Lately I was doing some test with an old project that included a scrollable window with +800 different controles (yes, that’s overkill).
Opening that window on my MBP 16 i7 takes approx. 4 seconds before it is shown.
Trying the exact same code in my MBP M1 took … well instantaneous (not even a second). It is a HUGE difference.
It is also noticeable with smaller projects I have, but that project really showed there is an incredible huge difference in displaying things (at least with Xojo apps).
Now there are two possible reasons:
The graphic drivers on the MBP M1 are very optimised for the M1
Or Apple is doing this on purpose (OS software wise) to make the M1 look like a better performance.
I read that apple engineers when they debugged the first M1 prototypes thought the battery indicator was not refreshed correctly because it didn’t move…
There isn’t much (other than the screen and SSD size) where my little Macbook Air M1 could not compete with my 2019 MBP 16-inch. A bit unfair though: the Intel excels in fan noises and I’m sure I will miss the heat of the device on cold winter days. Impressive device. I’m sure that in some areas (final cut pro like stuff, compiling, etc.) the MBP is a bit faster, but that’s almost not noticeable in daily use if you are like me and not doing much graph/video stuff.
To be clear, it isn’t the Xojo GUI that’s faster. It’s a big project compiled with Xojo that’s a lot faster showing/opening a window with a lot of controls. The difference is really huge compared to a Intel compiled version.
I still think Apple has optimised ARM stuff in macOS and ‘crippled’ Intel stuff. I have no other idea why a faster i9 MBP is performing worse regarding GUI operations.
Hi, I have now a MBP with a M1 chip,and a Mac mini with M1 but only 8GB Ram. The mini is now used by my daughter to finish her bachelor thesis. She uses a lot of Adobe, and is very impressed about loading speed of her projects, and the speed of working with a lot of graphics.
But , how many of the customers will switch to M1 ?
You could rule out GPU or graphics drivers as the reason by starting up in Safe Mode on both systems to compare. I suspect it may actually be the M1’s superior ability to burst single thread workloads during window init. You could also profile the code to see if something jumps out at you disproportionately between systems. Might also be a difference in the order of control initialization causing one window to draw earlier, or something that triggers extra processing during open events on Intel.
I am ready to upgrade my main work laptop as soon as the 16” MBP’s are ready to be sold. Thought for sure i’d have spent the money after WWDC this year, alas not ready. I’m not in a place where I need a testing mini even if I never use it for anything else, so I’m just waiting. About 90 seconds after they announce it I’ll be ordering one though.
It’s not - I have an i9 MPB and the results of a large XCode project takes low seconds difference running under 10.13 versus Big Sur on that same machine.
That same project on my 8GB M1 Mini is almost 5 minutes faster (7:33~ over 12:20~ for Intel).
Oh, and my MBP fans go into vacuum cleaner mode during the compile and the M1 is silent.
When the M1 came out, I wanted to understand how can they be so much faster, compared to 2019 Intel chips.
This is what I recall.
RISC has the potential to be faster than CISC, we saw this back in the days of the PowerPC. The switch to Intel was because Intel become more efficient.
Manufacturing size. All TSMC do is fabricate chips, this single track focus has allowed them to do much much better than Intel at shrinking die size. Not only does this alone make the chip faster, but also generates less heat. Rumor has it that since the M1, Intel have been working with TSMC to make Intel chips at smaller nano-meter than Intel could.
On chip memory, reduces path distance and Apple have used faster memory.
On chip GPU and unified memory reduces memory exchange, thus improving initial speed with pushing data to the GPU as it no longer needs to be “pushed”.
Faster SSDs and no more T2 Chip. The T2 chip was problematic, it was the cause of many issues during the 2017 ~ 2018 era of MacBook Pros. I read somewhere that it was actually slower than a dedicated SSD controller, to which Apple never improved speed in the last 4 years.
Improved read-ahead where the CPU looks at what it needs to do and makes better assumptions in shortcuts.
Better background task management, background tasks do take longer on M1, but the way how Apple schedules them, leaves cores available for foreground tasks, so it feels smoother. Scheduling on Intel spreads out background tasks across all cores.
If Apple had wanted to do so, they could have upgraded their Intel machines by using faster SSDs, faster memory, improved controller chips, AMD x86_64 processors (which are more efficient than Intel) and NVIDIA GPUs. Heck at this point they could even put in last years Intel chips and gain upto a 5% improvement over the two year old chips they still selling.
But they don’t want to do that, because in doing so, it would reduce ROI on Intel machines that they’re going to replace next year.
This is the beef I have with Apple AJ, the focus is on profit, they have enough cash to improve their Intel based offerings, especially while Pros wait for more capable machines, but they won’t.
There’s rumors that Apple’s having trouble with the next gen chips, and they may not be available until 2022 or 2023. Bandwidth is a big sticking point and is tangible performance improvement. But at this point, just rumors.
I am talking just about a compiled app were a window with a lot of controls is openend. Not about it’s main performance or speed. That’s still faster in my MBP i9.
If I remember correctly the M1 can look ahead to decode the next 8 commands, because they all have same length (4 bytes).
The intel one looks 4 commands ahead (one byte each), but has to throw away those decoder results if he sees a command is longer. Intel commands are between 1 and 15 bytes long!
The intel decoders more likely throw away result than they may use it.
It makes perfect sense. Loading is quicker, memory is quicker and thanks to the GPU being on chip and utilizing the same CPU memory, it means that view layers (one for each view) are quicker in getting to the GPU and appearing on screen.
That sounds familiar, we must have read the same article. This can be done with Intel/AMD, but it will cost them backwards compatibility, which is one of the advantages of the current x86 design. However Intel/AMD will adopt other advantages, which will improve the speed of x86 processors.
I fully expect Intel / AMD to surpass Apple again in the future, as this roundabout continues going round…
The problem with Intel/AMD is the CISC architecture and TDP (thermal design power). They have leveled off in the gains that they can make because of the thermal limits. The CISC architecture itself kind of handcuffs their efforts to make progress because complex instructions require a lot of circuitry and that in itself creates more heat. The M1 has far more headroom in TDP and therefore there is a lot more room for improvement. Unified memory architecture is a game changer by eliminating many bottlenecks thus increasing efficiency resulting in a lower TDP and faster execution times. When one sees huge cooling units strapped to the CPU to prevent thermal meltdown, then the dance with heat dissipation becomes the limiting factor which plagues Intel/AMD.
15 years ago Apple abandoned their own RISC chips, for the very same reasons you are describing now.
Which is what I mean by it going around and around, maybe in 15 years they’ll be back on x86 because they’ve once again reached limitations that they can’t seem to overcome. Or realize that they need x86 compatibility to increase their market share (again).
Worth watching this ARM presentation to Apple if you haven’t watched it before - these guys were way ahead of the curve 30 years ago. https://www.youtube.com/watch?v=ZV1NdS_w4As
Because AMD is also switching to the same RISC instruction set No point in switching to another manufacturer of another chip that you’re just going to have to switch away in another year when they stop bothering to improve them
Copying Files & Folders in MemoryStick is particularly slow. Better use AirDrop (*) or external HDD / SSD.
(*) Beware of AirDrop: it is not always reliable (some Files / Folders may not be copied if sent by groups). I even had a case where a single file copy was impossible; I never knew why.
No point in switching to another manufacturer of another chip that you’re just going to have to switch away in another year when they stop bothering to improve them