This man is a godsend for a ML master student like my self. comparison against older but more powerful models is also much appreciated. The most context the better.
CPU wise in multicore 14Core M4 Pro and base M4 Max are almost identical to maxed (16core) M3 Max, GPU Wise M3 Max is still quite more powerful than 14core M4 Pro and a little above the base M4 Max. Maybe if there is a good deal or refurbished M3 Max value is still quite good I'd say.
3:35 sooooo I'm no expert or anything but the relationship between NVMe SSD capacity and speed is actually quite logical when you understand how they work. The main reason is parallelism in the NAND flash chips. When an SSD's capacity increases, this means there are more physical NAND chips inside the drive. Each chip can handle read and write operations independently. A 2TB drive might have 16 NAND chips while a 500GB version of the same model only has 4 chips, allowing the larger drive to handle four times as many simultaneous operations. The controller uses a technique called die interleaving, distributing data operations across multiple chips simultaneously. This creates an effect similar to RAID 0 striping but at the chip level. The controller's firmware optimizes these parallel operations, so when more chips are available, it can achieve significantly higher throughput. This is why larger capacity versions of the same SSD model often advertise much higher speeds. It's not just marketing - it's a fundamental architectural advantage of having more physical NAND chips to work with. This scaling isn't infinite though. At some point, other bottlenecks like the controller's capabilities and the PCIe interface bandwidth become limiting factors. But within the same SSD model line, the larger capacity versions will typically offer better performance due to this increased parallelism.
@@RocketLR I was just about to say raid 0 across multiple chips. But hey. This guy beat me to it with a pretty good explanation. I’m still on my M1 Max with 2tb drive and it does 6gb r/w. Which is pretty impressive for a 4 year old machine.
the video mentions "the bigger the drive the faster they go" but isn't the speed also determined by the processor speed ? So to what extend can we know the faster disk speed is due to the bigger size rather than the faster processor ?
Your channel is absolutely fantastic and informative! I truly enjoy watching your videos. I have a small suggestion: could you consider adding summary tables at the end of your tests? It would make it easier to see the results at a glance and get a clearer picture. Thanks for the great content!
I always love your videos: clear, simple, precise, complete and detailed. The way you explain things is clear for beginners as well as advanced users and this is a gift you have. Thanks for this benchmark, it helped me a lot. Now i am definitely convinced to get a M4 Max... can't wait to try it out after years of Intel!
I was really fed up with other "reviews" on youtube, where someone just read the geekbench score, some specification and run some synthetic test on the machine. I really like your chanel because you test the machines with real life tasks, like Android and visual studio applications. Thats how a programmer chooses a machine, not based on some synthetic tests.😊😊
WebKit release build for Windows via Docker for Windows Containers, Intel i7-14700K, avg. 142W for CPU. Time is: 23m52s. Matched by M4 Pro, beaten by M4 Max. Impressive.
Can't wait to see the video of using a cluster of Mac Minis for Machine Learning. I was wondering if/how that would work as you could potentially get 4 base models for the price of one M4 Pro.
@@nf6sv I'm going to try again and add the comment but without links! maybe it was that. Before, I was saying that I got a 4tb SSD. Originally I went with the WD Black that does +7k speeds but it wasn't going over 5k speeds. Sent it back and got my money back, then I bought a 5k SSD and it ran at the same speed as the expensive one. I bought these: - Crucial P3 Plus 4TB PCIe Gen4 3D NAND NVMe - Qwiizlab Fanless 40Gbps USB4 Thunderbolt External SSD Enclosure Seems that the 7k plus speeds is only if you use the SSD directly into the motherboard. Maybe this can help you to save at least $100 as I did. Btw, I used a Thunderbolt 4 cable from WD, OWC and the cables that came with the enclosure and the OWC and/or cable that came with the enclosure worked perfectly.
Thanks for the review, long awaited. Next time you test them, can you check in the fan noise on the M4 which is talked a lot? I would like to see a regular workflow test like the ones you do. Everyone is doing heavy benchmarks and long exports which is not representive. How about running vms, compiling, writting code, multiple tabs etc... would the fan go crazy on the base M4?
I'm waiting on the m4 max ML video to buy one! I'm studying ML as my research subject in uni, normally I use linux just because of the hassle windows is for new packages and whatnot alongside the capacity for customization. That being said I've seen some other students with mac m1 max's and they generally have a good time with everything. Looking forward to your next video!!
To explain the drive speed, it has to do with the number of nand chips soldered to the board since the controller itself is identical in each machine. The base specs typically only have 1-2 chips to address with reads and writes with all others unpopulated, putting a hard cap on their speed.
Yes, other sources tested this. The 1TB NAND has four 256GB modules running in parallel, while the 512GB drives have two 256GB NAND modules. Hence, the theoretical increase in speed is 2X.
How does the 24GB ram treat you? What do you use it for? Im considering the $2400 MacBook Pro m4 but the 24Gb ram worries me about editing and making videos. Everything else i do is mostly daily use
@@ItsBoNik My work load is not RAM intensive, i did upgrade for future proofing. For video editing or 3d Rendering , go for anything above 24GB , highly recommended, to avoid frame rate drop and future bottlenecks
Amazing video, thanks! Exactly the info I needed. Would be great if you could summarize all results in a table overlayed on the screen after each run. Easier to imbibe such info visually
Bro u are the only one who is programmer who tests these macs so PLEEEEASE WE NEED YOUR OPINION ON NANOTEXTURE FOR CODERS (because it is said that the main objective of programming - text is kinda bad on nanotexture). Is the text code looking bad on it? And to what extent?
Man I’ve literally been waiting for this so I can decide whether to go for the m4 pro or m4 max. As someone who’s a SWE looking to get into AI, your context in these videos is super helpful. Thank you!
I would say now that M4 came out there might be some good deals on the old top specced M3 Max. it's just as good as M4 Pro and base M4 Max on multicore and beats both in GPU, the M4 pro by a quite a bit and the base M4 max by a thread. Talking this as a SWE who's getting into AI as well and running a top M3 Max and also after looking at multiple reviews and comparing a lot of data.
@@brunonascimentofavero6097 ah that’s interesting! Will check it out. But how would you find one that matches the configuration you want? Since Apple doesn’t sell it online anymore
Maybe someone pointed this out before, but if you try to recreate those results with the browser benchmarks it has a big impact whether you run the test for the very first time or not. I ran e.g. the jetstream 2 test on my M4 Pro MB Pro twice (in Chrome). The first time I got a score around 330 the second time it was 495 - so close to Alexs results. I assume Alex did not have to do that because he took over all the data from his previous machine...
About single core tests; You could rig a test with parallels. You can set only one core to parallels virtual machine. Then you can basically run any WIN11, Linux etc. software with a one core.
@AZisk, thanks for video. Please tell a little bit about fan noise. Is it really annoying or insane loud as they say everywhere? Have you really heard them doing usual stuff like web browsing, watching movies, listen a music, etc.. And the same question about a coding tasks: writing code in VS Code, compile anything and so on. The main question: is it really as bad as people say.
I waited so long for this to know if I should return my 16 inch M4 pro, but looks like its okay, but still can't say enough how great that m4 base model is
What about Parallels performance. I am torn between 14" Macbook Pro M4 Pro (14-core) and 13.8" Surface Laptop Elite X. I know M4 Pro is better on paper but what about when in Parallels - people rarely give their VMs 100% of CPU and Ram (not even possible). So how would they both fare in Windows ARM? A real life scenario with Docker, multiple Visual Studios running etc?
If you're outside in the sun on a typical summer day in the Deep South, then you'll find that your laptop begins to malfunction in a matter of minutes. The fans just can't keep it cool enough.
Great video -- very informative. Graphs were helpful. Curious for some follow-up on the battery life low power modes -- in my excitement for the Best Buy deals I got an M4 Pro when I'm a light user by all accounts. Debating taking it back for the M4 base chip (which I have 3 days to pick-up or refund) if that does have a significant bump in battery life. The price difference was negligible for me at their discounts.
Hey Alex. Since you have different Macbooks with M4 pro and M4 MAX chipa with different RAM, have you done any testing to check whats the max LLM parameter size if we run unquantized or 16bit quantized with usuable tokens per seconds. If you can do a video on that, really appreciate that. Thanka a lot. Keep up the good work buddy.
it would be nice if u could integrate the m1 mba again. that gives you a create idea how far the m series macbook have come in a really short amount of time
Very interesting test, very well executed. I would like to see some comparisons between different 128GB RAM models. Like a M3 Max vs a M4 Max. 14 inch vs. 16 inch would also be interesting. Especially for AI and ML tasks that actually use all this RAM and demand huge performance.
Hey Alex! This is perfect timing, as I'm having some heat issues with my Asus S14 UX5406 this week. I’ve decided to return it, and despite the Apple EU tax(ironically, Apple typically owes EU taxes), I’m considering getting a MacBook Pro! I’m still figuring out which variant to go for, though. Thanks for your help!
Not to get all political, but this is an example of the misunderstanding of how a Tariff works. Regardless of the tariff (aka: Tax) being imposed directly to the consumer, or to the country/company of origin…YOU are the one who will be paying the tax in the end. If the country of origin pays the tariff, then that cuts heavily into their bottom line. The only option that they have is to pass it on to you, the consumer.
@@scottkuzma125 I removed the Tax as in Europe, we include the tax on the prices of the goods. It worked out to about $340. I also looked at prices in some states in the US, and some states have a high tax rate, though.
Would you recommend the top MacBook m4 pro for 2399? Only thing that worries me about that one is 24gb memory. I do video editing with videos from my phone and drone. Video editing is the most i do, everything else is mostly daily use.
Can't wait for the machine learning video, Although the max is very powerful, try to include the other cheaper models (m4 and m4 pro) and give their verdicts as well on like capabilities to do machine learning .. since i am a student and don't have the budget to buy a max chip , but still want to learn ML and maybe run smaller deep learning and small parameters model. Please try to include it as well
I appreciate your work man. may I request, can you do a benchmark comparison on compiling/build/running processes inside docker using both arm build and x86 architecture version?
Thank you for your video. It’s very helpful. You had a video about running Windows 11 with Parallels. I’m considering to buy the 14inch M4 pro. At the moment I’m thinking to upgrade the CPU cores to make Windows with Parallel to run smoother. I will not be using the VM to perform heavy tasks. Just to allow me options to run some Windows specific apps with the same hardware. Hope you can give your opinion or make a new video about running Parallels with M4 Pro. Thank you
The iPhone and the Mac range of products share the same type of Nand flash, the MacBook Air has one slot, MacBook Pro 13 two slots and MacBook 14/16 has 4 slots. Depending on what drive size configuration you go with is how they populate them.
2 หลายเดือนก่อน
Please also show us the performance improvement using Android Studio and a large Android project. The difference between M1-Max and M4-Pro is particularly interesting.
I would love to see a color measurement between the older MacBook Pro and the new ones. Apple changed the display technology slightly on these, and didn’t bother to advertise it. They now use quantum dot technology. The new MacBook Pro should be capable of brighter colors, particularly in the greens and blues, as well as more accurate color steps between the three primaries. That should overall increase the color gambit and total colors achievable.
Any ideas for a 16” for starting to code and some gaming. Can’t decide between the 16” M4 pro 48/1tb and the 16” M3 max 36/1tb. There’s about $100 difference in price.
Hi Alex, given that you have several Macbooks. may I please ask you to check how different is the color uniformity across the screen for the models you have, do you notice any having a warmer screen on one side, or with blotches or redness, or anything like that, while looking at white content?
Like you I had the pimped out M2 Max and upgraded to the M4 Max (both 8TB with just a few hundred GB free). I've got 78 ollama models and a couple of dozen in LM Studio and raw HF models too. It's about 33-50% faster than my M2 but the 128GB over 96GB of the M2 means I can now run slightly larger or less quantised models. Keep making the videos, ping me if you want / need any ideas, frameworks, alternative numbers etc.
@@AZisk Most of it is taken up with my astrophotography (my hobby), a whole night of 30 second exposures on a 50MP camera that then need to be processed by a superb app called PixInsight that NASA use. It creates numerous new versions of each one, aligned, normalised and then subtracts the dark noise, and "light noise" for each image before stacking them. A single evening can generate a good 200+ GB of sub-images before the final image. Then new MBP is a superb machine for processing astrophotography, and of course for work in AI 🙂
Hi! I'm currently deciding between a few MacBook models and was hoping you could share your expertise. I'm planning to apply for a data science program in 2026, so I need a laptop that will work well for studying and practicing in this field The options I'm considering are: Macbook Pro M3 (18GB RAM) Macbook Pro M4 (24GB RAM) Macbook Pro M4 Pro (24GB RAM) Which one would you recommend for data science studies, considering performance and long-term use? Thanks in advance for your advice
Thanks for the great video. Can you also comment on the latest display upgrade information floating around where the new M4 MBP have a Quantum Dot display. Are you able to see less visible "ghosting" compared to the M2 Max?
Could fan behavior be affected by the High Power mode? It would be nice if “I don’t care about fan noise” = High Power mode and “I would rather not hear the fan as much” = Automatic/Low Power mode.
Fantastic test and presentation as always! - Now I need to decide if I can live with a big honking 16” laptop. I don’t travel as much as I used to, but I’m still kind of struggling with the idea of that big a jump in size and weight. (I’m going for either the same setup as you or that with 8TB. My use case will be doing actual work with local LLMs. I’m coming from a used loaded M1 Pro, haven’t nearly used its 8TB, but the fact it’s non-upgradable gives me pause. - OTOH, I could buy a pretty fast 4TB external for less than the incremental cost…)
Thank you for being a mac "reviewer" that doesn't only focus on performance in video editing and content creation 🙏🏻
That doesn't only focus on Geekbench and claims it is game over for AMD and Intel... lol
Yes! All videos are about video editing and all of that. Its really valuable to find these videos
@@ecko10fullAlex is just about the only reviewer on here who looks at development workflows. His Snapdragon X reviews are great.
@@TechGameDev What you said x 10000 ! You are so right! Geekbench is almost like userbenchmark web site.
@@dennisvillamor2806 yes cant wait for the machine learning test
The guy literally set up a laptop testing facility in his bedroom.
Not exactly his bedroom, but he does have to sleep there sometimes 😂
@@toadlguy🤣
This man is a godsend for a ML master student like my self. comparison against older but more powerful models is also much appreciated. The most context the better.
CPU wise in multicore 14Core M4 Pro and base M4 Max are almost identical to maxed (16core) M3 Max, GPU Wise M3 Max is still quite more powerful than 14core M4 Pro and a little above the base M4 Max. Maybe if there is a good deal or refurbished M3 Max value is still quite good I'd say.
ML inference is pretty good on these M4 machines but model training or fine-tuning still needs Nvidia horsepower.
3:35 sooooo I'm no expert or anything but the relationship between NVMe SSD capacity and speed is actually quite logical when you understand how they work.
The main reason is parallelism in the NAND flash chips. When an SSD's capacity increases, this means there are more physical NAND chips inside the drive. Each chip can handle read and write operations independently. A 2TB drive might have 16 NAND chips while a 500GB version of the same model only has 4 chips, allowing the larger drive to handle four times as many simultaneous operations.
The controller uses a technique called die interleaving, distributing data operations across multiple chips simultaneously. This creates an effect similar to RAID 0 striping but at the chip level. The controller's firmware optimizes these parallel operations, so when more chips are available, it can achieve significantly higher throughput.
This is why larger capacity versions of the same SSD model often advertise much higher speeds. It's not just marketing - it's a fundamental architectural advantage of having more physical NAND chips to work with.
This scaling isn't infinite though. At some point, other bottlenecks like the controller's capabilities and the PCIe interface bandwidth become limiting factors. But within the same SSD model line, the larger capacity versions will typically offer better performance due to this increased parallelism.
nice explanation
@@RocketLR I was just about to say raid 0 across multiple chips. But hey. This guy beat me to it with a pretty good explanation. I’m still on my M1 Max with 2tb drive and it does 6gb r/w. Which is pretty impressive for a 4 year old machine.
“I’m no expert,” and gives an expert-level explanation. Thanks G.
the video mentions "the bigger the drive the faster they go" but isn't the speed also determined by the processor speed ? So to what extend can we know the faster disk speed is due to the bigger size rather than the faster processor ?
The King with the best Testing Videos ever
Your channel is absolutely fantastic and informative! I truly enjoy watching your videos. I have a small suggestion: could you consider adding summary tables at the end of your tests? It would make it easier to see the results at a glance and get a clearer picture. Thanks for the great content!
Thanks!
thank you!
Thanks
thank you so much!
😁 "I forgot which is which, who cares, we'll find out!" 😂🤣😂 Like a boss!
I been checking every hour for this to drop, finally. Alex the GOAT 🐐
At 66Gbps, it would take only two minutes to clone the entire 1TB drive from one machine to the other. That's insane.
I always love your videos: clear, simple, precise, complete and detailed. The way you explain things is clear for beginners as well as advanced users and this is a gift you have. Thanks for this benchmark, it helped me a lot. Now i am definitely convinced to get a M4 Max... can't wait to try it out after years of Intel!
I was really fed up with other "reviews" on youtube, where someone just read the geekbench score, some specification and run some synthetic test on the machine. I really like your chanel because you test the machines with real life tasks, like Android and visual studio applications. Thats how a programmer chooses a machine, not based on some synthetic tests.😊😊
You have quickly become one of my favorite people to watch for insightful reviews. Great format and excellent attention to detail. Thank you 👍
I feel sad for not mentioning the Legendary Schwarzenegger 😔
2.0
@@AZisk I thought it was a 3.0; I don't remember it being so aggrandized
@@aeonlancer Rather 4.0 for 4 x fingers ;)
WebKit release build for Windows via Docker for Windows Containers, Intel i7-14700K, avg. 142W for CPU. Time is: 23m52s. Matched by M4 Pro, beaten by M4 Max. Impressive.
Can't wait to see the video of using a cluster of Mac Minis for Machine Learning. I was wondering if/how that would work as you could potentially get 4 base models for the price of one M4 Pro.
3:46 I got the M4 Max but with 1tb and it does run at that speed. +7k and +6k, so I believe it does change when you choose the +1tb SSD.
Thank you for the info. I've decided that my next laptop will be a M4 Max /1tb. I will use some external ssd for archival and projects.
@@nf6sv great! I bought 4tb SSD with T4 connection and it runs at 5k read and write! I will check which brands and let you know.
@ bro I added another reply but it’s gone :(
@@nf6sv I'm going to try again and add the comment but without links! maybe it was that.
Before, I was saying that I got a 4tb SSD. Originally I went with the WD Black that does +7k speeds but it wasn't going over 5k speeds. Sent it back and got my money back, then I bought a 5k SSD and it ran at the same speed as the expensive one.
I bought these:
- Crucial P3 Plus 4TB PCIe Gen4 3D NAND NVMe
- Qwiizlab Fanless 40Gbps USB4 Thunderbolt External SSD Enclosure
Seems that the 7k plus speeds is only if you use the SSD directly into the motherboard.
Maybe this can help you to save at least $100 as I did.
Btw, I used a Thunderbolt 4 cable from WD, OWC and the cables that came with the enclosure and the OWC and/or cable that came with the enclosure worked perfectly.
amazing engineering... Thanks Alex to perform all these tests. KR..
Thanks for the review, long awaited. Next time you test them, can you check in the fan noise on the M4 which is talked a lot? I would like to see a regular workflow test like the ones you do. Everyone is doing heavy benchmarks and long exports which is not representive. How about running vms, compiling, writting code, multiple tabs etc... would the fan go crazy on the base M4?
the finger device is awesome!
I'm waiting on the m4 max ML video to buy one!
I'm studying ML as my research subject in uni, normally I use linux just because of the hassle windows is for new packages and whatnot alongside the capacity for customization. That being said I've seen some other students with mac m1 max's and they generally have a good time with everything. Looking forward to your next video!!
Thank you so much for the review!
The code benchmarks are what I was looking for
The best reviewer for devs! Gotta love the thumb machine xD
Swartzen-Finger!
Finally, finally it's here. Been waiting for this over a month.
Thank you, Alex!
it’s not over
To explain the drive speed, it has to do with the number of nand chips soldered to the board since the controller itself is identical in each machine. The base specs typically only have 1-2 chips to address with reads and writes with all others unpopulated, putting a hard cap on their speed.
Yes, other sources tested this. The 1TB NAND has four 256GB modules running in parallel, while the 512GB drives have two 256GB NAND modules. Hence, the theoretical increase in speed is 2X.
What about the 256gb model? @@andyH_England
@@andyH_England Do you mind sharing those other sources? I tried to find but could not.
@@TheAyushSomani same here
Mate very impressive ,detailed, different kinds of tests 👏
Dude you do the best reviews. Love your channel.
Thanks dude.🙏
I will watch the video but first I need to like it, that’s how consistently good you are 🎉
I upgraded base M4 Pro to 24GB Ram and 1TB SSD - Same speeds on 1 TB SSD as 512GB ( R -3300MB/s , W - 2900MB/s)
How does the 24GB ram treat you? What do you use it for? Im considering the $2400 MacBook Pro m4 but the 24Gb ram worries me about editing and making videos. Everything else i do is mostly daily use
@@ItsBoNik My work load is not RAM intensive, i did upgrade for future proofing.
For video editing or 3d Rendering , go for anything above 24GB , highly recommended, to avoid frame rate drop and future bottlenecks
M4 Stravaganzza!!!! Holly Canolly Alex! Good job!
Amazing video, thanks! Exactly the info I needed. Would be great if you could summarize all results in a table overlayed on the screen after each run. Easier to imbibe such info visually
your youtube award on the wall is unaligned, it drives me crazy 😆
This is how you use a skill for the exact pupose. Great job with the setup and the video Alex
Awesome content as usual Alex! Can we expect an ML review soon? Would really like to know what memory config to get the M4 Max in.
You should call your starting gadget the “synchronator” 😂😂
Bro u are the only one who is programmer who tests these macs so PLEEEEASE WE NEED YOUR OPINION ON NANOTEXTURE FOR CODERS (because it is said that the main objective of programming - text is kinda bad on nanotexture). Is the text code looking bad on it? And to what extent?
Man I’ve literally been waiting for this so I can decide whether to go for the m4 pro or m4 max. As someone who’s a SWE looking to get into AI, your context in these videos is super helpful. Thank you!
Whichever you go. You're gonna love it for sure 🎉
I would say now that M4 came out there might be some good deals on the old top specced M3 Max. it's just as good as M4 Pro and base M4 Max on multicore and beats both in GPU, the M4 pro by a quite a bit and the base M4 max by a thread. Talking this as a SWE who's getting into AI as well and running a top M3 Max and also after looking at multiple reviews and comparing a lot of data.
@@brunonascimentofavero6097 ah that’s interesting! Will check it out. But how would you find one that matches the configuration you want? Since Apple doesn’t sell it online anymore
Maybe someone pointed this out before, but if you try to recreate those results with the browser benchmarks it has a big impact whether you run the test for the very first time or not. I ran e.g. the jetstream 2 test on my M4 Pro MB Pro twice (in Chrome). The first time I got a score around 330 the second time it was 495 - so close to Alexs results. I assume Alex did not have to do that because he took over all the data from his previous machine...
About single core tests; You could rig a test with parallels. You can set only one core to parallels virtual machine. Then you can basically run any WIN11, Linux etc. software with a one core.
@AZisk, thanks for video. Please tell a little bit about fan noise. Is it really annoying or insane loud as they say everywhere? Have you really heard them doing usual stuff like web browsing, watching movies, listen a music, etc.. And the same question about a coding tasks: writing code in VS Code, compile anything and so on. The main question: is it really as bad as people say.
This channel is a gem, diamon, gold and BTC
most awaited vid of the m4s🔥
I waited so long for this to know if I should return my 16 inch M4 pro, but looks like its okay, but still can't say enough how great that m4 base model is
M4 MAX is the new KING of laptop processors!
Just in the right time! I am thinking of buying one this end of the month!
I wish you added some Flutter tests as well, but overall thanks for the video!
What about Parallels performance. I am torn between 14" Macbook Pro M4 Pro (14-core) and 13.8" Surface Laptop Elite X. I know M4 Pro is better on paper but what about when in Parallels - people rarely give their VMs 100% of CPU and Ram (not even possible). So how would they both fare in Windows ARM? A real life scenario with Docker, multiple Visual Studios running etc?
If you're outside in the sun on a typical summer day in the Deep South, then you'll find that your laptop begins to malfunction in a matter of minutes. The fans just can't keep it cool enough.
Great video -- very informative. Graphs were helpful.
Curious for some follow-up on the battery life low power modes -- in my excitement for the Best Buy deals I got an M4 Pro when I'm a light user by all accounts. Debating taking it back for the M4 base chip (which I have 3 days to pick-up or refund) if that does have a significant bump in battery life. The price difference was negligible for me at their discounts.
Thanks for the WebKit build.
Also, call it The Arnold.
Hey Alex. Since you have different Macbooks with M4 pro and M4 MAX chipa with different RAM, have you done any testing to check whats the max LLM parameter size if we run unquantized or 16bit quantized with usuable tokens per seconds. If you can do a video on that, really appreciate that. Thanka a lot. Keep up the good work buddy.
it would be nice if u could integrate the m1 mba again. that gives you a create idea how far the m series macbook have come in a really short amount of time
Very interesting test, very well executed. I would like to see some comparisons between different 128GB RAM models.
Like a M3 Max vs a M4 Max. 14 inch vs. 16 inch would also be interesting.
Especially for AI and ML tasks that actually use all this RAM and demand huge performance.
First time seeing the channel, would love to see machine names next to numbers for easier reference, I'm used to charts but that's probably overkill
You're the best at doing this stuff... i thank you sooo much.
Thank you so much!!!! Someone finally comparing the M2Max!!
thank for the great review.
Love from Netherlands 🇳🇱
Hey Alex! This is perfect timing, as I'm having some heat issues with my Asus S14 UX5406 this week. I’ve decided to return it, and despite the Apple EU tax(ironically, Apple typically owes EU taxes), I’m considering getting a MacBook Pro! I’m still figuring out which variant to go for, though. Thanks for your help!
np. good luck.
Which config are you considering? I am kinda in the same shoes, tho i returned my s14 because i couldnt get on with windows again
Not to get all political, but this is an example of the misunderstanding of how a Tariff works.
Regardless of the tariff (aka: Tax) being imposed directly to the consumer, or to the country/company of origin…YOU are the one who will be paying the tax in the end.
If the country of origin pays the tariff, then that cuts heavily into their bottom line. The only option that they have is to pass it on to you, the consumer.
@@scottkuzma125 I removed the Tax as in Europe, we include the tax on the prices of the goods. It worked out to about $340. I also looked at prices in some states in the US, and some states have a high tax rate, though.
@nniklask Hi I decided to go for 16" Max 1 40 GPU Core, and 1 SSD, and 128 GB.
Dont flag the Arnold !!! 😂
Good review - Thanks !!
Great channel. You can take alcohol and wife the macs down. I even use it on the beyboards as well as the glass. They look like brand new when done.
Would you recommend the top MacBook m4 pro for 2399? Only thing that worries me about that one is 24gb memory. I do video editing with videos from my phone and drone. Video editing is the most i do, everything else is mostly daily use.
I enjoy watching your videos!!
thanks for the video! Awesome results and tests.
Can't wait for the machine learning video,
Although the max is very powerful, try to include the other cheaper models (m4 and m4 pro) and give their verdicts as well on like capabilities to do machine learning .. since i am a student and don't have the budget to buy a max chip , but still want to learn ML and maybe run smaller deep learning and small parameters model. Please try to include it as well
I appreciate your work man. may I request, can you do a benchmark comparison on compiling/build/running processes inside docker using both arm build and x86 architecture version?
Thank you for your video. It’s very helpful. You had a video about running Windows 11 with Parallels. I’m considering to buy the 14inch M4 pro. At the moment I’m thinking to upgrade the CPU cores to make Windows with Parallel to run smoother. I will not be using the VM to perform heavy tasks. Just to allow me options to run some Windows specific apps with the same hardware. Hope you can give your opinion or make a new video about running Parallels with M4 Pro. Thank you
Freaking hell finally!! :D been waiting for this video before choosing which to go for! Gracias Alex!
The iPhone and the Mac range of products share the same type of Nand flash, the MacBook Air has one slot, MacBook Pro 13 two slots and MacBook 14/16 has 4 slots. Depending on what drive size configuration you go with is how they populate them.
Please also show us the performance improvement using Android Studio and a large Android project. The difference between M1-Max and M4-Pro is particularly interesting.
Thanks for ollama test!!!
On your upcoming ML reviews - please do some SDXL tests as well - torn between getting a M4 Max vs a x86Nvidia 50x based system 😭
Great job. Itwould be easier to follow if you had the machines in order of speed.
I would love to see a color measurement between the older MacBook Pro and the new ones. Apple changed the display technology slightly on these, and didn’t bother to advertise it. They now use quantum dot technology. The new MacBook Pro should be capable of brighter colors, particularly in the greens and blues, as well as more accurate color steps between the three primaries. That should overall increase the color gambit and total colors achievable.
Crazy performance 👏
14" m4 pro 12c vs 14c, heat/performance trade offs please 🥺
Any ideas for a 16” for starting to code and some gaming. Can’t decide between the 16” M4 pro 48/1tb and the 16” M3 max 36/1tb. There’s about $100 difference in price.
Hi Alex, given that you have several Macbooks. may I please ask you to check how different is the color uniformity across the screen for the models you have, do you notice any having a warmer screen on one side, or with blotches or redness, or anything like that, while looking at white content?
Thanks for all this work, but I really wish someone would do a test like Topaz that uses the neural engine, to compare the machines.
Just tried it, and yep, it does use the ANE - great to see that.
Like you I had the pimped out M2 Max and upgraded to the M4 Max (both 8TB with just a few hundred GB free). I've got 78 ollama models and a couple of dozen in LM Studio and raw HF models too. It's about 33-50% faster than my M2 but the 128GB over 96GB of the M2 means I can now run slightly larger or less quantised models.
Keep making the videos, ping me if you want / need any ideas, frameworks, alternative numbers etc.
wow, you really need the storage!
@@AZisk Most of it is taken up with my astrophotography (my hobby), a whole night of 30 second exposures on a 50MP camera that then need to be processed by a superb app called PixInsight that NASA use. It creates numerous new versions of each one, aligned, normalised and then subtracts the dark noise, and "light noise" for each image before stacking them. A single evening can generate a good 200+ GB of sub-images before the final image. Then new MBP is a superb machine for processing astrophotography, and of course for work in AI 🙂
could you also mention the specs of each of the laptop aswell?. I mean the cpu and gpu cores? You only listed the ram and storage space.
good call. i added it to the description
@@AZisk thank you so much. It’ll make my shopping decisions easy ♥️
Hi! I'm currently deciding between a few MacBook models and was hoping you could share your expertise. I'm planning to apply for a data science program in 2026, so I need a laptop that will work well for studying and practicing in this field
The options I'm considering are:
Macbook Pro M3 (18GB RAM)
Macbook Pro M4 (24GB RAM)
Macbook Pro M4 Pro (24GB RAM)
Which one would you recommend for data science studies, considering performance and long-term use? Thanks in advance for your advice
M4 Max 64GB or 128GB for running LLM locally?
Awesome channel!!
NANDs are in RAID configuration: that might account to the increased speed in higher storage ssds
Will you compare the mac mini with the m4 macbooks? Just to see how it compares in these tests
Please do a stable diffusion test with either automatic1111 or comfyui between the macs and also a PC
Thanks for the great video. Can you also comment on the latest display upgrade information floating around where the new M4 MBP have a Quantum Dot display. Are you able to see less visible "ghosting" compared to the M2 Max?
I suspect that the reflections from such a bright flashlight so close to the screen inflated your luminosity numbers.
4:17 it's 1000 nits base max brightness and up to 1600 nits.
This guy returns are crazy
I was waiting for this exact video, Thanks man
Soon you will cross the Million mark InshaAllah
Love and thanks from Pakistan 😍
Could fan behavior be affected by the High Power mode? It would be nice if “I don’t care about fan noise” = High Power mode and “I would rather not hear the fan as much” = Automatic/Low Power mode.
The one with one hot spot is M4 (has only one fan) the other one is M4 Pro (got two fans).
Love all your videos.
i hope you know your content is deeply appreciated. thank you for taking the time.
Thanks for the kind words!
You should run a benchmark against the iPad pro m4 as well. Geekbench AI scored slightly higher on the iPad pro m4 than the iMac m4 base.
I would like too how these machines perform when an external monitor is attached and how support for 120hz is working.
Fantastic test and presentation as always!
- Now I need to decide if I can live with a big honking 16” laptop. I don’t travel as much as I used to, but I’m still kind of struggling with the idea of that big a jump in size and weight.
(I’m going for either the same setup as you or that with 8TB. My use case will be doing actual work with local LLMs. I’m coming from a used loaded M1 Pro, haven’t nearly used its 8TB, but the fact it’s non-upgradable gives me pause. - OTOH, I could buy a pretty fast 4TB external for less than the incremental cost…)
Can you exp them together with T5 and see how big a model they can handle?
🤔 i might just do that