bouncer
← Back

SPACE DESIGN WAREHOUSE · 25.2K views · 1.2K likes

Analysis Summary

30% Low Influence
mildmoderatesevere

“Be aware that the highly specific technical details about 'SOIC MH' and 'super cores' are speculative and serve to make the creator's purchase recommendations feel more scientifically grounded than they currently are.”

Ask yourself: “What would I have to already believe for this argument to make sense?”

Transparency Mostly Transparent
Primary technique

Appeal to authority

Citing an expert or institution to support a claim, substituting their credibility for evidence you can evaluate yourself. Legitimate when the authority is relevant; manipulative when they aren't qualified or when the citation is vague.

Argumentum ad verecundiam (Locke, 1690); Cialdini's Authority principle (1984)

Human Detected
95%

Signals

The content exhibits high levels of spontaneous speech, personal workflow context, and self-correction that are characteristic of a human creator. The technical speculation is delivered with a unique personal voice and humor that deviates from formulaic AI scripts.

Natural Speech Patterns Self-corrections ('Silicon silicon'), filler phrases ('All right, let's try this again'), and conversational tangents ('[ __ ] nerd tech tubers').
Personal Anecdotes References specific personal workflows like the 'magnetic mask in Final Cut Pro' and previous videos on his own channel.
Metadata Authenticity Description includes specific gear lists, a joke about a 'Klingon' secret vlog, and a distinct personality consistent with the transcript.

Worth Noting

Positive elements

  • This video provides a clear, accessible explanation of semiconductor 'binning' and the potential shift from SOC to chiplet architectures in consumer laptops.

Be Aware

Cautionary elements

  • The use of highly granular technical speculation (like nanometer measurements and specific core counts) can create a 'veneer of certainty' that masks the fact that the buying advice is based on rumors.

Influence Dimensions

How are these scored?
About this analysis

Knowing about these techniques makes them visible, not powerless. The ones that work best on you are the ones that match beliefs you already hold.

This analysis is a tool for your own thinking — what you do with it is up to you.

Analyzed March 13, 2026 at 16:07 UTC Model google/gemini-3-flash-preview-20251217
Transcript

I know all these computers are going to be here in like three days, but there's still time to speculate. First, I want to talk about the M5 chips, cuz the M5 base chip is basically just a spec bump with a huge asterisk on whatever these neural accelerators are on the GPUs. All right, let's try this again. It's pretty It's better than the last one. Previously on the Space Warehouse, I showed that the base model M5 MacBook Pro beats the M4 Max MacBook Pro, a $3,500 computer that's only one year old at local AI image generation. And it does that while sipping half as many watts out of its battery. Speaking of sipping, that alone is outrageous. But in the last few months, the only other regular computer task that I actually do on a laptop that seems to take advantage of these GPU accelerators in that same way is the magnetic mask in Final Cut Pro. The thing that cuts me out of the background of this shot so that I can put texts or images back there between me and the background. In Apple's new ads for these computers, they talk a lot about it speeding up AI stuff. I just don't know what AI stuff normal people use other than chat bots and coding bots, but those don't run locally in most situations. So, if you don't care about running AI image generation on device, and if you don't edit video, then who cares? It makes custom emojis faster if you use image playground. But is that worth spending another thousand bucks? It isn't. Or is it? No, it's not. I do have a speculation that this same superpower will apply to Apple's custom AI ondevice LLM that they simply must be working on. And I do think that these neural accelerators will be the key that unlocks basically a chat GPT type thing that runs on your computer in the background all the time. The faster SSDs support that theory because I would expect the LLM will load on demand when it's needed, but doesn't stay in the memory because those things use a lot of memory. So, these computers can swap out whatever's in the memory real fast, either to RAM compression or to the swap, and swap in these LLMs with these super fast SSDs, cuz I think on the Pro and Max models, the SSDs are now like 14 GB per second or something, which is close to the speed memory used to be like 15 or 20 years ago. Anyway, the existing M5 chip and the upcoming Pro and Max chips are two fundamentally different things now. Which means unlike with M1 all the way through M4, in those computers when you bumped up from the base model MacBook Air all the way to the most expensive MacBook Pros, you just get a bigger chip that had more cores. Like the M4 Max chip is just a really big version of the M4 chip, but they're different chips. The CPU cores are the same on all of them, but the chips themselves are different. The SOC, the system on a chip, well, not anymore. The M5 that's in the upcoming MacBook Air is the same chip as the M5 that's in the MacBook Pro. So the only difference between the brains of those computers is going to be thermal throttling and then the body stuff like a nicer screen, SD cards, more ports, and a bigger battery. But once you step up into the Pro or Max M5 chip, you are now waiting into Apple's first chiplet design. You've probably heard me or other [ __ ] nerd tech tubers use the term SOC, system on a chip. That's where the CPU, GPU, NPU, and RAM are all on one piece of silicon. Silicon silicon. Well, this is no longer that. The M5 Pro chiplet is now an SOIC system on integrated chips, a stacked die. Apple calls it fusion architecture. Basically, the M5 Pro and the M5 Max now start out as the same thing, an 18 core CPU with six super cores and 12 performance cores. More on that in a minute. sharing a chiplet with an NPU and Thunderbolt controllers. Then that chiplet is connected horizontally, not stacked, using TSMC's SOIC MH packaging. MH for molding horizontal with the GPU chiplet. And this second chiplet, the GPU one, is where you get the differentiation between the Pro and the Max. So now Apple can pump out the CPU chiplets that go with both and connect them to either a 20 core GPU single media engine and RAM controller for the Pro or a 40GPU 2X media engine 2X RAM controller for the Max. Also, most of what I just said is also speculation, but based on a ton of research from people close to TSMC and Apple and makes sense for the design. Apple used to release a little cartoon diagram of the chip itself when they would make these announcements, but that is conspicuously missing for this one. Hence the speculation on how it's all put together. Since you get the same CPU and NPU and Thunderbolt capability for both the Pro and the Max, and since we now know it's a fused die chiplet design, the rest of this sort of just falls into place logically. I'm going to talk a little bit about bend cores. Skip to this point in the video if you just want more M5 super core, performance core, efficiency core mumbo jumbo, and you don't want to learn something. This actually stood out to me with Schnazzy Labs breakdown of these products. He was talking about the Bend chips. A lot of tech review folks when talking about CPUs in general will use this term and just move on. And I'll bet there's a pretty significant chunk of people who kind of understand what it means but don't really understand what it means. Like I'm sure you get the idea that a bin CPU core is worse somehow. It's slower somehow. But I would like to explain briefly what that actually means. When TSMC is printing these chips, they make them out of these big wafers of silicon. To make the CPU, they literally just etch into the silicon. They use these obscenely precise mirrors and lights to dig microscopic channels that will later become the transistors and wiring pathways. If that wafer of silicon has a single defect, and we're talking about like a bump the size of a virus particle, people like to use the phrase a size of a human hair for the measurement of small things. But the 3nm process and that's interestingly not a measurement of any particular part in chipm. I guess it used to be, but now it's just a nomenclature. The smallest element on a chip is actually something like 48 nm. Weird. At 48 nanometers, you could fit something like 2,000 transistors across the diameter of one human hair. And if even one of those ised up in some way, if we're off by even one, 529,000th of an inch, that core might not work properly. So if you're printing a sheet of 18 core CPUs and two of those cores on one CPU just happen to land on a divot 1 529,000th of an inch wide, they get a little [ __ ] up and they don't work. And now in that 18 core CPU, 16 of those cores work just fine, but two of them are defective in some way. They're either too slow or they don't work. After they print one of those wafers full of CPUs, they'll test every core. And on the CPUs that don't pass fully, they'll disable the bad cores and sell those CPUs as slightly cheaper, slightly less core models of the same thing. And that's how we end up with the 15 core CPU option on the M5 Pro. Otherwise, I'd have to just throw out the CPUs that don't have Oh, we just gained 100 new subscribers. How about that? Otherwise, they'd have to just throw the bad ones away, the ones that don't have 18 out of 18 cores working. So, by sharing the CPU chiplet with both M5 Pro and M5 Max, now Apple can make the manufacturing process more efficient. They don't need a completely different design for the Max chips. They just fuse on a bigger GPU chiplet. Onto the chips themselves, Super Core. Evidently, there's a new core in town. And actually, I think the new core is the performance core. I think these super cores are what the performance cores were. But now the performance cores are like supercharged efficiency cores, but we still have the efficiency cores in the base model MacBook Pro and the new Air. Three core types or maybe four core types because the M5 iPad Pro still says performance cores and efficiency cores. There's no way Apple forgot to update their website, right? As it stands, the new M5 MacBook Air and the base model M5 MacBook Pro have a combination of these new super cores and the old efficiency cores. But the Pro and Max models have no efficiency cores. They just have super cores and performance cores. The rumors say that the new performance cores in the Pro and Max model will hit over 4 GHz, which would mean they really are fundamentally different from the efficiency cores in the lower models. I unfortunately ordered too slow and my M5 Pro isn't coming until the end of the month. Plus, my wife scheduled a vacation for exactly at launch time. And apparently I'm going to go up to the Northeast and spend a week in George Washington's wife's grandfather's cabin and the Watergate Hotel. What a week. Anyway, we already know now though that the M5 Air will have essentially the exact same performance as this computer, the M5 MacBook Pro, until you heat it up. But you know what? Most of you aren't going to heat it up. Video editors won't. I don't think audio people will. Office workers certainly won't. Small 3D modelers like custom 3D printer people won't. Honestly, in my opinion, the Neo takes the place of who should have been buying the Air all along. And the Air can pretty much replace the base model MacBook Pro. And if you need an M5 Pro or Max at this point, you have to know that already. There's no way this type of video is helpful to you. The M5 is more efficient than the M4 was. So, heating up an M5 MacBook Air will be even harder than it was before. Of course, you might be buying an Air to save money and then using your Air in a way that's designed for a MacBook Pro, like exporting gigantic things, which calls for active cooling, but then you should just expect the computer to feel normal when you're working on something and then be slower when you're exporting that thing. Apple really leaned on AI in their marketing for these. So, I imagine that's getting stuffed into every aspect of all new computers, maybe forever going forward. So, I do feel like I need to stand by my previous RAM recommendation requirement of 24 GB. These new GPUs are basically a video editor's dream. The Pro and Max models are going to be just so lovely to see as we get more and more automation processes for video editing that actually make sense to use. Like, I can't believe there's still not an initial cut editor for something like Final Cut Pro that'll just take out the spaces between sentences. But there is stuff like the new transcribe features in Final Cut Pro where as you import your footage, it's automatically transcribed into captions. And therefore, you can search your own footage for like a single phrase that you said somewhere in the video. it'll pop to that point in the video. And that's all going to be much faster with a huge GPU count when each GPU has its own neural accelerator. With the Pro that I ordered, my favorite magnetic mask tool is going to speed up like 300%. Which will be the first thing in a few releases of MacBooks that'll actually save a meaningful amount of time for my workflow. Apple says their media engine is significantly improved, too, so exporting should be a lot faster on these new machines, too. I don't know. The M4s were already too much computer for most normal people. So, I fully expect a whole suite of Siri 2.0. Maybe they'll call it Super Siri to come out this year and make use of this super futuristic chipleten enhanced computer. Goodbye. If you listened all the way to the end, there is a Klingon audio in this one. Today's subscriber number 82,690. Those just I made these the last three videos now. Little time stamps. But look at that. We're We're cruising cruising to 100,000. Join me.

Video description

We'll have real information VERY soon, but for now, we speculate. The M5 Pro and Max are both built on a Chiplet Design! TSMC SOIC MH package! Molding Horizontal. Basically the M5 Pro and M5 Max have the same CPU now (with a binned lower core count option for base model Pro) and then those chiplets are attached to a GPU chiplet with either 20 - 40 GPU cores and double the media engines and RAM bandwidth for the MAX. None of what Im saying here is final or proved yet, its just probably true, most likely, I think. M5 MacBook Air: https://amzn.to/4s0Oi9f M5 Pro Macbook Pro: https://amzn.to/4uj5lVy Ugreen External NVMe SSD enclosure: https://amzn.to/3OYKyX9 My favorite battery bank by Anker: https://amzn.to/4udOvHH I shoot with this camera: https://amzn.to/4riCRIQ Also there is a secret VLOG embedded into this video if you go to settings and change the language to Klingon, youll hear it. Join this channel to get access to perks: https://www.youtube.com/channel/UC_lGMZMTN7Oh9DRKV2eaMtg/join

© 2026 GrayBeam Technology Privacy v0.1.0 · ac93850 · 2026-04-03 22:43 UTC