The M chip is truly energy efficiency, I’m already amazed my Mac Studio M1 Max only draw about 15 watt during idle. I wonder for the Mac Mini full power wattage.
Now we need some MacOS feature to build a cluster of these so they can be treated as "one" Mac. This could revolutionize many things honestly... compute, resiliency and redundancy.
There already are things to cluster up even different kind of computers/OS. Microk8s or any other kubernetes clustering tool will allow you to gather the memory and compute of all the systems joined together
2 base M4 minis cost a dollar less than a M4 mini with double the base RAM and storage. Transparent, end-user-friendly clustering over TB4 would be an amazing feature.
Since M1 came out it only got better on Performance you get for that price, my 1k bucks pc got demolished by an 200 bucks cheaper and more power efficient mac mini
@@rebs4jezuseveryone is right. It’s a hardware issue because a majority of systems don’t have 2+ Thunderbolt ports. It’s a software issue because applications for creative professionals don’t support clustering BUT - for those of us that already run distributed workloads, the Mac mini has no competition. I used to keep 4 2U rack servers under my bed for my clusters, now more compute power takes up less space than a mousepad on a desk and doesn’t sound like a Boeing.
@ Partially agree when looking at the TCP/IP model transfer/connection is not only done through the physicallayer like thunderbolt/ USB C. There’s multiple ways to do this for example ethernet 10. The hardware that is part of the machine supports the concept, but the software is non-existent as of now.
What is the Sense of building such a M4 cluster? Will it also have a positive effect if I build such a cluster and use it for Gaming, Video editing, LLM, or other things
LLM use cases would be a little slow, since the programs that support the Apple MLKit don't support distributed inference. Maybe vLLM has changed that lately, but not in my memory.
They are connected with included Power Supply cable as you can easily see in the video. I don't think they can be powered through thunderbolt like a Macbook laptop.
These Mac Minis can be an incredible tool. You can potentially put these in a server and have them connect to any TV in your home, providing you with full computer system.
When the video about how much was the consumption when using them to the maximum? How will the cluster vs the RTX 4090 compare?. Incredible videos. Thank you!!
Well, Windows on an x86 processor (AMD or Sauna... sorry, I meant Intel) can run most professional programs, while Macs are not always supported. Windows mini PCs have their place in the world where compatibility is sometimes more important than price, performance, or efficiency. Price-to-performance matters none if you cannot boot to your most important work app(s).
I dont get it does this now work like one pc and combines the cpu and cpu by 5x! Can someone explain, ive also heard the term mac farms, are those different than this?
@saatvik7035 that's just using mac minis or studios daisy chained to act as nodes in a cluster for computing. You might have seen little towers of msc minis vabled with with thunderbolt to do machine learning which apple silicon is known to do well for way less than nvidia and others while also being a whole comp rather than just a gpu and taking less power. 5 of these compute 3 times better than 2 nvidia 4090s but at half the cost while also being able to do many other things and are not gonna break down or make noise and heat.
I wonder if the software to connect all the Macs together to do a process load for heavy tasks. Probably connect via thunderbolt 5 or Ethernet cable. Theres been server rack Macs in the past for server use not for processing heavy tasks though.
Bro please be quick I love to see your videos i was inspired to find my career in machine learning only because of you your videos inspired me and made me see my field as I was confused before which path to choose love you bro big fan ❤❤
@@alanwang8866 that's what kinda let me down about this video. still, can do lots with that bandwidth. edit: not to mention 80gb unified memory. not sure how much of thst could be utilised in the cluster, but hopefully something close to 65gb or more vram.
@@ZeerakImran better to ebay an older ai server and fill it with 24gb gpus if you want vram. M40s are $100 and still very useful for big models. Or you can get newer cards if budget is on the 5x M4 scale+. 3090s are $600 range which you can put in a cheap mining rack.
@Mooooov0815 😊😊😀😀😀😊😊😀😀😀😊😊😊😀😀😀☺☺☺lol it would be screaming fast as beck for software optimised for it. not at all distributed computing like servers it would actually be faster than a single computer that's the whole point. Like 2 media engines are better than one. Super computers felons must so computer are fast becaudevthey combine and distribute the load between more cores for cpu GPU npu memory media engines by taking the who process and splitting it between multiple workers instead of one fast worker it becomes multithreaded multi core this speeds up work flow for demanding tasks and allows for much more compute power. Sure for some single threaded quick tasks it would be slower. I'm sure you know if the software is written to take advantage of multiple gpus cpus media engines you will get more fps faster rendering. encoding etc Even with a simple 10g Ethernet cable. Now with tb5 it will be e en better that's basically pci5. All the data does not have to be on e sry machine the master machine splits up the workload to each machine so you have more cires ram gpu working on smaller parts finishing much faster like computers downloading with software that allows it to down load from two or more broadband accounts instead of one all put together at the end downloads instantly become twice as fast than from one broadband account even if the second account is slower. Surely you know thiswould be fast as beck each computer basically basically doubling performance fur multithreading test. Meaning you could add a Mac mini to your iPad pro or Mac air or pro or studio or even iPhone and instantly increase the gpu gpu boy cores and ram for software optimised for this basically now that some music software 3d cad scientific encoding rendering servers blender etc. However this would slow down dales of high end machines as you could simply add 2 mini pros power together instead of getting say an ultra and you can then add any apple silicon mac or tablet when more compute power is needed which would be cheaper than getting the premium machines. Sat a mini pro and a MacBook pro. That's up to 28 cores cpu 40 cores gpu 2 media engines double the ram when you need it for as a games complex 3d audio video. Or have that apart for mainstream single core lighter computing.
Hi Alex, I love your videos! I'm starting to work on a rack for Mac Mini M4s. I really like your tower setup. Did you use 3D printing for the supports? If so, could you let me know where to find the print file?
Thats just how ARM CPUs run, they can be far more effiecient since they dont idle the same way x86 CPUs do. Under heavy load the power draw will be high still though
at this point, all that's missing is the installation of the various Linux systems in place of Mac OS and making a cluster that gives meaning to this tower of ARM mini-PCs (closer to medium-high range smartphones)... will there be any?!?...given how much they cost. The amortization of the costs of a project "also" includes the energy consumption for the useful life cycle of the "nodes", but also the cost of hardware and the man-time for the system engineer who must distort systems born and sold to do something else with them. I appreciate the enthusiasm though
Mine draws 6 watts idling with everything set up the way I need it (networking etc); and even switched on with my 32” monitor combined they only draw +/- 50W! I can play emulated PS games at 1/6 the wattage of my PS5; I’m honestly thinking of selling the console as it’s just too expensive for long gaming sessions.
The highest draw i got with mine was a testing a new game in early access with 2 hx370 connecected pulled 70 watts total in thone and liberty. I dont use them for that i use for a host server. Average 5-7 idle average 25-40 with load. I want a rack like this eventually but price is really high atm.
Alex , please make tests with exocluster and lets project how would 20 mac minis in cluster do??! this is super interesting,,,seams as possibly better buy than 3090 cards..
With thunderbolt 5 make a cluster.
What does this cluster do for the owner. Is there software that can distribute it's load across all processors?
Yes I wonder if there could be like lightroom and dhare the export un many cpus
The network over Thunderbolt was added in version 2.
is there any something like "tb5 switch"?
M4 doesn’t have thunderbolt 5. You need Pro or Max chip.
I approve this. My 16 core Ryzen idling with the 4080 GPU draws like over 120 watts...it's no wonder my electric bill went up
Yes, but when you need to do something, you do 10 times faster
@@cfzerooo like what? Most applications still favor 1-2 cores at most
@@cfzerooo that’s nonsense - unless you tell us what use case.
@@macbugde 3d rendering
@@cfzerooo you realize the combined compute power of five M4 Mac minis absolutely demolishes a 16 core machine with a 4080?
The M chip is truly energy efficiency, I’m already amazed my Mac Studio M1 Max only draw about 15 watt during idle. I wonder for the Mac Mini full power wattage.
@SiriPod597 It's the ARM architecture.
Now we need some MacOS feature to build a cluster of these so they can be treated as "one" Mac. This could revolutionize many things honestly... compute, resiliency and redundancy.
There already are things to cluster up even different kind of computers/OS. Microk8s or any other kubernetes clustering tool will allow you to gather the memory and compute of all the systems joined together
2 base M4 minis cost a dollar less than a M4 mini with double the base RAM and storage. Transparent, end-user-friendly clustering over TB4 would be an amazing feature.
@@jeslinmx22 Incredible! Those upgrade prices are insane I know.
@@jeslinmx22 Non-nonsensical upgrade prices would be a truly amazing feature.
Except macos is a toy os, rendering it useless
If you cluster them together, are the new mac mini the best vaule for money in terms of computing power?
Since M1 came out it only got better on Performance you get for that price, my 1k bucks pc got demolished by an 200 bucks cheaper and more power efficient mac mini
Depends but most software does not support distributed system like that since they were not design for that.
@@scientist30 This is more of a software issue than a hardware issue no?
@@rebs4jezuseveryone is right.
It’s a hardware issue because a majority of systems don’t have 2+ Thunderbolt ports.
It’s a software issue because applications for creative professionals don’t support clustering
BUT - for those of us that already run distributed workloads, the Mac mini has no competition. I used to keep 4 2U rack servers under my bed for my clusters, now more compute power takes up less space than a mousepad on a desk and doesn’t sound like a Boeing.
@ Partially agree when looking at the TCP/IP model transfer/connection is not only done through the physicallayer like thunderbolt/ USB C. There’s multiple ways to do this for example ethernet 10.
The hardware that is part of the machine supports the concept, but the software is non-existent as of now.
I wonder whether 4 minis are a better deal than one speced out mini for ML? 🤔
What is the Sense of building such a M4 cluster? Will it also have a positive effect if I build such a cluster and use it for Gaming, Video editing, LLM, or other things
for the money of that you can get a pc with a 4090
LLM use cases would be a little slow, since the programs that support the Apple MLKit don't support distributed inference. Maybe vLLM has changed that lately, but not in my memory.
Can someone explain how the individual Mac Minis are connected to each other and what is required?
Awesome! Are they connected with thunderbolt?
most likely, like the other macbooks
They are connected with included Power Supply cable as you can easily see in the video. I don't think they can be powered through thunderbolt like a Macbook laptop.
@@helloukw I think he means like instead using a network wire UTP. Cause he previously connected the laptops through thunderbolt wires as LAN.
@@helloukw 🤦♂️
Are they all on independent monitors or is there some way you combine all the computing power into 1?
Alex I really love your videos and creative way of thinking 🎉
These Mac Minis can be an incredible tool. You can potentially put these in a server and have them connect to any TV in your home, providing you with full computer system.
you can do it with every computer, did you know it?
so? how much do they draw when "pushed to the max"? Or are you only gonna show them do NOTHING? Because my pc draws 0 watts when I turn it off.
Are you able to get them to work in parallel?
can we use them as one gaming beast?
One at full tilt draws only about 40w. It's pretty awesome.
I’m confused why the mini’s need its own power port if it can be through TB port?
How much would a setup like that cost? And what would require the use of 5 Mac Minis as opposed to 2 or so Mac Studios?
Full CPU load power consumption?
If you're curious you should research the story behind arm it will explain this.
Imagine if you can play game run demanding stuff on it, all run at the same time, share the loads together.
Amazing
When the video about how much was the consumption when using them to the maximum? How will the cluster vs the RTX 4090 compare?. Incredible videos. Thank you!!
What are you using as a rack?
It looks like 3D printed plastic.
printed pla
@@AZisk could you please share a 3d model?
@@anonymos3671 I just used this one from printables, and scaled it in the slicer to fit the m4: www.printables.com/model/36855-mac-mini-rack
waiting for the full video comparrison on how m4 perform in a llm cluster using exo.
It's because there is only the down on that is used, all the others are not connected to a screen ...
Push to the max 😮
how can i buy that rack?
How about in a deep sleep (like put to sleep and check in the morning, the consumption?)
and how to use it?
with mini pc on 5825u it'll take same electricity, but there you can do any linux on that cluster, but with macs🤷🏻♂️
Welp, this video was recommended to me and I’m hooked. Subscribed.😊❤
What is the usage case to cluster them like this?
Inter connect then with the usb ports ....
and people will still say ‘buy a windows mini pc’ 😭😭
@@blackxfiied_ which i also find funny as a windows user 💀😭
To be fair some Windows mini pcs can idle at a decently low wattage, that just all goes out the windows the minute they have a load
Or just a M-ATX build. Will be way better for gaming, but these are unbeat in terms of workload at this price point
@@mrdinosaurproductions9507 MATX is also nowhere near the same size, or is ITX. Those things are huge compared to actual mini pcs.
Well, Windows on an x86 processor (AMD or Sauna... sorry, I meant Intel) can run most professional programs, while Macs are not always supported. Windows mini PCs have their place in the world where compatibility is sometimes more important than price, performance, or efficiency. Price-to-performance matters none if you cannot boot to your most important work app(s).
I dont get it does this now work like one pc and combines the cpu and cpu by 5x! Can someone explain, ive also heard the term mac farms, are those different than this?
@saatvik7035 that's just using mac minis or studios daisy chained to act as nodes in a cluster for computing. You might have seen little towers of msc minis vabled with with thunderbolt to do machine learning which apple silicon is known to do well for way less than nvidia and others while also being a whole comp rather than just a gpu and taking less power. 5 of these compute 3 times better than 2 nvidia 4090s but at half the cost while also being able to do many other things and are not gonna break down or make noise and heat.
so they're all working as a one system or as multiple systems ?
Can't wait for the exo benchmarks on these and which models did you get.
It’s better if we can have a video about this cluster
They still have Mac mini new versions?
It did not seem to help to run larger LLM?
I knew he would have done that! Can’t wait for the MacMini 4 cluster video!
can you look into running multiple mac mini m4 in a cluster? using exo for example?
Any setbacks like repairability or issues with upgradability? Maybe a manufacturing defect?
It’s a apple product, upgradability practically doesn’t exist
Can you put exo on them and see hoe they compare to your previous mix and match ollama cluster?
I don't think I've ever seen a Mac cluster outside of a specialized data center before. What are you building with this? Local distributed LLMs?
can you power a Mac mini from a monitor\displays 90 watt usbc out, I know the MacBook Pro can be
Pro or base version?
may be pro
30w idle is what my 2012 i7 Mac mini does at idle. Can’t wait for that Linux distribution to fully work and then I’ll get an m4 mini
I wonder if the software to connect all the Macs together to do a process load for heavy tasks. Probably connect via thunderbolt 5 or Ethernet cable. Theres been server rack Macs in the past for server use not for processing heavy tasks though.
what are you running? do you have cluster software or just use them individually via kvm or something?
iE purpose?
Bro please be quick I love to see your videos i was inspired to find my career in machine learning only because of you your videos inspired me and made me see my field as I was confused before which path to choose love you bro big fan ❤❤
My base model M4 iMac draws about 2.8w idle.. x5 would be about 15w~.. are these M4 Pros?
what is the point of using all these mac minis together? can you run any apps at the same time?
May I ask why you would need so many imac minis?
HOW much watts do they use when maxed out????? Please put long length video in desc!!
1 m4 mac mini uses around 40 to 60 watts maxxed out
@@XashA12Musk Thanks! but what chip though? the m4 base model the m4 pro base model or the m4 pro upgraded?
@@Timely-ud4rm M4
You need a TB5 switch
That would be a epic cluster
6GB/s interconnect for cheap
base M4 Mini is only TB4 compatible bro...
@@alanwang8866 that's what kinda let me down about this video. still, can do lots with that bandwidth. edit: not to mention 80gb unified memory. not sure how much of thst could be utilised in the cluster, but hopefully something close to 65gb or more vram.
@@ZeerakImran better to ebay an older ai server and fill it with 24gb gpus if you want vram. M40s are $100 and still very useful for big models. Or you can get newer cards if budget is on the 5x M4 scale+. 3090s are $600 range which you can put in a cheap mining rack.
on max load it's like 40 watts each
You should be able to combine media engines CPU GPU npu cores bram SSD by connecting thunderbolt for distributed computing, but no
Would be slow as heck lol
@Mooooov0815 😊😊😀😀😀😊😊😀😀😀😊😊😊😀😀😀☺☺☺lol it would be screaming fast as beck for software optimised for it.
not at all distributed computing like servers it would actually be faster than a single computer that's the whole point.
Like 2 media engines are better than one.
Super computers felons must so computer are fast becaudevthey combine and distribute the load between more cores for cpu GPU npu memory media engines by taking the who process and splitting it between multiple workers instead of one fast worker it becomes multithreaded multi core this speeds up work flow for demanding tasks and allows for much more compute power.
Sure for some single threaded quick tasks it would be slower.
I'm sure you know if the software is written to take advantage of multiple gpus cpus media engines you will get more fps faster rendering.
encoding etc
Even with a simple 10g Ethernet cable.
Now with tb5 it will be e en better that's basically pci5.
All the data does not have to be on e sry machine the master machine splits up the workload to each machine so you have more cires ram gpu working on smaller parts finishing much faster like computers downloading with software that allows it to down load from two or more broadband accounts instead of one all put together at the end downloads instantly become twice as fast than from one broadband account even if the second account is slower.
Surely you know thiswould be fast as beck each computer basically basically doubling performance fur multithreading test.
Meaning you could add a Mac mini to your iPad pro or Mac air or pro or studio or even iPhone and instantly increase the gpu gpu boy cores and ram for software optimised for this basically now that some music software 3d cad scientific encoding rendering servers blender etc.
However this would slow down dales of high end machines as you could simply add 2 mini pros power together instead of getting say an ultra and you can then add any apple silicon mac or tablet when more compute power is needed which would be cheaper than getting the premium machines.
Sat a mini pro and a MacBook pro.
That's up to 28 cores cpu 40 cores gpu 2 media engines double the ram when you need it for as a games complex 3d audio video.
Or have that apart for mainstream single core lighter computing.
30w in idle, what it draws when its all running in full perf mode?
Can we make a cluster like this for after effects or blender render ?
You can combine performance of these?
Can you use these to game?
Baby, baby, baby! This has got me fired up for whatever video you have coming up!! And to think I only bought one of them haha
And I bought none 😂
Hi Alex, I love your videos! I'm starting to work on a rack for Mac Mini M4s. I really like your tower setup. Did you use 3D printing for the supports? If so, could you let me know where to find the print file?
he wont give it out. either he stole it or wants to make money from it. time will tell
Why are you running a rack of Mac minis?
Can we start using them for Bitcoin mining?
oh no, but you could use them for a cluster and possibly rent vps
@@reatcas no ASICS are for bitcoin mining you would never make any money mining on a computer now sadly.
What do you do with a rack of these?
Question is what cant you do
You can't triple a game
@@bosdagunslinger5479 No one cares about games in the real world. Time to grow up.
@@AKagNArun a 64-bit lab like you used to with the intel macs :-)
I like the part where you told us how much power it drew under load
Is the rack 3d printed? can you share the link for rack purchase or stl file if 3d printed?
a cluster of 3-5 mac mini m4 pros with 64gb ram and 10gbps networking used with exo...
How did you turn them all on!!!??? 🤣
Thats just how ARM CPUs run, they can be far more effiecient since they dont idle the same way x86 CPUs do. Under heavy load the power draw will be high still though
Where is the full load!!! The suspense!
Please check when all of them are asleep (deep sleep after a few hours I guess) or hibernating (deep sleep mode)
Can someone explain to a pure beginner
what am I looking at
what's the advantage of this
😊 thanks in advantage
THIS COULD BE AN AMAZING NAS
How do I use this to play Steam games as if they were on Windows?
how can buy one of the rack ?
How can I buy one of the racks?
nuc does 5w in iddle to.
Wow thing eats power when idling?? Weow didn't see that one coming
You had my attention, NOW, you have my interest 🤨🤔🧐🙂
at this point, all that's missing is the installation of the various Linux systems in place of Mac OS and making a cluster that gives meaning to this tower of ARM mini-PCs (closer to medium-high range smartphones)... will there be any?!?...given how much they cost.
The amortization of the costs of a project "also" includes the energy consumption for the useful life cycle of the "nodes", but also the cost of hardware and the man-time for the system engineer who must distort systems born and sold to do something else with them.
I appreciate the enthusiasm though
My old daylight bulb was eating more 😮
Are you planning to do some cluster computing or something
the shitty M chip isn't good for much, but it uses very little power
Se importar com energia pra que? Só o valor desses M4 já pagaria muitas contas de luz...
Mine draws 6 watts idling with everything set up the way I need it (networking etc); and even switched on with my 32” monitor combined they only draw +/- 50W! I can play emulated PS games at 1/6 the wattage of my PS5; I’m honestly thinking of selling the console as it’s just too expensive for long gaming sessions.
The highest draw i got with mine was a testing a new game in early access with 2 hx370 connecected pulled 70 watts total in thone and liberty. I dont use them for that i use for a host server. Average 5-7 idle average 25-40 with load. I want a rack like this eventually but price is really high atm.
5 Mac 30 watt??? Omg❤
You should print a cheese grater chasis for it. And maybe splurge on some $17 wheels.
Nice. Looking forward to what you can come up with this monster
The mac only outperforms in singlethread, I have 2 hx 370 stacked 16tb total. Draws almost no power, insane performance with almost no power cost.
but how will they draw if they don't even haave hands?
Proxmox dream machine. I hope we see apple silicon development for this soon.
Why do you need 5 units?
Can you put proxmox on them?
Can we use mac minis 4 for render farm?
Do they all have 10Gb ethernet?
Is content coming soon on the m4s or..?
Alex , please make tests with exocluster and lets project how would 20 mac minis in cluster do??! this is super interesting,,,seams as possibly better buy than 3090 cards..
The only issue is powering on the device, so never turn it off could be the solution. Use the button only once.
Okay, you have more than enough Macs. Outa send me one of them my dude! 😊
they draw so much that these pc became artists