Buying a Mac for AI: The Only Guide You Need

Comments · 149

• AL

  @alexgrankin · 13 days ago · pinned

  100,000 views, thank you all! As promised, the tool is now open source: <a href="https://github.com/alexgrankinukr-hash/llm-sizer">https://github.com/alexgrankinukr-hash/llm-sizer</a> (MIT). Add a machine, correct a number, or run it yourself. The tool stays free at <a href="http://theaibridges.com/tools/llm-sizer">theaibridges.com/tools/llm-sizer</a>, and since the video it gained a Reading tab (the wait before the first word), linked machines (two Sparks, up to four Macs), and the Flash Next builds that run on 64 GB. Thank you for the comments. <br><br>P.S. New video on Mac Read &amp; Write Speed breakdown for all mac generations will go live later this week! Stay tuned:)

  3

• RI

  @richardr.5177 · 2 weeks ago

  I was waiting for the new Mac Studios for local AI experimentation and development.<br><br>The M5 Ultra with 256 GB is the first model that really feels powerful enough and reasonably future-proof. The 512 GB version even more so. But at €11K to €17K?, there are very few local AI use cases that justify the investment.<br><br>Even the M5 Max with 128 GB is already over €6K.<br><br>Maybe the smartest move is simply to wait another year or two and see how fast local AI and the hardware needed to run it evolves.

  5

• AL

  @albuslee4831 · 3 weeks ago

  You are an excellent primary school computer science teacher I would have wanted back when I was a kid.

  8

• MA

  @magicalcat2000 · 3 days ago (edited)

  The explanation was very easy to understand, and your tool has helped me choose what I can afford versus what I would like 👍

• AJ

  @ajoneal-individual2222 · 2 weeks ago (edited)

  Bar none, this is the best, most practical layout and explanation of this information.<br>I&apos;m glad that you didn&apos;t put too much emphasis on the fact that the M1, M2, etc all run at nearly the same memory speeds - I&apos;m hoping to snag some older hardware at a lower price for the same token speeds and people trade &quot;up&quot; for the M5s.

  3

• TH

  @TheGijzzz · 3 weeks ago

  Its exactly what you say at the start, the models that are considered frontier models this moment are four months later open models. As we speak China released new open models every month and lowers the bar for the hardware all the time. China knows the future of AI will be decentalized and that is why they develop AI for home users. China has also fired up its memory industry so they will provide the world with more memory chips so prices of hardware will come down next year. There are also projects to share compute hardware power among users like darkbloom. And the current big general LLM’s can be cut into pieces so general modells can be loaded as smaller specialized parts so you can use them on light hardware. There are plenty of possibilities to alter the AI software so consumer computers can handel it . China is working hard to make it possible to get AI in the hands of consumers while US big tech is doing the opposite because their whole business model is based on centralized AI in the hands of big tech so you will have to pay them for token costs and compute power. Most consumer and small business AI use doesnt need frontiermodels. Lets be honest we dont need to develop our own x-box games or create our own movies. We do need an personal assistent that reminds and makes appointments or answers the mailbox. So yep for 90% of use small open models that can run on lightweight hardware are the future.

  34

• SU

  @SuperJaegermaster · 3 weeks ago (edited)

  again, great video and fantastic work &nbsp;for purchase decision. Subbed !

  10

• JO

  @joncorreia · 3 weeks ago

  what a goldmine, thank you for building this and sharing. so excellent!

  3

• ME

  @melkyoretor1824 · 10 days ago (edited)

  Wow this is insanely clear and to the point. I finally understand all the different parameters and how to make my choice. I have been agonizing over the M5 Max 128GB and M5 Ultra 96GB. Thanks to you and your charts, I&apos;m going to go for the Ultra for the extra bandwidth since 128GB doesn&apos;t get me a &quot;superior&quot; model (without degradation) at 250k context window(at least not today). And that&apos;s not even considering QWEN will continue to get better at the same size. This is exactly the kind of information I was looking for (on youtube) and couldn&apos;t find. So clear and to the point. Liked!

• SI

  @SimaKumari-w9o2i · 12 days ago

  This channel is exactly what I needed It helped me understand better.

• SP

  @SpaceUnicornProductions · 2 weeks ago

  I have been waiting for a tool like this! Awesome work and great video! :)

• VA

  @VastIllumination · 3 weeks ago

  This was very helpful

  3