#apple #MacBookProM5Max #M5Max #MacStudio
The new M5 Max MacBook Pro is more than just a spec bump; it’s a portable AI data center. In this video, we put the maxed-out 128GB M5 Max head-to-head against the ultimate desktop powerhouse, the Mac Studio M3 Ultra.Instead of relying on cloud subscriptions like ChatGPT or Gemini, Apple’s new Fusion architecture and Neural Accelerators allow creators to run massive AI models 100% locally and privately.To see what this 40-Core GPU is really capable of, I ran the “Trinity Stress Test.” We pushed a 70-Billion parameter Llama 3.3 model in LM Studio, generated extreme high-res 2048×2048 images using FLUX.2 [dev] in Draw Things, and rendered 8K RED V-Raptor footage in DaVinci Resolve—all at the exact same time, while running purely on battery power.The results will surprise you.
Image Generation Prompt For FLUX.2
A hyper-realistic scene of a futuristic city at night, with neon lights reflecting off wet streets. A detailed character stands in the foreground wearing a chrome visor, with intricate futuristic armor reflecting the city’s lights. Drones fly in the sky, and a massive digital billboard displays an alien language. The scene is cinematic, with volumetric fog and dynamic lighting.
Chapter Markers
00:00 – The Most Consequential Mac Since M1
01:12 – Your Private AI Workstation
02:38 – Pushing a 70B LLM Locally (LM Studio)
04:28 – Next-Level Image Generation (FLUX.2)
05:52 – Crushing 8K RED Footage (DaVinci Resolve)
06:33 – The Ultimate Multi-Tasking Flex
08:13 – Mind-Blowing Battery Performance
09:40 – Which M5 is Right For You?
This is an absolutely phenomenal review. As a Korean tech enthusiast, I watch a ton of international tech channels, but this is hands down one of the best reviews I’ve ever seen. The way you tested real-world multitasking and provided actual battery consumption data was exactly what I needed. I’m honestly shocked this video doesn't have millions of views yet. I have no doubt your channel is going to blow up. Keep up the amazing work!
Best review I’ve seen – you’ve got the sub looking forward to new content
Nothing big happened.
Problem is video generation. Apple silicon sucks for creating AI video, Nvidia destroys it.
Can we stop with the "This Changes Everything!" claims? My coffee tasted the same as yesterday, my car drove just as well as before, my dog still likes to go on walks, Trump said something stupid again today. See? Almost nothing has changed.
How was the overall temperature of the macbook pro? Was it running really hot?
First of all amazing work and these videos are very helpful. Thanks a lot in advance.
I edit videos from Sony cameras 2-3 layers of 4k slog3 4.2.2 10 bit around 15-20 min videos with some effects, colour grading, masking, noise reduction.
I am torn between 14” m5 max (36gb binned version or I should go with max 40gpu and 64gb ram) how much performance difference I will see? And this future proofing with 64gb realistic ?
I can spend some extra money for 40core and 64gb ram if it really makes sense.
Looking forward for your guidance.
Unfortunately, none of the local LLMs I've tried that will run on the 64GB M5 Pro are nearly as capable as Claude Opus, Gemini 3, etc.
please do video M5 Air 16gb 512gb vs Neo 8gb 512gb
wow the m5 max can do flux2 at the speed of my old 3060ti! cool to see its possible though, just wish this unified memory could be used in same way as nvidia cards :3
I just got my M5 Max Macbook Pro today. I love it
I would love to see you run some local A.I. video tests with higher resolutions and durations on the M5 Max.
You won't know how fast you can drive or how far you can go, until you feel the power.
SSD speed benchmarks? thanks
@5:45. No. You still have more cores on the m3. The difference is the neural arch in the m5 GPU cores.
I have a 14" MacBook Pro M3 with 8GB RAM, but I need more storage and performance pretty soon (for my major and gaming). I'm thinking about buying the new 16" MacBook Pro with the M5 Pro this week since it has a bigger screen, more ram AND storage, and is just a better machine overall. I could sell my current MacBook (it's in perfect condition) for around $1k. My only concern is I've heard rumors about a future “MacBook Ultra” or something coming later this year, and I’d be annoyed if something way better came out right after I buy this. I’ve already waited longer than I wanted to — should I just upgrade now or wait even longer 😭?
can you test wan2.2 or ltx2.2 for ai video generation
Mac Studio M5 Ultra/Max + MacBook Air M5 is the best combo.
sorry m1max for life here.
I have a M1 Pro 16" 1Tb storage and 16gb ram . when I got this I was listening to conten creators saying most people only need the bass model so I got this and I ran out off ram most times I use the m1 . So now I just ordered the M5 max 18core cpu 40core gpu and 128 gb ram 4TB storage so hopefully this will be great . I love the MacBooks the build quality is out off this world
Horses for courses.
So I do not do video as I have my M1 Studio ULTRA for large photo processing. My workflow is, I have all my subject folders on external drive and use Bridge to activate ACR for processing of 212MB files and then go over to PS for usually minimal processing in order to save as PSD or TIFFS and the export of a 1.5MB jpeg for sharing on the net. So far I see no need to upgarde. I do use an M4 15" PBAir when on the road, but mainly for running external backups of images with very minimal on-the-road processing.
For sure is really interesting what they are doing with these refreshes! I’ll wait to replace my MBP M1 Pro with a redesigned MBP
7:45 excellent! I was going to suggest using barcharts as it’s so instant to see which is the faster machine, and here they are
Edit the RED 8K Raw in a 8K timeline with color, effects, and final export time please?
I'm going to buy a Mac Studio M5 Ultra, and I'm going to call it Jarvis.
The problem is open source has fallen way behind now. Also those run times are so slow. If you have serious work, get a real GPU.
We want M5 pro vs M5 max video!! You're very good at this, i'm waiting for this video from your end more than MKBHD as i can learn a lot from it
Took how long to generate Flux 2 images?? My 4090 is under a minute, i think. I’m not 100%. Quick enough to where I don’t think about it too much. Long enough that I use SDXL unless I need/want the flux looks. But the time listed in this review are unusable. For image generation. Fur sure.
I'm convinced Bobby is AI-generated now
At last a real review instead of unpacking and reading apple-provided specs and "opinions", thanks!
As amazing as these improvements are I just cannot justify spending this amount of money when there are much cheaper and faster cloud based solutions. Not to mention state of the art models are either close source (no local inference options whatsoever), or open source with over 600 billion parameters, completely out of reach of current consumer hardware's capability. The only argument for spending this amount of money for local AI is privacy, which is simply not an important enough reason for most people. Anyway that's my take.
It's also important to note that context length is a major bottleneck for inference that is not talked about enough. Just because you have enough ram to load a large language model does not mean it is usable in a meaningful way. Software solutions such as sub-agents can address this issue but they only work in the cloud where there's virtually unlimited memory resources.
Another marjor limitation of AI is knowledge cutoff, which forces you to use third party web scraping tools and integrate them into your AI platform, so you end up having to rely on subscription services anyway.
Also since tech companies are buying up all the ram and gpus to build data centers, we might as well capitalize on their infrastructure.
Going to run your Flux benchmarks on my 60 GPU core M1 Ultra with 128GB ram. Gramps M1 is still around.
same you can do on m4 max … waste time this video
i m getting into india's one of the best colleges for cse based courses , would be doing a lots of online courses and works over ai , coding ; also aiming for summer internships at ivy leagues
for all of this , will m5 max be overkill ? or should go with macbook pro m5 baseline ?
pls give me some clarity 🙂
Alex Ziskind can learn a lot from you! These are actually real-world related tests for end users, also thanks for including Draw Things (Stabel Diffusion)!!
Help, Should i buy the M4 Max ($3,499) or M5 Max ($4,349) both from Costco?
looks like apple won AI. Can shove copilot up microsoft's ass
You are close, but missed completely the why. What happened to create this hardware? Well, Apple was working on a new project, the Vision, and the engineers showed it to Tim Cook and one said a fateful line, "…and of course it doesn't play games." Tim was incensed, the Vision was to be his swan song as he retired. Well, Tim let a lot of engineers go, and Apple did something unusual, they joined the Blender Consortium by dumping a whole bunch of money on a public domain software company to fix Apple's graphics problem. In an alarmingly short time, Blender figured out what Apple had been doing wrong. Apple had hard coded the incorrect algorithms into silicon, and it couldn't handle the new Blender code. So, a new version of Metal was introduced (Metal 4) and Apple went to TSMC and gave them a blank check and said, make this work. M2 was the old design. TSMC started work. M3, too low a yield. M4, they found the interconnect for the Ultra was a bottleneck so they dropped the M4 Ultra. TSMC was also working on a new tech which showed promise for the interconnect, called Fusion, and it worked! Now at the same time, AI started to be a big thing, it used the same graphics algorithms, and Apple found quite by accident that their Blender Consortium code for the Vision also did AI, just like NVIDIA. By accident Apple had figured out a way to do AI better than their biggest enemy, and enemy they are, and do it with almost no power. The race is on to see if TSMC can make enough of these new chips, reliably, because you see, there is no limit to the Ultra design. You can do a double Ultra and more, it's all the same to this new tech. All because some engineer said, "…and of course it doesn't play games."
Great review, really appreciate testing in real world scenarios instead of using benchmarks.
Wow apples gpu is still far behind Nvidia Cuda severely. Granted the 5090 is the size of the MacBook
Thank you for a rare early and legitimate review.
FYI when comparing LLM (large language model) performance, what you really want to look at is not the time the prompt takes to run, but the prompt processing & response tokens per second numbers. Those are really the only things that matter, since the length of the response can vary significantly even running the same prompt.. and ESPECIALLY if you're using a 'thinking' model like qwen3:35b-a3b or glm-4.7-flash.
BTW, go for MoE models (like the ones mentioned above), they run MUCH faster than 'dense' 70b models like featured in this video
I bought M1 years ago and I bought M5 thinking.. they only had the 14 inch, but this was not a M5 pro like my M1. Apple is so sneaky and now they don't want your M5 computer as a return, to buy the M5 pro. They keep doing these things they are going to make their clients pissed off. I would have bought a 16 inch but it only came in 14 inches.
Bro, I just discovered you. You rock and I love your style. Instant sub. You and SPACEDESIGNWAREHOUSE are my favorite tech YouTubers.
Was Llama faster than MLX??
What was your RAM usage like? I feel like no one shares that important info.
ARM-based macs are cool, but the possible directions you're exploring through this YouTube channel is way cooler. I found your real-life scenario test examples immensely helpful as context beyond pure marketing info or tech-spec analysis. Thanks Bob!
Bro, you rock. Instant sub.
I have the M4 Pro and it does pretty good.
Can’t wait for the M5 Ultra. I hear it’s gonna be touchscreen! 😮
Just preordered 48gb ram m5 pro with nano screen ….
32gb is more than enough for next 5 years, which will pay off for itself for most people.
on-pram LLMs, still have ways to be optimized, but with the crazy windows builds [ ram pricing ] , its better to lock in like Tim Cook did, and go banger abit with the RAM pricing .