposed to also be made for ARM CPU's without the compiler, is likewise being pushed into 2021. Both systems can do it with the compiler but lose performance because of it. I'll speculate that both companies are waiting for the new chips to arrive and then to keep working on the software side to achieve full compatibility.
https://www.somagnews.com/amd-may-release-arm-like-ryzen-cpus/
Intel already has the big.LITTLE chip in Lakefield. AMD is working on it. This doesn't directly relate to the ARM CPU's per se but should make the Windows OS portables be more efficient while maintaining decent performance.
Competition is a wonderful thing. For the consumer.
And sneaking from behind is Samsung.
http://www.businesskorea.co.kr/news/articleView.html?idxno=50279
http://www.businesskorea.co.kr/news/articleView.html?idxno=50279
Power, power and more power. I mean, less power.
NVIDIA is attempting to get into the ARM based server business, with a little help from its friends.
https://fudzilla.com/news/51345-nuvia-phoenix-cpu-is-faster-than-all-arm-x86-competition
NVIDIA is attempting to get into the ARM based server business, with a little help from its friends.
https://fudzilla.com/news/51345-nuvia-phoenix-cpu-is-faster-than-all-arm-x86-competition
Multi-core processors allow for massive parallel programming. This is where ARM chips can catch up with Intel and AMD. A single chip superiority is still with Intel/AMD but the multi-core performance depends on a number of other variables, such the cost of CPU/GPU, energy consumption and the software compatibility.
Below are some examples of the multi-chip solutions with several familiar names being mentioned.
https://builtin.com/hardware/parallel-processing-example
And for more technically inclined (of course, some had worked are are working in this field and know way more than covered here). Nonetheless.
https://computing.llnl.gov/tutorials/parallel_comp/
Below are some examples of the multi-chip solutions with several familiar names being mentioned.
https://builtin.com/hardware/parallel-processing-example
And for more technically inclined (of course, some had worked are are working in this field and know way more than covered here). Nonetheless.
https://computing.llnl.gov/tutorials/parallel_comp/
joema
Active member
It appears plausible that Apple Silicon CPUs will eventually use higher (maybe significantly) core counts than Intel. However (as indicated in one of your links) this is constrained by Amdahl's Law: https://en.wikipedia.org/wiki/Amdahl's_law
Some things you just can't efficiently parallelize. Intel does currently have superior single-thread performance and with their new "Sunny Cove" micro-architecture this is expected to further improve.
Amdahl's Law is fundamental, but viewpoints and technical approaches change. When debating whether a few faster CPUs were better than many slower CPUs, Seymour Cray once responded "If you were plowing a field, which would you rather use: two strong oxen or 1024 chickens?"
However the supercomputer industry has totally changed to massively parallel designs, whether on the CPU or GPU side, or both. So for the type of algorithms they use they found ways to greatly parallelize those. It's unclear if that can happen for more conventional business software. But maybe it's just institutional lag from software developers who haven't had access to many-core machines or development tools which streamline coding for those.
Thomas Smet
Veteran
If I could train the chickens I would use them. This isn't the most accurate comparison because they are very different animals that don't have the bodies for the same kind of work. You could have 10,000 chickens and it would still be a disaster. I think multiple cores is a very different scenario where the big core and little cores and the exact same kind of animal. So in this case its more like two really big oxen or 1024 smaller oxen. Both sets capable of doing the same kind of work.
I always thought a better analogy is vehicles. You can have a four seater car and a 12 seater van. The van will get 12 people to a destination at the same time as 4 people in the car so you get more people there. Once you add four cars to the mix however you are now able to transport 16 people. Each car can carry less people but together they still transport more. More gas and cost of course but it can get more done. Those cars can be smaller and less expensive as well and may get closer to the cost of the large 12 person van.
As it is CPUs are not really like a 4 person vs 12 person car. They tend to be much closer in clock speed. Like 4.0 Ghz vs 3.2 Ghz which is nowhere near 3x faster. Multi-core tends to only take a small speed hit. Some CPUS like arm are much slower per core but again they can do the same kind of work.
What I see in the CPU race is that the ARM chips are catching up in the single core performance to Intel/AMD but are still one half to two thirds slower. However, one can probably put in a 16-core Big/Little ARM for the same price as Intel/AMD and get a superior performance on most tasks that can be computed in parallel. Which, by now, are most tasks.
But to ensure full compatibility with every popular software package, Apple, Microsoft and Google need to rewrite a lot of software from the x86 to RISC. Otherwise, the performance suffers when grounded through a compiler.
And then it becomes a hardware/software race. Amazon, who needs a huge amount of server power, has moved to ARM with own Graviton series of chips. NVIDIA is building its own servers with a heavily redesigned ARM. Then there are Ampere and Marvell. And then the question for an individual consumer becomes about the higher end will moving lower or the lower end higher. Off the latest news, Huawei is releasing a 24-core ARM chip workstation that performs on a level of a $440 i9-9990. Ampere has 80 and 128 core chips. And a few years from now, 80 cores is what one might get in a high end desktop. And it won't have Intel Inside.
More importantly, though, it'll be a wholly competitive market with many players vying for the slice of the pie and the Wintel dominance will be an ancient memory.
PS. Apple is supposed to be putting in a custom AS 16-core into the next generation of Macs, while Microsoft is waiting for a Qualcomm Snapdragon 875. Samsung, as mentioned a few days ago, is seeking partners, having not done so well with its Exynos line of processors (and the Euros are not happy about that).
Last edited:
combatentropy
Veteran
Apple's A13 seems to match them. That's what your article said: "Apple scores close to a single-threaded performance of 10 nm Sunny Cove and wins against AMD's Zen 2 7nm and does it under 4W." --- Nuvia Phoenix CPU is faster than all its ARM X86 competition. Another source echos it: AnandTech Testing Finds Apple’s A13 Chip Essentially Matches Intel and AMD Desktop Performance
Now elsewhere I did find some comparisons where x86 chips won out. It's hard to nail down, because there are so many models of x86 chip to compare. I would not be surprised if the fastest x86 chips, often bought by gamers and such, outrun the fastest ARM chips. But among median x86 chips (even desktops), it looks like ARM (at least Apple) has matched them, and has done so with a fraction of the electricity.
P.S. If you really want to geek out about computer processors, your research is incomplete until you have read about RISC-V.
JAMedia
Well-known member
less electricity is going to mean less heat. So less cooling. Quieter laptops. Cortex parts don't usually run with cooling systems like x86 do.
not really. It's not one for this race.
Besides there are dozen's of other m/cpu's out there. I think x86 is less than 4%of the processors in the world.
Kupeng 24 core, 2.6 GHz ARM based chip "roughly" matches up with Intel i9-9900K. I9-9990K is an 8 core, 16 thread, 5 GHz chip. So, basic arithmetic places the per core performance of the Kupeng chip at about one third. That's in the multi-core test.
Single core scores - say that twice - for Kupeng are unavailable (or probably available in Chinese).
The Microsoft Surface Pro and Samsung Book S, both running off Qualcomm's Snapdragon, were tested as being significantly slower than the corresponding Intel chips but that's due to other reasons (OS problems, compilers, etc.).
This is why both Apple (MacOS) and Windows (Windows 10 X) are delaying the introduction of their ARM based releases. They need to make sure everything matches from both the software and the hardware point of view.
PS. Various sites tested the server chips by Ampere and Marvell but those are battling Intel under the TCO (Total Cost of Ownership) premise. Competition is good.
Samuel H
Major Contributor
The ARM cores inside an Amazon Graviton 2 server can match the single-core performance of AMD Zen1 and Intel's Cascade Lake: each can win or lose in certain tasks, but all of them win at something and none is clearly above the others (maybe the intel one is slightly ahead, but Cascade Lake is a 2019 microarchitecture, while Zen1 is 2017; the Graviton 2 uses Neoverse N1 michroarchitecture, which is 2019 as well but is running at less GHz here).
https://www.anandtech.com/show/15578/cloud-clash-amazon-graviton2-arm-against-intel-and-amd/5
https://www.anandtech.com/show/15578/cloud-clash-amazon-graviton2-arm-against-intel-and-amd/5
iain.bason
Well-known member
I think it might be more accurate to say that single thread performance of even lackluster CPUs is now adequate to perform most tasks that can't be parallelized. In other words, most of the tasks that require massive amount of computing power are easily parallelized.
It takes a lot of human effort to match the performance you get from a compiler even for a small computational kernel like a dot product. As a consequence very little software is written in assembly language. I would expect that far less than 1% of the code in most applications will need anything other than a recompile to get the best performance out of ARM. The bigger problem will be that the programs probably all have bugs that will be exposed because of subtle differences between x86 and ARM, and it will take the developers some time to discover and track down all of them.
The "server frequencies" is an important caveat. It's sort of like comparing Veloster N and Porsche 911 in rush hour traffic. It's an imposed limiter.
Intel also does this.
https://www.pcworld.com/article/357...erformance-then-battery-life-for-laptops.html
And on the same topic of power consumption and heat.
https://www.forbes.com/sites/gordon...pgrade-iphone-11-pro-max-update/#7f9076fc6051
A14 is supposed to be a major jump off A13 and Mac Books will/should have a different/upgraded CPU versions than smartphones, with more high performance rather than power saving cores.
https://www.forbes.com/sites/gordon...pgrade-iphone-11-pro-max-update/#7f9076fc6051
A14 is supposed to be a major jump off A13 and Mac Books will/should have a different/upgraded CPU versions than smartphones, with more high performance rather than power saving cores.
Samuel H
Major Contributor
In the anandtech article I linked, for single-core performance, the 2019 ARM at 2.5GHz is matching a 2017 Zen1 at 2.5-2.9 GHz and a almost matching a 2019 Intel Cascade Lake at 2.9-3.2 GHz. It's true that standard desktop parts will run at higher rates, and I don't know if the ARM can do that, but it's definitely not "a third of the single core performance" or anything like that. It's in the ballpark. Make one tweaked for your energy budget and with plenty of cores, and it will be competitive. You can find my posts at the beginning of this thread and see I didn't think this would be the case, but I'm not so pessimistic about Apple's move to ARM anymore. It may just work. Or maybe it will be slow and just make batteries lighter, we'll see XD