Another " I don't know what to buy" thread.

[neomorpheus]

Sitting on a dilemma.

I need to "new" Mac and these are my options with a budget of up to 1200 US$.

• Base Mac Mini M4, with needing a third party SSD upgrade.
• Upgraded Mac Mini M4 24 GB, 256 MG SSD with needing a third party SSD.
• Mac Mini M4 Pro base.
• Mac Studio M1 Max 32 gb/512GB SSD, which goes for around 940.
• Mac Studio M1 Max 64/1tb, which goes for around 1200.

I am not a photo or video editor.

Plans are:

• Running LLMs. Nothing major, just to learn.
• some VMs, again to test and learn.
• Browsing.
• Teams, discord, different offices suites.
• Programming learning.

So, suggestions as to which one to buy?

PS I already have a nice gaming PC, so gaming is not a need on the Mac side.

Edit Added a but more info.

 

[m1mini]

All these prices are education store prices and you can buy from it even without being affiliated with any educational instittiuon.

If you already have a desktop, I can suggest a macbook air, they come with just enough for your needs. base is 899 on apple education and sometimes even cheaper @ amazon and best buy. up the ram to 24 for best results

13-inch MacBook Air - Sky Blue

Get free shipping on a MacBook Air laptop supercharged by the M4 chip when you order from apple.com. Select a model or customize your own now.

www.apple.com

My ultimate suggestion if you want to get a mac mini, is to go with base m4 and up the ram to 24 its 699 dollars that way on education store. You will have 500 dollars left for your budget

(YOU WILL NEED THE EXTRA HEADROOM FOR SOME VMS AND TO RELIABLY RUN 13-20B MODELS)

Mac mini

Mac mini with the M4 and M4 Pro chips. Built for Apple Intelligence. With front and back ports. Financing options available. Buy now from apple.com.

www.apple.com

You will have 500 dollars left for your budget, with that you can get

4TB Samsung 990 EVO Plus SSD (They are 250 dollars rn on amazon!! usually goes for 300-350)
OWC Thunderbolt 4 - USB4 NVME enclousure

And enjoy 4.256 tb of storage without having to upgrade the ssd



Mac Mini M4 Pro base is actually 100 dollars above your budget with even educational discounts,

Mac mini

Mac mini with the M4 and M4 Pro chips. Built for Apple Intelligence. With front and back ports. Financing options available. Buy now from apple.com.

www.apple.com

 

[neomorpheus]

Thanks for the suggestion, i added a bit more info to the original post.

 

[maflynn]

lans are:

• Running LLMs. Nothing major, just to learn.
• some VMs, again to test and learn.
• Browsing.
• Teams, discord, different offices suites.
• Programming learning.

Given you want LLMs and VMs, those two items will require RAM, so that largely excludes the base Mac Mini. If you upgrade the ram and storage of the base Mac Mini, you're inching ever so closer to the M4 Pro Mini, and so that makes the logical sense to me.

If chatgpt is to be trusted, the M4 Pro comes out on top regarding specs and I believe performance in both single core and multi-core and especially the neural engine, i.e., AI/LLMs. The M1 Max came out in 2022, personally, I'd not consider 3+ old computer when the specs of the newer models outpace it.

 

[neomorpheus]

Thanks @maflynn

I dont plan in running LLMs or VMs all the time, just play with them, so I figure that somewhere between 24 and 32 should be good enough.

The reason that I am considering the Mac Studio M1 is because they are still available new and the minimum RAM is 32 GB and costing around US$900 and with 64 GB going for around 1100. Plus, it already has a 10 GB NIC and way more GPU cores, which I read still matters and are still faster than the M4 Pro, since there are more, unless I'm mistaken in those metrics.

Also, reading many, many posts, it looks like the performance of the M1 is still good compared to the M4s on these tasks.

Granted, I do agree with you that its not necessarily a good idea to buy what is essentially a 3+ years old computer for that much money.

But Apple's greed is forcing the hands of their loyal customers with insane pricing.

The problem is , I still need to have a Mac, for career reasons and wouldn't mind having whatever I buy, as my primary system.

 

[MacCheetah3]

• some VMs, again to test and learn.

In many uses, assigning 2-4 CPU cores and 4-8GB of RAM is sufficient. So, typically, not a huge resource hog.

• Programming learning.

A lot of software development has modest requirements. Most files are text-based, of course, or images. Some of the working files and caches can grow into the gigabytes. However, storage really only bloats if you’re using VMs/containers or (otherwise) performing wide compatibility testing. For example, with simulators going back to iOS/tvOS 15 and having downloaded the Predictive Code Completion Model, my Xcode developer footprint is ~50GB.

• Browsing.

How impactful (i.e., how inefficient or problematic) Web browsing has a lot to do with bad habits. That is, if you use tabs/windows as bookmarks or rarely restart/reboot the app/system. (As someone who learned the consequences the hard way by ignoring the logic, I’ll continue to shamelessly criticize such behavior.) Don’t misunderstand, in general, the modern Web is stuffed with garbage, from several aspects — by the way, this is also relevant if you intend to pursue Web development. Nonetheless, having bad Web (browsing) habits is a substantial cause of problems.

• Running LLMs. Nothing major, just to learn.

On this topic, it’s best if I point you to another source. I have no knowledge or interest in local LLM dabbling.

Best Mac For Large Language Models (MacBook, Mac Studio, Mini)

Find out what is the best Mac model with apple silicone (M1/M2/M3 chips) to run large language model inference locally. And learn how does it perform.

www.hardware-corner.net

 

[neomorpheus]

In many uses, assigning 2-4 CPU cores and 4-8GB of RAM is sufficient. So, typically, not a huge resource hog.

I know and running one VM is not that bad.

But when running more could be an issue or having one VM doing some real intensive work.

Nonetheless, having bad Web (browsing) habits is a substantial cause of problems.

Agreed!

I have no knowledge or interest in local LLM dabbling.

Its just something to mess with, not to take it too seriously, since I am not a programmer and the real money is for those people in the AI field.

I just like to have a little freedom to play with it on my own hardware.

 

[roman.stapunov]

Mac Studio M1 Max 64/1tb, which goes for around 1200.

This option looks a sweet spot for your budget, but if I were you I would buy m4 pro base. For your use cases it will be enough and will be supported longer.

 

[JSRinUK]

Its just something to mess with, not to take it too seriously, since I am not a programmer and the real money is for those people in the AI field.

I just like to have a little freedom to play with it on my own hardware.

I had no clue about LLM until I got my MacBook Pro. I very quickly decided that I wanted to dabble more and got a Studio to complement it.

My advice for dabbling would be to not go with a base M4 model. The base M4 has the slowest memory bandwidth and, although I haven’t used a base model, I wouldn’t recommend the slowest bandwidth. And, typically, you can’t use the full RAM for GPU so base 16GB will feel restrictive.

I feel my MBP (M4Pro, 24GB, 1TB) would be the lowest “dabbling” spec. You can scrimp on the SSD if you plan to store your LLMs on an external SSD.

I don’t have experience of the older models. You can run tiny LLMs on very low-spec machines (even on a phone) but you wouldn’t get encouraging results.