There is Something Apple Didn't Tell Us About M5

The base model M5 MacBook Pro is here and it is on fire. We just have to get serious right off the bat here. I'm going crazy over here at the space warehouse. I don't want to lose your attention early. The M5 base model MacBook Pro beats the M4 Max MacBook Pro so far at two things that I found, one more significant than the other. First up, Final Cut Pro magnetic mask. That's the tool where you can have your computer cut you out of the background so you can do things like put words behind you in the room without having to be shooting on a green screen. I actually use this tool a lot. I set up a 44 second section of my last video and had these two computers fight it out. And I feel like I need to reiterate this is the base model M5 MacBook Pro. Obviously, it's the only one that's out. This is an M4 Max. So, this thing has 10 performance CPU cores. This thing has four. This has 32 GPU cores. This has 10. This should have something like three times the overall total processing power as this computer does, even though it's a one generation behind. This is the Max model. set up the magnetic mask, got them both going, and wowee, the new M5 can do this faster. The M4 Max is pulling a total of about 11 to 12 watts, and only a little more than three, three to three and a half of those is the NPU. The M5 is instead using between 16 to 20 watts, and 3 1/2 to four of those watts are in the NPU. The M5 did this in 52 seconds. The M4 Max did this in 55 seconds. Has there ever been a thing where the base model next generation Apple silicon can beat the max model of the previous generation? I got this M5 MacBook Pro yesterday and started testing it out against everything. And it was doing the usual stuff you'd expect. It beat the M4 Pro base model by about 20% in benchmarks, exporting video, all that. And I'm going to do a direct comparison video after this. M5 base model, M4 base model. But this revelation was just too much, too fast. I have to get it out with the machine learning stuff. This new computer is like a 300% leap. I'm confident that there is something Apple's not telling us about this chip. Well, I think they did say it by mentioning the neural accelerators on the GPU, but I don't think they made a big enough deal about it. This is the base model M5 and it's not just the magnetic mass. This thing also beats the M4 Max at image generation. And actually, this is the new M5 iPad Pro, and it beats this thing at image generation when you're using the image playground. The iPad Pro will produce images faster than the M4 Max does on the image playground. And that's such a specific skill and one I really don't care about at all. But that's not the point even a little bit. The M5 base model beats the M4 Max. Not just the M4 Pro, but the Max at this borderline useless skill of putting together these little cartoony pictures of me and my friends on the image playground app. I can describe a thing and this base model M5 MacBook Pro will spit out an image faster than this $3,500 computer that's only a year old. I have a prediction. I think Apple GPT is almost ready. the fact that I can talk to my Alexa now conversationally. If you're unaware, Amazon released its new Alexa Plus, but for some reason only on the devices that have screens. So now there's Amazon's own version of Chat GPT just living in here that I sometimes just have a little conversation with when I'm at my desk. I think Apple's basically done with theirs finally. But I think they wanted to do something no one else is doing yet, and they want it to run on your computer instead of in a data center. And I think whatever is happening here with this performance is the key to making that possible. Someone made some sort of a breakthrough and then just didn't mention it. How is this the least announced chip release of recent memory? Anyway, these two things seem to be majority NPU stuff doing things. So, this sent me down a rabbit hole of ondevice image generation. And I ended up at an app called Draw Things. It's on the app store. I downloaded Quen Image 1.0. I don't know what I'm doing. This is pretty much like how on ChatPT you can just describe something and ask it to make a picture of it. And in that case, a bunch of GPUs in some data center somewhere will spin up and it'll create an image for you, send it back to your computer. Except this does this on device and it does it a lot slower. The Mac is doing all of the data cent's work. Like you can completely turn off your Wi-Fi, put your computer on airplane mode, and it'll keep spitting out images. First, I put the M5 MacBook Pro against the base model M4 MacBook Pro and it doesn't just do the like expected 20% better. It's not even 50% better. This thing can create an image in like a third of the time. It should be noted that this process is not super fast. It's not like the image playground app. This computer takes 8 minutes to make one picture, the M5. But the M4 MacBook Pro took over 22 minutes, almost 23 minutes. Same settings, same prompt, same image generation model. You can see while this is running on your screen, M5 on the right, M4 on the left. It's hitting the GPUs at 100%. The NPUs aren't even lighting up for this. So, I don't know what to say about that. The M5 will start out at about 30 watts. But then, unfortunately, it looks like this thing's got a little bit of a thermal throttling problem due to the single fan design, I'm sure. And it settles down to around 20, a little over 20 watts for the duration of the test. So, the M4 is using a little bit less in the high teens, but I mean only a couple watts less. We'll speed this up. And done. 8 minutes and 18 seconds. Meanwhile, the M4 MacBook Pro at this point has only finished 12 out of 30 samples. It's not even halfway done. This is more than 2 and 1/2 times faster. This is a big deal. So, then I figured, okay, fine, but there's no way the M4 Pro will lose to a base model at something that I was under the impression is just raw GPU churning. So over on the M4 Pro Mac Mini back behind me there doing the same thing. It's managing to pull a steady 30 watts. So 50% more power and it finished 50% slower than the M5 base model Mac. The M4 Pro did this exact same job and it finished in over 13 minutes. So this thing obliterates the M4 base model. This thing crushes the M4 Pro chip. So the only thing left to do was to put it up against the M4 Max. Yes, there are deep gouges in my M4 Max. That was a close one. I was careful to make sure all the settings were exactly the same because there's a lot of settings and I don't really know what I'm doing. But same picture size, same resolution, same number of steps, same sampler, same model, same prompt. Even though the M4 Max is chugging down over 40 watts, essentially twice as much power, twice as much electricity as the M5 base model, it took 8 minutes and 22 seconds to finish. The M5 base model beats the M4 Max at image generation using half the power. Do you realize what this would mean if OpenAI could run their entire operation at half the amount of electricity? This is wild. The M5 Max is going to be an absolute image creation, machine learning monster unleashed upon this world. This computer only has 16 gigs of RAM. This one has 36. My mind is boggled. Mark my words. Apple's up to something. Subscribe to my channel. I'm going to be testing the crap out of this M5 MacBook Pro against everything else that comes out until, I guess, forever. Thank you. Goodbye. fire. [sighs] [snorts] Woo. I started using like a cotton pad on the back and it soaks up a lot more of the little gas.

My mind is blown! Im testing my new M5 MacBook Pro and it can do AI image generation as well as Final Cut Pro Magnetic Mask FASTER than the M4 MAX MacBook Pro. Not the M4 Pro, the M4 MAX! I don't think there's ever been a time before this where a base model Apple Silicon Chip has beat its previous generation Max chip at anything. This link just goes to Amazon main page, if you use it it supports my channel to the tune of 2% without costing you anything :) https://amzn.to/474h9Bs