Unfortunately there are only angry and laughing face reactions neither of which are that good in my opinion. A simple thumbs down and an eye-roll emoji would be good additions.
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.
Intel Alder Lake vs. Apple M1
- Thread starter leman
- Start date
- Sort by reaction score
Unfortunately there are only angry and laughing face reactions neither of which are that good in my opinion. A simple thumbs down and an eye-roll emoji would be good additions.
Are you unaware that Cinebench literally is an application benchmark? It's the raytrace engine from Maxon's Cinema 4D app packaged up as a benchmark. It should be fully representative of Cinema 4D rendering performance.
Cinebench is fine for answering the specific question "how fast will Cinema 4D render on my computer". It's not a good general purpose CPU benchmark for more abstract questions like "is Apple's Firestorm core generally better or worse than Intel's Golden Cove core". You should never rely on a single specific bench for that kind of thing.
Crossplatform benchmarking for the purpose of answering that second question has other pitfalls, too. One of them is relative optimization levels. Cinebench R23 / Cinema 4D R23 were the first versions Maxon shipped which could run native on Apple Silicon. Do you believe that R23's raytracer has been as highly optimized for Apple's CPU as x86? I don't, I'd bet the R23 release prioritized stability and correctness of the port over performance.
Be nice if they could do that for the large iMacs generally, including those that don't have high CPU core counts (and thus may not be considered "Pro" models). Copying from something I posted elsewhere:
Since most programs are still single-threaded, anything beyond about 8 performance cores is overkill for the overwhelming majority of Mac buyers. [Of course, higher core counts are needed to compete in the pro desktop space.]
However, precisely because most programs can't distribute their workload across multiple cores, they rely on single-core CPU speed to get the job done. Thus, even with these new chips, there's still some wait time when doing heavy-duty processing. Consequently, I expect any power user (a much larger group than just the Mac Pro buyers) could benefit from far higher single-core speed.
Apple's done a fine job offering the high single-core speeds they have, especially with such low power consumption. But if Apple wants to improve the user experience for most of its customers, in addition to increasing core counts, they'll also find a way to further improve single-core speed—particularly in their desktop models, where they aren't quite as constrained by thermals and efficiency.
Probably what they will do with M2 Pro and Max is increase frequency because the process node difference will be less than previous chip releases, i.e. instead of 14nm to 10nm 4nm difference or 10nm to 7nm 3nm difference to 5nm 2nm difference to 4nm 1nm difference. M2 is supposed to be 4nm process node and I doubt they will be adding cores but instead will increase the frequency since the difference from M1 is just 1nm. How much they can increase the frequency on single core I am guessing will be big to keep impressing but I am just guessing!
It depends almost exclusively on the CPU. It doesn't read from the SSD, nor is it heavily affected by memory performance or other platform aspects.
When using actual applications as benchmark, there are also OS aspects to consider. For example, Apple enabled QoS CPU scheduling on the M1 Macs, which prioritizes foreground tasks and makes foreground apps seem "snappier" and thus may affect application benchmarks, but at the cost of background tasks. They did not enable this on the Intel Macs (although it would be technically possible), nor does Windows 10 have a comparable feature (Windows 11 does though).
It was historically actually a pretty good indicator for CPU performance. When you start making benchmarks with a mix of many different aspects like Geekbench, you always run into the problem of how to set the weights to calculate the overall score. It's almost impossible to avoid biases.
I have no idea. Do you? There are obviously many variables. E.g. it's possible that the compiler that the developers used on one platform is just better than the one on the other.
In any case, Geekbench also shows Alder Lake clearly ahead of the M1, so simply dismissing the Cinebench results as an anomaly without explanation doesn't seem well founded.
Apple's quality of service scheduling features have been part of Intel macOS since at least 2014 in 10.10.
QualityOfService | Apple Developer Documentation
Constants that indicate the nature and importance of work to the system.
No, it never was. Lots of forum warriors have claimed it is, and you may have incorporated that into your belief system, but that doesn't make it so. Nobody serious about trying to fully characterize a CPU's performance should rely on a single application-derived benchmark kernel.
How does arbitrarily deciding that one specific raytracing engine is the golden standard for performance testing resolve the problem of bias being hard to avoid? You're just deciding to be super biased, but in denial about it.
You certainly can criticize GB's design choices, weighting, and so forth. I don't think it's perfect myself. (Its own author probably doesn't think it's perfect.) But if I want some benchmark data to try to resolve any performance question other than "how fast is Cinema 4D's raytracer going to run", I'd choose GB5 over Cinebench in a heartbeat. If nothing else, since GB5 consists of many subtests and publishes individual scores, you can extract a subset of data which is more directly relevant to what you're interested in knowing.
With Cinebench, you can never know anything beyond "how does this one raytracing engine perform". Raytracing isn't a great predictor of a wide variety of real world problems.
I don't think you understand the point of questioning the utility of Cinebench. It's just not a very good general purpose benchmarking tool.
Agreed, a smaller process should allow shorter delays and thus higher frequencies. But I was wondering if, in addition, the architecture could be modified to optimize performance and efficiency at a somewhat higher frequency for their desktop chips.
I.e., consider their current process. For laptops, they found a great sweet spot at 3.2 GHz. But, if they weren't designing for laptops, might they have been able to, with the same process, produce an equivalent desktop chip (with the same unified memory architecture, assorted coprocessors, etc.) that was designed to run as efficiently as possible at, say, 3.8 GHz? It wouldn't be as efficient as the mobile chip, but it might still offer outstanding efficiency for the desktop category.
Last edited:
I too am genuinely interested in this topic.
Why 3.2Ghz currently?
Are these chips able to run at a higher frequency? Or is this the absolute fastest they can run? If so, what is causing the limit?
If power/thermals isn’t an issue, what else is a factor in frequency scaling the CPU?
They could have run at 4.0 GHz, but Jon Ive decided 3.2 was an aesthetically more pleasing number...something about refrences to ? and the shape of Apple's new HQ building. Too bad he didn't prefer ? ?.
To give you a more serious answer: I assume that's just the particular optimization they arrived at for performance and efficiency while designing the chip. It probably resulted from some complex combination of performance and efficiency targets, along with the results of their empirical testing during development.
I believe the OP said it's been opined part of the reason for the chip's extraordinary efficiency is that it's been highly optimized to run at that frequency. So that particular design might become both unstable and inefficient at higher frequencies—which is what motivated my question about whether a different design, with somewhat different optimization choices (but that adhered to their general unified architecture) could be created to operate at higher frequencies.
Last edited:
The behavior of CB23 on the M1 is known and been noted by reviewers: M1 P-cores are underutilized. No it isn’t going to E-cores, but for some reason the P-core is not ramping up all the way on that specific test. The why is unknown. I’ve seen some claims that CB23 gets lower scores on Intel Macs in macOS than those same machines in Windows through Bootcamp. But I’ve not seen that actually tested by any outlet so I’m dubious until it’s tested. CB23 is a bit of an outlier as a result - not an extreme one I grant you, but one nonetheless. I would be softer than @mr_roboto in that I would say normally CB23 is an okay benchmark, but he is right that strictly speaking it only tests one aspect of a CPU’s capabilities. Right now I’d say that, in other benchmarks which don’t have the problems CB23 has, the new ADL chips are beasts at CPU ray tracing. They are a huge improvement over the previous Intel generations and, yes, the M1 Max in raw performance.
Truthfully, if I wanted to be biased I could actually choose to highlight the CB23 results, rather than discard them. The reason is because while the relative performance, the raw scores, of the M1 Max doesnt look as good here as in some other tests, the number of joules the Max uses to run the benchmark is lower than it should be relative to Alder Lake. Yes, the Max is more efficient overall than the i5, but no it isn’t like 6-7x more efficient in most ST tasks. That would be pretty unfair to claim that’s representative of the Max’s efficiency advantage in most workloads because the firestorm’s uarch is simply not that much better than Alder Lake! It’s only about 4-5x more efficient in ST. ?
Last edited:
@mr_roboto has a few good posts.
Geekbench score for M1 Max. It's great
This is not exactly how us chip people would describe it, and there are a few minor mistakes, but it is conceptually correct. We rarely think about individual transistors; we're mostly concerned with gates. Gates are logic circuit elements built from at least two (in CMOS logic, anyways)...
forums.macrumors.com
Geekbench score for M1 Max. It's great
In turn, people like me often work on this from the other side without ever thinking about junction physics. Chip design requires a lot of contributions from different specialists! The main knob they might have to get more frequency out of the Firestorm core with no changes is boosting...
forums.macrumors.com
In short, Apple *might* be able to run more power through the core to up the frequency a lot, but even if possible the cost of doing so likely isn’t worth it. This is especially true in bigger chips where what you want is to maximize throughput. Right now Apple has a sweet spot where they can basically just keep adding CPU cores, run them all at pretty close to peak frequency, and not have to be as concerned with power draw or heat. An interesting follow up question though is why Apple only lets their cores run at absolute peak frequency when only one core is active regardless of the thermal capacity of the system? ?♂️
Last edited:
Maybe you can't relate because you've never left your house or something. Laptops are computers that get used OUTside.
How do you think the M2 will be compared to the M1?
on cpu level at most 10% improvement ?!
on the gpu, A15 is around 20% better than A14 just based on core vs core or just because the A15 iphone pro has 5 cores?
Im asking because , if the gpu core a15 is better than the gpu core a14...then the M2 will gain not just 2 extra more gpu cores, but also an better single gpu core...so the 8 gpu cores from the M2 will be slightly better than the 8 cores gpu from the M1, and if you want even more i heard the M2 will have the 10 cores gpu...so even more performance
Bottom line, what do you think? the M2 will have a better gpu core but also even more gpu cores than the M1?
PS: of course if the rumours are true
Last edited:
Little interesting Cinebench R23 score results.
Alder Lake i9 12900k limited at 35w and disabled 2 P-cores to simulate laptop version of Alder Lake with 6+8.
vs M1 Max at 30w
12900k - 14288
M1 Max - 12326
Alder Lake i9 12900k limited at 35w and disabled 2 P-cores to simulate laptop version of Alder Lake with 6+8.
vs M1 Max at 30w
12900k - 14288
M1 Max - 12326
We don't know whether they are underutilized or simply not very good at handling this specific kind of workload. The M1s are based on a highly focused design originally made for phones after all, not rendering workstations.
How come? I mean, the numbers speak for themselves. M1 Pro/Max is cooler, quieter, runs at the same performance on battery and battery life is significantly better. And the performance is almost the same in benchmarks - and that’s not counting Apple special stuff like ML or video decoders.
So in what world would Intel processors be better for Apple?
M1 performs amazingly well in high-complexity workloads such as scientific computing, compiler code, databases and also, yes, rendering. Cinebench is pretty much the only outlier among the common benchmarks. My suspicion is that its code does not provide enough opportunities for deep out of order execution which is not optimal for Apple. Another point in favor of this hypothesis is that Cinebench scales extraordinary well with SMT, which again suggests that a single thread does not fully utilize a CPU backend.
I don't doubt that it's good at rendering, but that doesn't mean it has to be equally good at handling the workload generated by Cinebench. For example, rendering highly complex scenes benefits hugely from high memory bandwidth (where the M1s benefit from their integrated RAM). But the scenes rendered as part of Cinebench aren't very complex, so it may more heavily depend on the raw CPU performance than a realistic rendering task.
I don't know. 50% multi-threaded in Geekbench is nothing to sneeze at.
What I was trying to say is that Firestorm is a very good all-round core. There are some areas where it has some deficits (arbitrary precision computation is one if I remember correctly), but it generally runs all kinds of workloads excellently.
But that's exactly the thing: when running Cinebench, "raw CPU performance" is not used — the chip backend spends a lot of time idling.
That's an excellent result, much better what I have expected! I would have preferred to see some SPEC results though. Looking forward to seeing mobile Alder Lake benchmarks.
Intel processors would be better for backward compatibility and running Windows apps. That functionality has been taken away. If you need to run abandonware, the Mac is a poor choice. Microsoft will bend over backward to keep old apps running. Apple will give fair warning and tools for third-party devs to update their software, but won't hold back changes to the OS to accommodate those devs who won't.
This is a major shift in software development, and the reason why you're seeing more ISVs turn to a subscription model. Back in the day, you would write your app, toss it over the fence and never worry about it. You might release a new version every now and then, or maybe not. When software was largely distributed over physical media, this made sense. Now that software is mostly delivered online, we're seeing more incremental changes over time. Some devs and users have adapted to this model, others have not.
Already we've seen Apple deprecate frameworks like OpenGL. They've transitioned the Mac from 68K to PPC to Intel and now arm64. They dropped 16-bit support, then 32-bit support. Some software users paid for no longer runs. Or in many cases, users are relying on applications that the vendor no longer supports. With Intel Macs these users could use Windows via BootCamp, or at least run it natively in a VM x64. Now the only choice is a VM running ARM and emulating x86/x64.
Apple often will rip off the band-aid rather than drawing things out. They did the same with moving to USB-A, and then USB-C, the 64-bit transition, and others. Microsoft will have difficulty getting ISVs to support arm64. Apple has simply given them no other choice.
You know, it's kind of funny. For years now Intel fans (and even Intel reps) kept saying that Cinebench isn't a real benchmark. Now that Intel is crushing Apple in Cinebench the tables seem to have turned and the Apple fans are saying it.
You know, it's kind of funny. For years now Intel fans (and even Intel reps) kept saying that Cinebench isn't a real benchmark. Now that Intel is crushing Apple in Cinebench the tables seem to have turned and the Apple fans are saying it.
Cinebench is popular and it’s a decent tool to test throttling under high load. But it was never a particularly good benchmark, especially given its limited scope. But it was only last year that it became apparent how „not good“ it is really, when people looked into CPU utilization.