theenterprisemac

[theenterprisemac]

Is the Minisforum AI X1 Pro Better Than the M4 Mac Mini?

I want to compare the Minisforum AI X1 Pro to an M4 Mac Mini, Apple's cheapest machine. Let's start by looking at cost:

AI X1 Pro ranges from: $929 to $1369

M4 Mac Mini ranges from: $499 to $4269 (If we create an approximate configuration to match the AI X1 Pro's middle configuration then: $2289.)

What about benchmarks:

AI X1 Pro manages: 2973 single core, 12758 multi-core

M4 Mini manages: (base) 3748 single core, 14467 multi-core (high-end matching the AI X1 Pro) 3829 single core and 22495 multi-core

As percentages the M4 Mini is 26-28% faster single-core and between 13-76% faster multi-core. Looking at VRAM: the Mini can allocate natively up to 75% of its memory to the GPU so at most 48GBs. You can change this in software to get up to 56GB. The percentage scales as the amount of memory you have gets smaller. The AI X1 Pro can also allocate 75% so at most 72GB VRAM. However, in practice I am not sure if this is achievable. I have also seen claims that the limit is 50% which would make the peak 48GB. The VRAM question is a bit hard to pick a winner on.

Looking at the NPUs: the AI X1 Pro manages 50 TOPs to the Minis 38 TOPs a difference of 31.5% better performance for the AI X1 Pro. However, the Minis NPU is more usable and more supported as of 04/2025.

Storage isn't worth mentioning since of course you can upgrade the AI X1 Pro. Also for RAM the AI X1 Pro also wins given that it supports up to 128GB of RAM.

The answer as to which is better is not actually cut and dry. At the low-end the Mini beats the AI X1 Pro in benchmarks, but its starting configurations is half the RAM and a quarter of the disk space. Once you make them reasonably comparable in specs the Mini cost $2289 and AI X1 Pro is only $1229 which is 53% of the price.

Really the answer is: what machine is going to work best for you. That can mean a lot of things from cost to platform preferences etc. I think that is the fun thing about having options—comparisons like this are almost meaningless. It actually really just depends on preferences and to a certain extent needs. However, both machines are very powerful and suitable for a wide range of needs.

[theenterprisemac]

Category - Terrible Ideas for Mac Management

So as you'd guess this comes from someone who hasn't been in the trenches and if they "have" it has been at a management level.

So what's wrong with this idea? Well a lot of things. First issue is that you are feeding your information into an LLM and expecting to get actionable insights out of it. This is silly because LLMs get things wrong ALL the time! Just today I asked a basic question about the open source Linux kernel–source available, and Chat GPT got it wrong. It specified the wrong file–wasn't even close.

Second issue I have with this is: it is promoting more of this no code nonsense. I love easy, but using someone else's code where you can't even see how it works is just a bad practice. In this case the recommendation was n8n, which is open source, but how many people are really going to look under the covers?

Finally, I would love to understand what "insights" do you expect to get out of feeding this to an LLM that you can't get with some automated Python analysis or some quick work in Excel? Additionally, how often do these reports really change? In my experience there is a lot of data that is a regular same report, and some one-offs. The amount of time you spend building custom reports is mostly mitigated by how you collect your data–not feeding it to a gibberish factory.

Once you are collecting the right data you can munge and report in all manner of ways, and none of them require you to use an unreliable gibberish factory. Bottomline I don't see what this is actually solving? What report is this going to make for you that you can't do without this process? If you are an administrator you can get this out of Jamf. If you have to give reports daily, then you can munge this just as easily without the LLM bit.

[theenterprisemac]

M4 Pro Mac Mini

Now that I have been using the M4 Pro Mini for a little bit I have some pros and cons to share.

Pros

M4 Pro is very powerful. Lots of performance cores and quite substantial performance boosts in computationally intensive tasks.

The form factor is impressively small.

You get a good selection of USB-C and Thunderbolt ports with your usual HDMI and Ethernet (which can be upgraded to 10GbE, and you should).

Machine stays quiet for the most part. The fan does a decent job cooling the machine.

Lots of memory! 64GB max finally! Means even more VRAM. Without any software modification you can get 46GBs of VRAM.

Fast storage, as usual, but it's always good to see. Even better since it's socketed storage.

Cons

M4 Pro compared to an M3 Pro isn't necessarily all that impressive of a performance jump.

You can't freaking rack mount it in 1U...🤬

You lose the 2 USB-A ports from the 2018 Minis which for some of us is kinda annoying.

The machine does not have the same thermal head room that you got with the older Mac Mini design.

Why is it still only 64GB? I'll tell you why–because Apple uses that to get people to spring for the MAX variant of the chip 😡.

Still doing that Apple proprietary thing for their storage.

[theenterprisemac]

M4 Pro Mini LLM Performance

It will do a 70B parameter model, if you have 64GB ram, at ~5 tokens per second. It consumes roughly 46GB of ram.

[theenterprisemac]

Mac Mini M4 Pro

NVIDIA 5090 median price $3900 and 32GB VRAM. Mac Mini with M4 Pro, 10GbE, 64GB unified memory and 2TB storage $2899 giving you 48GB VRAM. Yeah I think the Mini might have the edge here.

[theenterprisemac]

Have Care With Your LinkedIn Profile

I recently encountered a profile of someone who has been out of a job for more than a year. They are in the Mac IT space, and while my heart goes out to them–the reason why they still don't have a job was obvious.

Obviously, I have no intention of mentioning names, making it possible to figure out who I am talking about or castigating this person. With that in mind I want to give some broad strokes ideas of things to avoid on your profile–especially when looking for a job.

Some people say avoid the open to work banner–I haven't had an issue with it, but to be fair I have no empirical data. This is one that you should test empirically and see what works best.

If you have a side hustle make sure that it's clear that it's a side hustle. Especially, if you have been doing it for a long time. Seeing something that isn't obviously a side hustle; that you've been doing for a long time and are still doing, while looking for a job, sends the wrong message. Here is what I mean: it immediately begs the question of why is it on your profile, and if you have been doing it for so long why are you looking for a job?

If you have been laid off that's fine to say, but if the reason you were laid off is remotely your fault, or can be construed that way don't be explicit. A potential employer may ask, but you don't need to broadcast it on your profile. If you say: "I was laid off" doesn't imply anything negative. If you say: "I was laid off because I didn't do X, or couldn't do Y" then you have an issue. To an outside observer you look incompetent–or at least potentially so.

Be careful about putting numbers on your profile. Things like I saved some number of hours of work, or cut costs by some percentage etc. If you do, then you have to be able to back that up with evidence. If you can't, then get rid of it ASAP! I have interviewed people who put that sort of stuff on their resume, and I always ask–if they don't have a good answer, then that is a significant moment of pause for me.

Getting more certifications etc while unemployed is great! Do it! However, think about what you are getting. Look at where you are; where you want to be, and consider if these certifications align with the direction you are going in. If not, then still get them, if you want, but maybe leave them off your profile until your new job is sorted. There are other ways to show growth without getting certifications. Having a long list of not obviously relevant certifications is not going to help, and actually hurts you, because you look unfocused.

Be careful what you share on LinkedIn. It is all well and good to post, but consider what you are posting, and what a potential employer might think. I am not saying it should have an impact, but I am saying it does.

If you have been in IT for a long time, then be careful spending too much time trumpeting accomplishments that don't align with what someone might expect from someone with your years of experience. For example: you have 20 years of experience, then perhaps don't say I resolved 3000 tickets this year. Maybe instead focus on the work that those tickets represent–not the quantity.

Whatever you do–don't sound desperate. Job hunting is brutal and sounding desperate won't help.

If you want to use something like LinkedIn as a way to show thought leadership, then be careful to make posts that align with that. Posting links with surface comments, or summaries is not thought leadership–you need to show you have your own reasoned opinions.

Distinguish between statistics of your environment and accomplishments. For example: don't list the fact that you managed X number of devices as an accomplishment. List that as a statistic for scope of your activity. The accomplishment would be something like: I rolled out a new OS, or security tool, or automated some such thing.

Hopefully this is helpful. If you disagree, then tell me why.

[theenterprisemac]

Blowing Away Kandji Marketing Fog

Allow me to blow away some of this fog that Kandji's head of EMEA sales is trying to blow up your nose. Let's start with the only relevant paragraphs in this post:

"First, what are Smart Groups? Smart groups are a seemingly powerful feature in Jamf Pro, enabling administrators to dynamically group devices based on specific criteria (chip type, user group membership, whether it has a specific app installed, etc). However, their complexity and the challenges they introduce can often lead to frustration for IT teams."

I love the use of "seemingly powerful"–what does that mean? Jamf smart groups aren't seemingly powerful. Between the long list of attributes you can filter on, custom attributes and the ability to use boolean logic with parentheses you can robustly group your fleet for whatever purpose.

Are smart groups more complicated than assignment maps? Yes. They also do a lot more. This is the key point that the sales manager leaves out. Assignment maps aren't easier to use and equally powerful. Assignment maps are easier to use and are way less powerful. I discuss this in my review–post Kandji trial.

The problem boils down to Kandji giving you no equivalent to extension attributes–that you can use to group computers together. Kandji's list of attributes you can use in their assignment maps covers the basics for the simple cases. However, anything outside of the obvious is going to be a struggle.

Yes, you have tags, but Kandji doesn't allow you to run a script on a machine and then have the output of that script appended to the computers record so you can now identify all machines that have that feature.

At the time of my trial the list only included:

OS Version

Enrollment Type

Chip Type

File Vault

Supervision Status

Mac Family

Asset Tag

Serial Number

Tags

It doesn't take tremendous imagination to figure out that there are a myriad of situations that would be impossible to solve for with the above list–especially given that tags aren't extension attributes they are just tags that you can label a computer with, and not a way of appending additional fields to a computer's record.

So let's finish blowing the fog out of our noses:

Smart groups are complicated compared to assignment maps. But, they aren't rocket science and if you are one of those Mac admins who complains about their "complexity" you need to do some self reflection.

Don't mistake simple for better. Sometimes it is, but that assumes that simple does what you need. A sales manager with no actual IT experience isn't going to have a clue about the corner cases and will happily sell you a second rate non-solution–just because they don't know what they don't know.

This approach by Kandji of trying to build themselves up by tearing down Jamf is both pitiful and wrong headed. A competent Mac admin is going to see right through this noise.

Mac administration isn't supposed to be easy. It's supposed to enable people's work. Complaining about boolean logic and all the options of smart groups reflects poorly on the profession. There are issues with Smart Groups, but over complexity isn't one them. Smart Groups are very powerful, and sometimes you can't achieve the required power level in a simple paradigm. Smart Groups implementation could use an overhaul in how it is represented, but the underlying power is absolutely necessary!

[theenterprisemac]

Sad But Tends to be True

It's unfortunate, but a recurring theme that I notice in the YouTube tech space is that videos that are titled something to the effect of "The N Best Products for Some Use Case". You know the type, and in my experience most of those videos tend to be ways to push through Amazon affiliate links and satisfy sponsors in the guise of a recommendation video.

I really wish that these videos were more about recommending a good product and less about force feeding affiliate links. I also wish "long-term" review didn't mean I used it for a couple months.

To me this is a troubling trend. I don't begrudge people a living, but I feel like more diligence is required when pumping out this stuff. So many times the recommendations are just a rosy gloss over the details sales pitch.

I really wish there were more videos where the creator looked at a number of products; did an in depth comparison, and then made a recommendation–where the affiliate link was for their recommendation–not for everything they talked about in the video.

[theenterprisemac]

Whacky Speaker Recommendation

"…just how bad some of these Macs are with multiple Bluetooth connections this is more of a Mac problem and not a Sonos problem but the connection drops are so infuriating that if you do go this route make sure you don't have a ton of competing Bluetooth devices…"

So wait let me get this straight; he is connecting the speakers with Bluetooth and they drop out a lot. Then he says:

"…you can plug it in directly via USBC cables and some specific adapters that Sonos sells but it does end up impacting the audio quality and the volume levels quite a bit so l don't really like to use it that way i actually just don't recommend doing that at all these speakers though…"

So, not only does the Bluetooth drop out a lot, but the direct attach is garbage–so why is he recommending them? Is anyone else confused by this? Are we sure this isn't just a way to get you to click on those Amazon affiliate links in the video description?

Oh, and in case you are wondering the answer to the above question about the affiliate links is: yes. You can watch the entire video–it honestly makes me itch having so much sunshine blown at me, but the video is just a way to pump some Amazon affiliate links and give a nod to the sponsor of the video SetApp which I have talked about before.

[theenterprisemac]

PCCVRE - Networking

So I don't lose this information:

Within PCCVRE Apple has three routing options:

Router2 - which supports two nodes it seems in the same chassis

Router4 - which supports four nodes in the same chassis again it seems

Router8Hypercube - which supports the full nodes from two chassis the second chassis being a "partner chassis"

Not sure if there is support past 8 nodes working together. However, if you think about this from the standpoint of it being M2 Ultra or better CPUs I suspect this is quite substantial horsepower for running inference.

The communications are encrypted within the collection of nodes doing the work. There is a concept of a leader and nodes that provide data that the leader combines into the final output.

[theenterprisemac]

In Case You Thought JAMF Innovation Wasn't Dead

Just in case you thought Jamf was still innovating–just further proof of the mode they are in. Don't innovate–buy some company doing something you want to add to the portfolio–slap the Jamf badge on it–call it a day.

I wonder how long this one will take to integrate. The last few have taken positively forever.

[theenterprisemac]

Replacing the Last of my Intel Mac Minis

I have been thinking about this for a while now, and I have finally decided that the best bet for me is to replace my 2018 Intel Mac Minis with a pair of M4 Mac Minis. Even with two they are still cheaper than a single Mac Studio that would be sufficient for the purposes both satisfy.

The specs I am planning on:

M4 Pro 14 Core

64GB of RAM

2 TB Drive

10Gb Ethernet

Would the high end M4 Max processor be nice–sure, but I can't justify the cost especially since these machines are used for reverse engineering primarily.

I still hate the power button location, but I think this will be just what is needed to dump the last of my Intel Mac Minis from 2018.

[theenterprisemac]

LinkedIn Posting No More

I have made my decision. For the time being I will no longer be posting on LinkedIn–minus the occasional post. My posting will only be on this blog.

This has been a nicer experience, and getting away from LinkedIn and the noise has been refreshing mentally.

[theenterprisemac]

AI Hysteria Strikes Again

Let me be clear, there are some real risks with AI being able to generate convincing content that is not real. Image, video, and audio generation are all of great concern. However, we all need to take a collective deep breath and evaluate the actual situation. In the realm of image generation for every realistic, even on close inspection, AI generated image there are thousands of garbage photos.

The above post is such an example. On the surface it looks real–it looks good. However, not even looking that close you can quickly see the problems:

The hands look like they went through a blender.

The feet have six toes.

The shadow doesn't make sense.

The watch is just a mumbo jumbo of pixels.

I am not even looking hard and this is what I see. So, do we need to be concerned, yes. Is this a cause for panic–maybe not yet. What we have is cause for caution. However, the soberness we need to approach current "AI" with is the same soberness that we need when, and should be, approaching every new technology with.

"AI" as it is now is an intermediate step to something else, and is not truly transformational. However, some of the ideas that are being developed will lead to something properly transformational down the road.

If we all adopt a properly sober mindset we can get through and have fun with this "AI" craze–or we can let it get the best of us. I for one intend to enjoy it, and not let it get the best of me. I think the above image is a good example of the mindset you should have: don't just believe–test and validate–then believe.

[theenterprisemac]

This Makes No Sense

I really wish this post didn't paint the great good that will come of dumping the current design of Jamf Pro for something modern with the icky brush of irrelevant fallacious logic.

Jamf made the decision to make their new management stuff micro services. They also chose to design it to require Jamf SSO. This was not the only way to do things. Jamf is trying to pave the way for no on-prem Jamf. You can read between the lines and see that.

You can see just how much Jamf and Kandji are converging in the way they do things–as expected with these kinds of products–given that Jamf and Kandji try to solve narrow solution space with shockingly similar designs.

Kandji is in design closer to Jamf NOW. Jamf Pro is now becoming more like Kandji; adopting a cloud first approach and now using the idea of micro services. I assume Kandji does the same thing–they just started there–which makes sense given that Kandji only dates back to 2018.

At the end of the day this move from Jamf should be a good one, but I just wish they didn't feel the need to insult my intelligence by trying to tell my Jamf SSO was necessary for these new ideas.

[theenterprisemac]

Rolls Eyes Shrugs and Replies

I don't know what other people's reasons for saying Kandji is for small businesses, but I can tell you from my experience the reason Kandji is a small business solution: it is too confined in the solutions it can provide.

Kandji just is not that flexible. Kandji very much sells the Kandji version of Mac management. If your problems and solutions fit in their solution–great. I assume there are some businesses even some larger ones that this works. However, in the enterprise, having such a limited tool chest that doesn't allow you to design beyond the solutions that the product engineers designed for–is a poor fit for a complex enterprise environment typically.

I encountered this all through out my testing of Kandji. Even simple problems that you can solve in Jamf with an extension attribute are an absolute nightmare to solve in Kandji.

I also want to take a bit of a jab at the "we've beefed up our infrastructure" line. It makes me laugh when they say that because as you can read and see here–a mere two years ago Kandji didn't even bother doing static analysis on their code. So square that with advanced enterprise solution.

I also want to poke at the initial comment about how businesses managing thousands of Apple devices use Kandji, and therefore Kandji is ready for the enterprise. This is such a sad argument–the companies might just be doing iOS devices which is very different from the Mac. There are also companies who just want easy button Mac solutions–which Kandji is compared to Jamf. However, you the reader should not be taken by such waffle. The right solution is the one that solves today's problems and can solve tomorrow's problems. I think this is where Kandji falls short. It may solve today's problems, for the right use case, but solving tomorrow's problems–I am less certain.

[theenterprisemac]

How to Identify AI Charlatans

Some general rules of thumb for how you can identify someone as being an AI charlatan:

Are they saying that current LLMs are AGI (Artificial General Intelligence) or close to AGI?

Do they have an opinion on fair use in AI training that cheapens original work and scholarship?

Do they recommend you use LLMs to generate the bulk of your content such as social media posts?

Do they claim that AI is going to revolutionize education in the immediate future?

Do they think of LLMs as a mentor, teacher or other type of learning aid for new ideas and concepts?

Do they force AI into everything even when it doesn’t really make sense?

If they don’t distinguish between the value of a small vs large model for quality of answers.

If they claim that AI is going to take away jobs in a big way.

If they think that AI is a substitute for expert human knowledge.

Do they think that LLMs reason in anything approaching the way a competent human reasons and can achieve similar outcomes?

If they gloss over or outright ignore the level of crap that LLMs spew and don’t think it’s a problem or one they can brush by. Think do they emphasize the need to check the LLMs work and really understand the space before using the LLM?

If they view LLMs as an end state not a temporary way point to somewhere else.

Hopefully, this helps you pick out the right voices from the sea of noise. There is a lot of noise and lots of charlatans—keep a careful eye out and your journey through the AI bubble is going to be much smoother.