Thanks for sharing your findings. The GPU having an MMU is definitely a good thing for the contiguous memory question. As for sharing the same translation tables, on second thought that can't be possible because at any one time you could have a different app with different incompatible memory mappings on each CPU core. That's what probably makes it necessary to use a separate virtual address space on the GPU, otherwise pointer would need to be like tagged with the owning application.
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.
Does an M1 with 8GB of Ram out perform an I7 with 16GB of Ram?
- Thread starter TheOutspokenMan
- Start date
- Sort by reaction score
Oh, that's already the case today. Applications do not share virtual memory space, this is strictly isolated. I don't know the hardware details of how this is implemented, but I don't think Apple's approach will be particularly sophisticated, it's probably the controller loading an TLB for particular application on a GPU core and sending off some work. I'd be surprised if Apple GPU can concurrently execute threads from different processes, but I have not investigates this. Apple does have some patents about GPU context switch, but it's not really clear to me whether "context switch" here means switching between work submitted by different applications or simply different thread groups from the same application.
On a related note, some GPUs do have more sophisticated virtual memory management. For example, Nvidia has unified virtual memory with page fault detection — this allows you to pass pointers between the CPU and the GPU without any marshalling, the GPU will stall and issue a DMA transfer if you are accessing a page that has been modified by the CPU.
i don’t think this is accurate at all. if you have two 4K or higher displays hooked up to an M1 mac mini with 8 GB of universal RAM you’ll see even at startup 6 GB taken up before you launch an NLE application, a video file compressor/transcoder or a bunch of DTPs apps or CAD or whatever is going to them require lots of memory to run efficiently.
my old i7 MBP from 2015 with 16 MB didn’t see system resources eating so much performance away as my M1 mini. the mini M1 makes up for it in pure throughout of the CPU and GPU but there’s no way my next M2 studio/mini or M3 MBP purchase will have anything less than 16 GB whatever the apple tax is on memory. it’s just two important to skimp on memory and that’s been the case since i started using Macintosh with the 512 KB second release of Macintosh. i used to DYI my RAM upgrades as soon as i could afford to with 3rd party retailers of quality DRAM.
A 4K framebuffer is just 32MB. even if you go full floating point with tripple buffering you still only need 400MB. That one need a lot of VRAM to drive high-res displays as such is an urban myth. It's just that if you run multiple displays you will probably have more content (i.e. application windows), and that's what will drive your RAM usage up (but even than video content doesn't use much RAM).
Strictly speaking, @chabig is 100% correct. The amount of RAM has nothing to do with performance as having more RAM won't allow you to do work any faster. How much RAM you have only becomes a problem if your active work doesn't fit in it. Think about it this way: if all you need to do is transport a small box, having a bigger truck won't help — you want a faster truck. The size of the track only becomes relevant if the box does not fit inside, which might make it necessary to make multiple trips. Or another way to look at it — it's not that more RAM improvers performance, it's that insufficient RAM (for the given task) might hamper performance.
myth or not i turn one display off at times just to stop my GUI stuttering. and i’ve no idea why 6 GB of universal RAM is being used before i start up a big memory chewing application. sure i load extra fonts with suitcase and have a bunch of tiny drivers for tablets and stuff some of which i suspect might leak memory but 6 GB out of 8 GB being used means swap comes into play quickly even before Safari and Firefox put their massive footprints on my memory load.
so leaving video RAM out of it, you’re saying that memory swapping doesn’t slow a system down? there’s a lot of benchmarking videos i’ve watched that show when memory swapping happens, but slow downs happen.
i do run lots of apps contiguously ans i do have to restart my mac several times a week simply because it’s locked up so bad i can’t force quit from applications and memory has jammed up solid.
what i’m saying is the 8 GB of RAM is insufficient for a lot of real world computing tasks. especially if you have a dozen productivity apps open at the same time and maybe a photoshop or indesign type app. i run aground on this daily so i know it from experience. i keep activity monitor open all the time. the biggest offenders are Browsers which seem to cache everything rather than kill pages’ JS off. i once used a plug-in for sleeping old tabs in Safari and it did seem to improve performance noticeably.
Do these 6GB include or exclude the buffer cache? Anyhow, the operating system itself has a large number of background processes, the window manager, the Finder, the Dock, you name it. They all need system memory. Point is, it's more than GPU related things that use a lot of RAM on startup before opening a single application. Another point is that having more displays attached doesn't just consume more RAM, it also consumes more RAM bandwidth that is not available for the CPU anymore. But I'd agree that 8GB is nearly always not enough for a powerful machine like an M1 Mac so nearly everybody should get at least 16GB to prevent swapping all the time.
That's not at all what @leman said. He was saying that an 8GB system and a 16GB system with the same CPU will exhibit the same performance as long as the working set fits into both systems' RAM, i.e. in the case of no swapping.
How are you connecting your displays? I thought M1 only supported one external display? Stuttering can have many different causes, including your machine being RAM started because of applications you run, but you won't get into a memory starved situation just by using a couple of 4K displays. Bandwidth might be more of an issue as a 4K display at 60FPS needs roughly 2GB/s (before bandwidth compression etc.), and M1 only has 60Gb/s.
By the way, the amount of memory used is not indicative of anything without a more detailed analysis, since RAM is used opportunistically. I have 32 GB of RAM on my laptop, and currently 27GB is shown as in use, even though I don't have any heavy apps open. But looking at the vm_stat output, only actually 10GB is being actively used, the rest is caches and speculative stuff.
That's the second part of my post. You won't get heavy swapping if you have sufficient RAM for your workloads. From what you describe, I fully believe that you need more than 8GB (or maybe even more than 16GB if you do heave CAD work). But that's exactly the point: it's about how much RAM one needs, not about more RAM giving you better performance.
Stop worrying about what you see in activity screen. I have 64GB Ram on my M1 Max MBP. It shows I have little over 10 GB Free with browser, word, preview and couple of IDE open. When I run Vm_stat it shows I have zero page throttling, barely any swapin and swap out, though it shows over 9 GB swap space. Most of my memory pages are used for cache, active caching and inactive, available to use if needed.
What you describe is not having enough RAM for the tasks, not because you have 4K displays.
The 'Memory used' readout in Activity Monitor doesn't tell you much - if Mac OS sees RAM sitting around doing nothing it will use it for things like file cache. The important figure is the "memory pressure" which will show if the system is having to do a lot of swapping to disc, which is what kills performance.
If you actually needed a 16MB MBP in the past then, no, you probably shouldn't get an 8MB M2/M3. Apple silicon is efficient, but it's not magic. Anyhow, the Studio starts at 32GB and anything with a M1/M2 Pro starts at 16GB. I'd rather hope that the release of the M3 sees the end of the 8GB models - its 2023 (maybe 2024 before we see M3) and 16GB is not an unreasonable minimum spec for a laptop costing the thick end of $2000.
Two 4K displays is a lot of pixels to shift - and if you're using scaled mode everything is being rendered at 5K internally and then downsampled by the GPU. Bear in mind that, although the GPU in the base M1/M2 may thrash Intel's integrated graphics that's a pretty low bar, and while it hits above its weight, it's not a particularly powerful GPU in the grander scheme of things. If I wanted two 4K or better displays (esp. to run in scaled mode) for any sort of graphical work I'd probably be looking at a Mx Pro, at least (if not for the better GPU, because it had the extra ports to connect the things to & offered the possibility of a 3rd display). Which makes the 8GB thing irrelevant.
The base M1/2 (non-pro/max) is, after all, the entry level Apple Silicon processor for the MBA and lowest-end Mini, so it's quite optimistic to see it as the 'modern equivalent' of a higher-price-point (when new) Intel Mac.
Memory pressure is a heuristic that's sometimes right and sometimes wrong. With modern fast-ish SSDs, it's easy to get into situations where memory pressure is green but the Mac is unusably slow.
The most common mechanism I've seen is repeatedly reading files that are larger than memory size. If you read the files sequentially, it only takes seconds or at most tens of seconds to read through the files. In contrast, even when you are actively using multiple apps, you often leave them in the background for several minutes.
Because there are order-of-magnitude differences between the frequency you are using the files vs. the background apps, macOS thinks it's better to use more memory for disk cache and swap out the apps you are not using at the moment. When you switch to the background app, it must be swapped in. Because SSD latency is high, it often takes from seconds to tens of seconds before the app is working normally. But because you are still reading the files, and because disk caching and swapping are more or less the same thing, macOS thinks that there is only minimal swapping and everything is fine.
That sounds more like an Apple problem to me, but there are definitely i7 CPU's that are a lot faster than the M1. The last i7 (10th gen 10700K) from Apple in my iMac actually out benchmarks the M1 Pro 8 core, and current i7's (13th Gen 13700K) are over twice as fast. And since my iMac has 128G of RAM, it can do quite well at the heaviest of jobs and I use it instead of my Mac Studio Max for almost everything.
The M series CPUs aren't the miracle performers people think they are around here. They are more efficient and performance per watt is better, not doubt of that, but faster overall, no.
Problem? That's simply because Apple stopped putting Intel chips into their laptops after starting to use their own.
That sounds more like an Apple problem to me, but there are definitely i7 CPU's that are a lot faster than the M1. The last i7 (10th gen 10700K) from Apple in my iMac actually out benchmarks the M1 Pro 8 core, and current i7's (13th Gen 13700K) are over twice as fast. And since my iMac has 128G of RAM, it can do quite well at the heaviest of jobs and I use it instead of my Mac Studio Max for almost everything.
The M series CPUs aren't the miracle performers people think they are around here. They are more efficient and performance per watt is better, not doubt of that, but faster overall, no.
I thought this discussion was about mobile CPUs? Of course a newest desktop i7 with 200+ watt power draw will outperform the 15-20W M1...
But if you look at the newest ultramobile U-series Raptor Lake i7 CPUs (such as the flagship 1365U), those are still slower than the two year old M1. Even the P-series are just slightly faster (and that only in naive multicore workloads).
There was no mention of mobile or not in the original post.
And anyway, the newest i7 mobile, the 1280P is faster than the low end M chips, and then there's the i7-13700HX which is technically a mobile chip though it's pretty fast.
My comment about performance per watt is my response to your question though, the M series is great for low powered devices.
Exactly, intel does have lower power chips, but that doesn't mean that M chips are faster performers overall. The original question was a bit too open ended and flawed (M series chips aren't faster than all i7's but some i7's are slower, and that doesn't even count RAM) to be answered easily, but if performance is the major decision point, it's clear what type of processor you get, and if battery life and power are your major decision point, it's also clear which type you get!
My main decision point is pure software, so I own machines that run what I want and need, and that means I have both types of processors.
As often happens with your posts, this is a mix of counterfactual and irrelevant things. First, let's cover the counterfactual, M1 Pro 8-core (6+2) vs i7-10700K iMac (8+0). You were not speaking the truth here:That sounds more like an Apple problem to me, but there are definitely i7 CPU's that are a lot faster than the M1. The last i7 (10th gen 10700K) from Apple in my iMac actually out benchmarks the M1 Pro 8 core, and current i7's (13th Gen 13700K) are over twice as fast. And since my iMac has 128G of RAM, it can do quite well at the heaviest of jobs and I use it instead of my Mac Studio Max for almost everything.
iMac20,2 vs MacBook Pro (14-inch, 2021) - Geekbench
browser.geekbench.com
Next, the irrelevant. M1 Pro 8c is a ~40-50W laptop chip. The 13700K is a 253W desktop beast with 16 cores (8+8). Nobody should care about comparisons between chips with a 5x (or more) ratio in power budget.
If we take a look at the i7-13700HX, also an 8+8 Intel chip but with power limits set at laptop levels, and compare it to M2 Pro 8+4 rather than M1 Pro 6+2, what do we find? M2 wins.
HP OMEN by HP Laptop 17-ck2xxx vs MacBook Pro (16-inch, 2023) - Geekbench
browser.geekbench.com
Last edited:
Fair enough, but I wasn't going by geekbench, and what I looked at was pretty different. But even given your benchmarks, the i7 didn't lose by much, and it's 3 generation sold. I do wonder if those numbers you quote are correct. And looking at the geekbench browser, the i7 10700K is definitely faster than the M1 Pro. 7273 vs 8845. So something's got to be off on however you searched for that link. The processors look right, but the results don't match geekbench's own number by processor.
10700k - Geekbench 5 CPU Search - Geekbench
browser.geekbench.com
m1 pro - Geekbench 5 CPU Search - Geekbench
browser.geekbench.com
(The M1 Pro looks closer to the 10 core part, not the 8 core which is what that page says). Geekbench seems to have a problem.
Anyway, you really shouldn't say people are saying counterfactual (lies) so readily, you never know, you might be the one that's wrong, and *any* benchmarks aren't consistent enough to be proof enough to call someone a liar.
That's irrelevant to me (total power). I like performance and a few more dollars, a month is cool if it makes me more money. You might not think that way which is okay, but you absolutely don't have a right to say I shouldn't think like I do.
In the benchmarks I look at, the M2 Pro isn't even close to the 13700HX, maybe 75% of the intel part. (Passmark and yes, I know you guys hate that, but it fits more with my experience). And looking at geekbench, multicore, the 13700HX is even higher than 75% faster. 11999 vs 18419! Single core it's faster too, but I don't really care about that. 1954 vs 2151.
>HP OMEN by HP Laptop 17-ck2xxx vs MacBook Pro (16-inch, 2023) - Geekbench Browser
LOL.
What I find interesting is how close the i7 in my iMac is to the speed of the M1 Pro -- it felt like that to me, but I don't run benchmarks to tell exactly. And yes, I noticed what you posted was from geekbench6 and what I posted was from 5, but I have a problem with the way geekbench displays results, you can't just look up by processor and doing it by machine vs. an average of results from many more than one machine makes more sense to me -- with one machine you just don't know enough, it might be underclocked, or constrained cooling, or many different things that can effect a benchmark.
Last edited:
If you want to compare scores to the ones I posted, you must use GB6 results. GB scores are normalized to a baseline or reference computer, and GB6 updated to a new baseline computer, so its scores aren't on the same scale as GB5.
I used results for an iMac with the 10700K specifically to match what you own and seemed to be posting about. It isn't surprising that there might be higher 10700K scores out there. The 10700K is a multiplier-unlocked chip, and overclocking enthusiasts are much more likely to download and run benchmarks than the general population. Since GB's free version always uploads results, OC'd runs are probably overrepresented in GB's database (especially for 'K' suffix unlocked CPUs).
Also, many late model 27" iMacs went the opposite direction of overclocking. Intel has been ratcheting TDP ratings up in its consumer desktop chips for many years (by quite extreme amounts recently, hence these silly 250W chips), and Apple seemingly didn't want to redesign the prosumer version of the 27" iMac to accommodate this. Late model 27" iMacs with high end processors had a power cap set somewhere around 80W-90W regardless of Intel's TDP rating, which for the i7-10700K is 125W.
(the iMac Pro did get a better cooling system to handle up to ~140W TDP Xeon chips, but Apple never brought that internal redesign to the mainstream 27" iMac)
Thanks to prior history with you, I'm quite comfortable saying "counterfactual". You often double and triple down when unequivocally proven wrong, so why should I trust you to argue in good faith?
You can brag all you like that you can afford the extra power bill (LOL); that's irrelevant to my point. You're trying to be smug about a 250W space heater beating the CPU performance of a ~50W laptop SoC, and I absolutely do have the right to point out that you're being absurd. Even more so when considering that the Intel chip is mostly CPU cores, while the Apple chip is more GPU than CPU. These are products designed for very different purposes.
M1 Ultra is the only Apple Silicon design which tries at all to address the market that wants high multithreaded CPU throughput. It's not a perfect comparison point, as its die area is skewed even further towards GPU cores than M1 Pro, but it's as good as we can get. It beats the 13700K's multi-core score in both GB5 and GB6.
ASUS System Product Name vs Mac Studio - Geekbench
browser.geekbench.com
ASUS System Product Name vs Mac Studio - Geekbench
browser.geekbench.com
There you go telling me what to do, that's not acceptable.
Now that's a fair comment, and I'll keep that in mind in the future.
I haven't heard that, but that's why I don't like your posts (or geekbench itself) about specific machines on Geekbench 6, any machine can be so capped or OC'd (assuming a processor that can handle OC) In any case, my i7 doesn't feel sluggish compared to other machines at all, if its capped, the other things about it like amount of RAM and speed of disk more than make up for it, and that brings to mind the second thing I don't like about geekbench and benchmarks in general, they give a false impression of throughput depending on whatever machines they test.
I don't know of a better way to compare processor speeds, so I'll refrain from posting any other benchmarks. They just aren't a reliable enough test to prove anything at all. But then I'm a software guy anyway and my needs pretty much dictate what kind of machine I want for what.
Sorry, when I'm wrong, I admit it, but you saying I'm wrong isn't enough, you have to convince me I'm wrong with evidence. This post is certainly closer, but you can't just call me a liar like that and expect to have a civil discussion where you can convince me how I'm wrong. Calling someone a liar is pretty much ending any civil discussion and convincing. The way you're going, you wont convince me of anything, just keep that in mind.
That's not my intent. I run the numbers about what costs what and computers are a tiny portion of the electricity usage where I live and work, not even enough difference to care about between a high end machine and a power sipper -- it's just tens of dollars a year, where other things take up MUCH more like heating and A/C, traveling, commuting, industrial machines where I work...
Nope, it's VERY relevant.
The smugness is only on your side. (and the intel chips have iGPU as well and how most of the machines I control are configured to use them) As for designed for different purposes, I disagree, both are designed as computer processors for general purpose computers.
So far. We'll see what Apple has up it's sleeve for the Mac Pro -- to service that market they'll have to do something different. As for the M1 Ultra, the 13700K isn't the top end intel chip by any means....
You are comparing numbers from different benchmark versions which are known to be incompatible, and when someone politely points it out your reply is “don’t tell me what to do”? Is this really the standard we want to go by in conversations? Everybody makes honest mistakes and I think we should be able to correct each other in a respectful and polite fashion.
And yes, the stock 10700K is slower than M1 Pro, significantly so.
Yep, don't tell me what to do. As I went on in that message, I said why I don't trust his numbers, and that's the way it is. Just because he can cherry pick a number that "proves" his point doesn't mean squat to me.
No, it isn't, but I expect a lot more for polite conversation than you do apparently.
I never said the 10700K wasn't slower btw (if I did it was a mistake, I was wrong obviously, even I know that.), but significantly slower I would have to disagree on. I was actually surprised that it wasn't significantly slower to me. The only way I can tell the difference is on long running tasks and my i9 10th gen also) desktop is the fastest at encoding and such. It does have discrete video though. I don't even do jobs like that on my iMac or my Studio. I use the iMac because of the monitor on almost all every day type tasks, especially things that need a VM. The iMac has more RAM than anything else I have too.