... and when it scores faster than most PC laptops from 2021, since the 2021 MacBook Pro beat them.
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.
[CPU only] Apple M1/2(Max/Ultra) (TSMC 5nm) vs AMD Zen 4 (TSMC 5nm) - Technical Analysis
- Thread starter exoticSpice
- Start date
- Sort by reaction score
Why does the number of cores matter? What matters is power consumption, and it seems that Apple has lost most of its advantage on that front.
Eh, not quite. Number of cores do matter since that would tell the multithreaded performance. I forgot where I've found but 4 cores is the sweetspot for parallel tasks for the base M series chips. And also 28W is for that AMD chip vs 23W for M2. The extra 5W for ~2% performance gain is eh.
This conversation is getting ridiculous. People, stop throwing around arbitrary numbers and start thinking what you are measuring and for which purpose. "Better" is not an objective criterium. We have some here who only area about absolute performance (often according to some very specific metric), some who only care about performance per core, some who only care about maximal performance for a given thermal bracket, etc... and together this makes a huge mess.
Number of cores does not tell you the multithreaded performance, benchmarks do. And none of these things are trivial.
Why would that be? E-cores are decent contributors to parallel throughput, 4x E-cores add an equivalent of another P-core on many workloads. If you are interested in maximising the multicore performance, spawning only 4 threads on a base M-series wastes the performance potential.
Number of cores does not tell you the multithreaded performance, benchmarks do. And none of these things are trivial.
Why would that be? E-cores are decent contributors to parallel throughput, 4x E-cores add an equivalent of another P-core on many workloads. If you are interested in maximising the multicore performance, spawning only 4 threads on a base M-series wastes the performance potential.
I should have said scaling. 4 cores scale quite well but adding more on the base one will make it not worth it and 8 cores sometimes make it slower. I forgot where I have found it but IIRC it's on some software that's very niche.
True. Even looking at the breakdown, the higher multithreaded score shows that it isn't faster on most cases. That said, I'm not trying to be going on the deep end but looking at what normal people see.
That's not surprising. An E-core is only 1/3 of P-core performance if I remember correctly. And you are right that it will depend on the workload. That's why measuring is so important. Assymetric CPU configurations, dynamic frequency scaling and shared CPU caches make all of this very non-trivial.
That said, I've heard they are still trying to optimize the software so maybe they will improve the performance more. I just hope the Chandra team will take note on this. At least NRAO bother to update their software for Macs.
Something like this? The developers of Anaconda performed a benchmark on M1 and got this graph:
Anaconda | A Python Data Scientist’s Guide to the Apple Silicon…
Even if you are not a Mac user, you have likely heard Apple is switching from Intel CPUs to their own custom CPUs, which they refer to collectively as "Apple Silicon." The last time Apple changed its computer architecture this dramatically was 15 years ago when they switched from PowerPC to Intel…
www.anaconda.com
It's on some kind of support forum. I don't know where since it's a few months ago to see the progress of a possible future software in my course.Something like this? The developers of Anaconda performed a benchmark on M1 and got this graph:
View attachment 2199013
Anaconda | A Python Data Scientist’s Guide to the Apple Silicon…
Even if you are not a Mac user, you have likely heard Apple is switching from Intel CPUs to their own custom CPUs, which they refer to collectively as "Apple Silicon." The last time Apple changed its computer architecture this dramatically was 15 years ago when they switched from PowerPC to Intel…
The results given by that user have 8 cores sometimes regressing.
Last edited:
If "a core is a core" as you claim, then there would be no need for P and E cores in either Intel or ARM-based chips. But that clearly is not reality.
The 16 E-cores on the Intel 13900K are significant contributors for multi-threaded performance/benchmarks. For example, Cinebench R23 is around 1.68x faster with P+E cores vs P cores only. Without all those E-cores 13900K is just a lower tier 13700K with half the E-cores.
Not everyone's two legs are the same length so are you going to also start claiming unfair that the uneven leg runners are one leg?🤣
Not everyone's two legs are the same length so are you going to also start claiming unfair that the uneven leg runners are one leg?🤣
That decision is because Intel try to compensate for their weakness. If they reduce the E-cores, 7950X would be a lot faster re on multi-threaded but they can even it out for more P-cores at slightly lower clock speed. That said, it wouldn't need more E-Cores if they bump up the # of P-cores and lower the clock speed. 13900K would still be the higher SKU than 13700K if they try to reduce keep the # of E-cores and bump the # of P-cores instead.
E-cores help the Mx last longer on battery at the cost of lower performance. Complaining that the Mx is penalized in multi-core tests because of the E cores is as laughable as complaining that the 7840U is penalized in battery life tests because it only has P cores. Each company chooses what type of cores it uses to achieve its goal.
In your attempt to defend your "a core is a core" claim, you literally pointed out the differences between the cores. But it seems that in your mind, those differences are meaningless even though they have real-world implications on multiple levels.
It wil actually penalized on battery life. The commentator have said that it runs at 28W. I imagine going down to 20W will incur a lose of performance.
I guess people complain that a comparison between a 4+4 M2 and an 8+0 7840U in a multicore benchmark is unfair because two E cores perform similarly to a P core. So they may believe that a fairer comparison should be between an M2 and a 6+0 chip.
But the fact that two E cores perform like a P core in a multi-core comparison doesn't mean that the M2 has eight cores. Those four E cores in the M2 can run four light tasks and that helps the Mx last longer on battery life in everyday tasks.
So, the comparison between 4+4 M2 and 8+0 7840U is fair because both chips have eight cores, since a core is a core regardless of whether it is a P or E core.
I believe the common take is that 4 E-cores for AS is equivalent to 1 P-core.
I'm with @Xiao_Xi on this one. The problem stems from the fact that people compare different things. If one wants to compare advancements in CPU cores specifically, Apple is obviously far ahead AMD in everything but possibly some niche vector throughput workloads. But if one talks about performance, it makes sense to compare products, not cores. Yes, Apple uses a 4+4 core configuration, which potentially puts it at a disadvantage in multicore benchmarks against a 8+0 core configuration of mobile Zen4, but that's just part of different product tradeoffs and design decisions. Both M2 and 7840U are comparable products, with similar die sizes and manufactured on essentials the same process node, and as such I think it's perfectly fair to compare their relative merits.
AMD's Zen4 has very good energy efficiency (much better than Intel's latest offerings and comparable to latest Cortex-X ARM series), so using 8 main cores gives AMD a scalable product that can deliver some good multicore throughput at very low energy cost. Especially compared to Intel, AMD is in a very good position, because Intel simply cannot compete in this lower power bracket. But I think this is also where Apple's silicon superiority becomes clear. In single-core, the 7840U is still struggling to beat the now two years old M1 (despite a 2x difference in power consumption), and in multicore, Apple's current 4+4 config is sufficient to match AMD's 8-core (except in some niche benchmarks where the x86 CPU can play it's advantages like higher clocks). And of course, there might not be a large performance difference, but Apple can pack that kind of performance into a passively cooled chassis — I don't see AMD doing this any time soon. That's what dedicating large blocks of the core to power-saving technology gives you.
In the past, one decisive advantage of AMD was cost. AMD was able to offer these 8-core SKUs for a very good price, making them a good choice for many users who value versatility. Zen3 was already a kind of discount Apple Silicon if you want (just with a significantly lower single-core performance, which Zen4 fixes). But from what I understand, Zen4 Phoenix is a fairly large die, and I am wondering whether AMD can keep their price advantage.
AMD's Zen4 has very good energy efficiency (much better than Intel's latest offerings and comparable to latest Cortex-X ARM series), so using 8 main cores gives AMD a scalable product that can deliver some good multicore throughput at very low energy cost. Especially compared to Intel, AMD is in a very good position, because Intel simply cannot compete in this lower power bracket. But I think this is also where Apple's silicon superiority becomes clear. In single-core, the 7840U is still struggling to beat the now two years old M1 (despite a 2x difference in power consumption), and in multicore, Apple's current 4+4 config is sufficient to match AMD's 8-core (except in some niche benchmarks where the x86 CPU can play it's advantages like higher clocks). And of course, there might not be a large performance difference, but Apple can pack that kind of performance into a passively cooled chassis — I don't see AMD doing this any time soon. That's what dedicating large blocks of the core to power-saving technology gives you.
In the past, one decisive advantage of AMD was cost. AMD was able to offer these 8-core SKUs for a very good price, making them a good choice for many users who value versatility. Zen3 was already a kind of discount Apple Silicon if you want (just with a significantly lower single-core performance, which Zen4 fixes). But from what I understand, Zen4 Phoenix is a fairly large die, and I am wondering whether AMD can keep their price advantage.
More like one E-core is roughly 1/3 of the P-core, but at 1/10 of power consumption if I remember correctly. The power consumption of those cores is absolutely insane. We are talking about 0.3-0.5 watts or so.
Yeah. I've read that thread here and I can't believe how little it sips power even in extended period of time.
That said, I hope Apple increase their P,-cores for the base models in M3 and beyond.
I had read that one P core performs like between two and four E cores. For example, the Anaconda graph above shows that one P core performs like three E cores. The graph shows that four P cores can achieve 60 and the eight cores reach 80. So, each P core adds 60/4=15 and each E core adds 20/4=5.
Apple has shown that a hybrid architecture with P/E cores has advantages and that's why AMD will use it next time.
Well, iPhones has been Apple's hybrid core architecture test beds for a few years, since the A10 I believe.
I think this is fairly unlikely since the design strategy works fine. Yes, they won't win any multicore benchmarks with a 4+4 designs, but objectively speaking, their design works very well for thin and light notebooks. You can excellent everyday performance, with burst capability of much larger computers, paired with outstanding battery life. People who buy MacBook Air probably aren't going to run long chains of parallel computation on it, so it's a moot point to begin with. And for a casual photo/editing session it performs as well as anything else in the same price category anyway.
Th only problem with the 4+4 design of course is desktop. But I think Apple might as well throw in the towel and simply market the base iMac/Mac Mini as home/office class computers, where sustained multicore is not really needed. Gaming might be a bit more problematic, but that can be addressed by making chips with higher clock ceiling.
Aren't Apple marketing the iMac and Mac Mini as office and home class computers? Their iMac ad shows home usage.
I'm just hoping that the node shrinkage and innovative semiconductor processes will make it a reality of going beyond the current config. That said, maybe they will invest the silicon to bigger cores and higher clock speed.
Yes, they do
It's just the question where are they going with all that. Their desktop strategy in the last couple of years was... inconsistent at best. Will they leave those models as entry level pricey lifestyle products, will they try to improve the performance to match mid-level PCs, or will they continue treating them as an afterthought? Unfortunately, the rate of die shrinks appears to be slowing down, which puts a hard limit on what one can do. Not to mention that new processes are getting really really expensive. There are also some obvious scalability problems associated with Apple's large SoC strategy...