Loading...
Video Topic
This video is about a home network rack setup and mentions products like a smart switch, Bitcoin lottery miner, Dell Optiplex micro PC, M1 Mac Mini, Dell R630 server, GPU and NAS, LattePanda, and Zima OS.
Caption
This is a follow-on from my video yesterday showing you my network rack. I had a couple of people ask to see the server rack and what I'm running to this is a slightly updated version from the last time I did it.
Performance Category
Above Average
Score
4.5/5
Shares: 4/5
Comments: 5/5
Retention: 5/5
Views: 5/5
Likes: 5/5
Followers: 5/5
Script: 2.5/5
Total Views
7835
Likes
320
Shares
7
Comments
24
Duration
3m 1s
Traffic Sources
For You
7,514
95.9% of views
Follow
133
1.7% of views
Personal Profile
102
1.3% of views
Search
55
0.7% of views
Direct Message
16
0.2% of views
Others
16
0.2% of views
Sound
0
0.0% of views
Performance vs Median
Views
+81.5%
Likes
+125.4%
Shares
+40.0%
Comments
+166.7%
For You Traffic
+154.4%
Profile Traffic
0%
Search Traffic
-17.6%
Audience Origin
Non-Followers
62.0%
4,858 views
Followers
38.0%
2,977 views
10.5% of followers reached
Audience Growth
New Followers
15
Performance vs Median
+200.0%
Transcript Available
Transcript
Yesterday, I showed you my network rack.
Let me give you a rack tour, uh, of my 2026 setup.
So coming from the network rack downstairs, we come into a switch.
This is a smart switch.
Every switch in the house is smart, although it's only one gigabit.
And we have our patch panel.
On top here, we have our Bitcoin lottery miner.
I got that for, like, there's better odds of winning three Bitcoin with this than playing the lottery.
And an access point for this, my office and, and work area.
Here, I have a Dell 3050 Optiplex micro, uh, mini PC.
This runs Coolifi, and I use it for, um, I wanna expose, uh, for work or, or any sort of public projects.
This is a M1 Mac Mini.
I've tried, but I'm trying to convert this into a local AI machine.
We'll see how it goes.
The Dell R630.
This is essentially the compute machine.
This runs Penpot, local Git service I did a thing on the other day.
Prickett for T- for video management.
So essentially, anything that requires high compute, 'cause this has got two CPUs in there and, like, 128 giga RAM, when RAM was cheap.
And X server stuff is always cheaper than gaming stuff.
But anything that requires high compute lives in this one.
This is the media server.
All of the R stack.
It has a GPU in there.
It has about 80 terabyte of drive.
Well, less, 'cause I've had a few fail, but this essentially is a NAS as well as all of the media server.
My project server, so anytime I do a setup or a project using Unraid, it's all on here, so all the time.
Here we have a, I'm yet to kind of get around to, I've got a ton of IoT stuff and it's kind of plugged in but it's not quite plugged in, and so I will get around to doing home assistant properly at some point.
So the other day I built this LattePanda running Zima OS.
Um, this is gonna come in and project server, so I'll set up stuff using Unraid and then the, the terminal when I need to, and also Zima OS.
Did I say Zima OS before, Caza?
Zima OS as well, so trying to give a little bit more stuff.
And for anyone watched my last one, I still haven't solved my UPS problem.
Really need to solve the UPS problem.
But at this point, we are pretty much, like, this is a finished, finished as much as home labbing will ever be finished, but this is a relatively finished rack.
I will need more drives.
I have to update drives.
I've had a couple of drives fail.
The only other thing is I, I need inference machine.
My plan is next month, apparently, LattePanda is releasing a thing called a MU.
It's, uh, it's like this compute, but the m- mu, mu, mu, the m.2
will take a PCI4 input, which means I can put it into an eGPU, grab an old power supply, and that's going to be dedicated AI inference running, uh, whatever I can fit into the 24 gig VRAM on the 4090.
I do a ton of stuff with AI, especially I wanna turn this into an agent device, so OpenCore, whatever, but something else that'll run agent, and I wanna use local inference because if you're doing a ton of stuff and you're using tokens from a third party, uh, it can get real expensive.
And that's it as well.
So this was just a cheap IKEA shelf unit, screw-in rack shelves, so
Let me give you a rack tour, uh, of my 2026 setup.
So coming from the network rack downstairs, we come into a switch.
This is a smart switch.
Every switch in the house is smart, although it's only one gigabit.
And we have our patch panel.
On top here, we have our Bitcoin lottery miner.
I got that for, like, there's better odds of winning three Bitcoin with this than playing the lottery.
And an access point for this, my office and, and work area.
Here, I have a Dell 3050 Optiplex micro, uh, mini PC.
This runs Coolifi, and I use it for, um, I wanna expose, uh, for work or, or any sort of public projects.
This is a M1 Mac Mini.
I've tried, but I'm trying to convert this into a local AI machine.
We'll see how it goes.
The Dell R630.
This is essentially the compute machine.
This runs Penpot, local Git service I did a thing on the other day.
Prickett for T- for video management.
So essentially, anything that requires high compute, 'cause this has got two CPUs in there and, like, 128 giga RAM, when RAM was cheap.
And X server stuff is always cheaper than gaming stuff.
But anything that requires high compute lives in this one.
This is the media server.
All of the R stack.
It has a GPU in there.
It has about 80 terabyte of drive.
Well, less, 'cause I've had a few fail, but this essentially is a NAS as well as all of the media server.
My project server, so anytime I do a setup or a project using Unraid, it's all on here, so all the time.
Here we have a, I'm yet to kind of get around to, I've got a ton of IoT stuff and it's kind of plugged in but it's not quite plugged in, and so I will get around to doing home assistant properly at some point.
So the other day I built this LattePanda running Zima OS.
Um, this is gonna come in and project server, so I'll set up stuff using Unraid and then the, the terminal when I need to, and also Zima OS.
Did I say Zima OS before, Caza?
Zima OS as well, so trying to give a little bit more stuff.
And for anyone watched my last one, I still haven't solved my UPS problem.
Really need to solve the UPS problem.
But at this point, we are pretty much, like, this is a finished, finished as much as home labbing will ever be finished, but this is a relatively finished rack.
I will need more drives.
I have to update drives.
I've had a couple of drives fail.
The only other thing is I, I need inference machine.
My plan is next month, apparently, LattePanda is releasing a thing called a MU.
It's, uh, it's like this compute, but the m- mu, mu, mu, the m.2
will take a PCI4 input, which means I can put it into an eGPU, grab an old power supply, and that's going to be dedicated AI inference running, uh, whatever I can fit into the 24 gig VRAM on the 4090.
I do a ton of stuff with AI, especially I wanna turn this into an agent device, so OpenCore, whatever, but something else that'll run agent, and I wanna use local inference because if you're doing a ton of stuff and you're using tokens from a third party, uh, it can get real expensive.
And that's it as well.
So this was just a cheap IKEA shelf unit, screw-in rack shelves, so
Scorecard Complete
Action & Impact Complete
Script Scorecard
Perceived Value
51/100
Compared to Average
Average
Category Scores (0-5)
Credibility
4/5
Progression
2/5
Specificity
4/5
Authenticity
4/5
Hook Clarity
1/5
Value Density
2/5
Audience Match
2/5
Cognitive Load
1/5
Payoff Clarity
1/5
Problem Specificity
1/5
Action & Impact (1/5)
topic introNEGATIVE~0s
"Yesterday, I showed you my network rack. Let me give you a rack tour, uh, of my 2026 setup."
RetentionRetention dropped from 1% to 0.7% within 2 seconds and continued to decrease rapidly.
Likes0.56% of the total likes were at 0s, but likes dropped sharply after.
RecommendationThe introduction was not captivating enough, leading to an early drop-off. Start with a visual or intriguing hook to capture interest immediately.
claimNEUTRAL~15s
"On top here, we have our Bitcoin lottery miner. I got that for, like, there's better odds of winning three Bitcoin with this than playing the lottery."
RetentionRetention held at 0.29% compared to a 0.03% average drop per 5s elsewhere with no significant change.
LikesNo noticeable spike in likes following this claim.
RecommendationWhile the claim is intriguing, it lacks visual engagement. Consider combining bold claims with dynamic visuals or demonstrations.
demonstrationNEUTRAL~40s
"This is a M1 Mac Mini. I've tried, but I'm trying to convert this into a local AI machine."
RetentionRetention held at 0.19%, which aligns with the trend at this time segment.
LikesNo spike in likes was observed following this statement.
RecommendationTo create more intrigue, pair demonstrations with practical usage scenarios or outcomes to captivate the audience.
claimNEGATIVE~120s
"But at this point, we are pretty much, like, this is a finished, finished as much as home labbing will ever be finished."
RetentionRetention fell to 0.05%, signalling a further drop compared to earlier sections.
LikesNo noticeable change in likes during this segment.
RecommendationEnsure strong engagement towards the end by suggesting future improvements or unexpected value this setup brings.
otherNEUTRAL~135s
"My plan is next month, apparently, LattePanda is releasing a thing called a MU."
RetentionRetention remained stable at 0.04%, which is consistent with later parts of the video.
LikesNo notable spike in likes following this statement.
RecommendationIntroduce future content or product reveals towards the end to keep viewers intrigued about upcoming developments.
Actionable Tips (1/10)
hook clarityScore: 1/5
Reasoning
The opening does not quickly answer why a viewer should stay unless they already watched a prior video. In the first 3 seconds, retention falls from 100% to 58%, and by 5 seconds it is down to 44%, which is a severe early drop. That aligns with a weak hook: 'Yesterday, I showed you my network rack. Let me give you a rack tour...' is continuation-based and low-stakes rather than giving a clear promise or surprising payoff.
How to Improve
Open with a concrete payoff and novelty in line 1, such as: 'This $X home lab runs my AI, NAS, media stack, and Bitcoin miner—here's exactly what each machine does.' That gives a reason to keep watching even if someone did not see the previous post.
Relevant Extract
"“Yesterday, I showed you my network rack. Let me give you a rack tour...”"
audience matchScore: 2/5
Reasoning
The script eventually signals homelab / networking / self-hosting viewers, but it does not filter for them immediately. Early retention dropping from 100% at 0s to 70% at 2s and 51% at 4s suggests many viewers did not know this was for them quickly enough. The niche becomes clearer only once terms like 'smart switch' and 'patch panel' appear.
How to Improve
Name the audience in the first sentence: 'If you're into homelabs, self-hosting, or local AI, here's my 2026 rack setup.' That would help the right viewers self-select before they swipe.
Relevant Extract
"“So coming from the network rack downstairs, we come into a switch. This is a smart switch.”"
problem specificityScore: 1/5
Reasoning
This is mostly a descriptive tour, not a clearly framed problem-led video. There are a few real pains later—UPS issue, failed drives, token costs—but they are introduced deep into the video rather than anchoring the narrative. With a 181.9-second runtime and first-15-second retention sliding to 30%, the lack of a central tension likely reduced motivation to stay. Comments (24) and shares (7) are also modest relative to 7,835 views, which suggests the topic did not spark a strong problem/solution response.
How to Improve
Choose one main tension to structure the tour around, e.g. 'I wanted one rack that could handle media, local AI, and self-hosting without cloud costs.' Then make each device a step in solving that problem.
Relevant Extract
"“I still haven't solved my UPS problem... I've had a couple of drives fail... if you're doing a ton of stuff and you're using tokens from a third party, it can get real expensive.”"
payoff clarityScore: 1/5
Reasoning
The viewer is told they are getting a 'rack tour,' but not what insight, lesson, or result they will walk away with. That weak payoff is reflected in the sharp first-10-second decline from 100% to 33%. A plain tour can work for an interested niche, but the script does not promise practical takeaways like costs, architecture decisions, or why each machine exists.
How to Improve
State the payoff explicitly: 'By the end, you'll know how I split AI, storage, and compute across this rack and what I'd change if I rebuilt it.' This turns a tour into a useful blueprint.
Relevant Extract
"“Let me give you a rack tour, uh, of my 2026 setup.”"
value densityScore: 2/5
Reasoning
There is useful detail throughout, but the ratio of insight to filler is weak, especially early. The script includes throat-clearing, repeated qualifiers, and side comments ('uh', 'essentially', 'kind of', self-corrections), which is costly in a 3-minute video where retention is already down to 30% by 13-14 seconds. Engagement is decent on likes (320 total, about 4.1% like rate), indicating some viewers valued the topic, but the low share count (7) suggests the information density was not packaged tightly enough to feel highly shareable.
How to Improve
Trim filler and make each sentence either identify a component, explain what it does, or explain why it matters. For example, replace meandering setup with a consistent pattern: 'Device -> role -> reason it lives here.'
Relevant Extract
"“This is essentially the compute machine... So essentially, anything that requires high compute...”"
specificityScore: 4/5
Reasoning
This is one of the strongest areas. The script uses concrete nouns, brands, capacities, and workloads: Dell 3050 Optiplex, M1 Mac Mini, Dell R630, 128GB RAM, 80TB, Unraid, Zima OS, 4090, 24GB VRAM. That kind of specificity supports authority and likely contributed to the decent like count from a niche audience. The score is not a 5 because some terms are unclear/mis-spoken and some references may be too context-dependent without explanation.
How to Improve
Keep the concrete nouns, but pair them with plain-English labels so less technical viewers can follow: 'Dell R630—my main compute server,' 'Unraid project box,' '4090 inference machine for local AI.'
Relevant Extract
"“This is a M1 Mac Mini... The Dell R630... 128 giga RAM... It has about 80 terabyte of drive... 24 gig VRAM on the 4090.”"
credibilityScore: 4/5
Reasoning
The script sounds like it comes from real hands-on experience. It includes real constraints, hardware trade-offs, failed drives, and future upgrade plans. That kind of grounded detail builds trust. The like rate is solid for the view count, which supports that viewers who stayed found the creator credible. It is not a 5 because the delivery includes some unclear phrases and unfinished thoughts that slightly weaken perceived precision.
How to Improve
Add one or two explicit decision rationales early, such as cost, power, noise, or performance trade-offs. That would make the expertise even more obvious to new viewers quickly.
Relevant Extract
"“X server stuff is always cheaper than gaming stuff... I've had a few fail... using tokens from a third party, uh, it can get real expensive.”"
authenticityScore: 4/5
Reasoning
The script feels human and unscripted, with real frustrations and imperfections. Lines about the UPS problem, failed drives, and trying to get around to Home Assistant make it sound honest rather than over-polished. That authenticity likely helped convert some niche viewers into likes and comments despite weak retention. It is not a 5 only because the roughness sometimes drifts into confusion rather than charming honesty.
How to Improve
Preserve the candid tone, but tighten wording so the honesty stays while the meaning stays clear. A lightly outlined talking structure would keep the authentic feel without so many verbal stumbles.
Relevant Extract
"“I still haven't solved my UPS problem. Really need to solve the UPS problem.”"
cognitive loadScore: 1/5
Reasoning
The spoken delivery is hard to parse in real time. Many sentences have stacked clauses, self-corrections, vague referents, and repeated fillers. That likely contributed heavily to the retention crash from 89% at 1s to 58% at 3s, then 41% at 6s and 33% at 10s. Once viewers have to work to decode what a machine is or what role it serves, they swipe. The likes-by-second data also front-loads heavily at 0s and falls to near-zero after a few seconds, which suggests the opening lines did not sustain easy comprehension.
How to Improve
Use shorter sentence templates: 'This is [device]. It does [job]. I chose it because [reason].' Cut filler words and self-corrections in edit or with retakes. For technical terms, say the simple label first, then the brand/model.
Relevant Extract
"“Did I say Zima OS before, Caza? Zima OS as well, so trying to give a little bit more stuff.”"
progressionScore: 2/5
Reasoning
The video generally moves from top to bottom through the rack, but the narrative progression is weak. It does not build toward a clear conclusion, insight, or transformation; instead it feels like an inventory with side branches. Given the long duration of 181.9 seconds, strong progression was especially important, yet first-15-second retention settles around 30%, indicating viewers did not feel enough forward momentum. Shares are also very low (7), which often happens when a video lacks a strong distilled takeaway.
How to Improve
Organize the tour into a simple arc: 'networking -> compute -> storage -> AI plans -> one lesson learned.' Add mini-transition lines that signal progress and reward: 'Now for the server that does the heavy lifting,' 'Here's the part I'd rebuild differently,' 'And here's the upgrade that fixes my biggest cost problem.'
Relevant Extract
"“This is the media server... My project server... Here we have a... IoT stuff... So the other day I built this LattePanda...”"
neil allan269What is the bitcoin lottery thing you have ?
1
Creator
Apr 8, 2026its called a lottery miner, it just solo mines bitcoin and its got a 1 in 2000000 chance of just hashing a coin, which is better than paying for lottery tickets. its novelty really, but who knows.
1
stevoau7Cool setup Benjamin 👍 storage is always and issue now we have to sell kidneys🤪
Olliemay I ask what specs you are running in your jellyfin server as I am making now own on a mini pc but would like to move to a big one so I can put a GPU in for the transcoding and I'm not sure what the specs I would need
1
Creator
Apr 8, 2026oh, the gpu in this is o my an 8gb thing, nothing special at all.
1
Ollie8gb is that all is needed okay I was looking way to high then as I only wanted to do 1080p it's the transcoding that's killing my mini pc when more then 2 people are using it this is so helpful to know thank you so much 😊
1
Creator
Apr 8, 2026yeah 8gb gpu is more than enough to do multiple 4k streams.
1
Mach AttackHoly terabytes batman! 800TB is nuts, are you hosting the Netflix catalogue locally? 😂
1
Creator
Apr 6, 2026800.. I'd love that, no 80 only!
redmasterwhat is your power consumptions pr month Just for this rack?
2
Creator
Apr 6, 2026honestly not that much. 65kwh over the last 30 days, for everything In that rack.
1
redmasterwoow that Just $4 in power consumptions. you have reading meter? how about noise? or you have replaced with quiet fans?
1
Creator
Apr 7, 2026I have smart power plugs from the wall, and track usage of power, so yeah, super cheap. I do have an account run in the room, and that adds some cost (but I'm in Thailand so heat is an issue). fans are ok, it does make some noise but honestly never too much, but if you notice that audio is a little odd, the background sound reduction was working hard.
BEYANKI 🇦🇺🦘❤️🏆💪Mate that’s a powerful set up. I’m trying to find ways to get mine done but I like your setup 💪💯
3
Creator
Apr 6, 2026tha is, it's all second hand, Facebook marketplace and whatever I can scrape together really.
2
BEYANKI 🇦🇺🦘❤️🏆💪Yeah that’s what I have to do start looking around for parts. The rack is impressive I saw one but they wanted 899 for it I thought I can make that. I never thought of repurposing my Mac mini which is a really good idea.
Creator
Apr 6, 2026do what you can with what you've got. you can run docker desktop on the Mac and remote I to it and its a server.
NicholasRagenice rack, babe 😘
1
Creator
Apr 6, 2026lol, thanks
Creator
Apr 6, 2026not sure why but the edit just skipped words for some reason...
user1947284628472May I ask how come many Aussies move to Philippines?
1
Creator
Apr 6, 2026Not sure, I'm I Bangkok and have been the last 15 years.
user1947284628472Oh that’s right. I see so many Aussies living in thailand as well.
1
Creator
Apr 6, 2026its easier to make a life in another country where you need to learn a whole other language and culture than make a life in Australia I belive.
user1947284628472Oh that’s interesting
1
Sebastian KlingkHow is the internet connection to outside of Thailand like Europe etc? On Phuket it is often very laggy
1
Creator
Apr 6, 2026I don't really have an issue. I use my own upstream dns, and when I have tunneled service I use pangolin on a Thailand based vps.
Second-by-Second Performance
Total viewers and likes aligned with spoken words.