---
title: 'I built a 2500W LLM monster... it DESTROYS EVERYTHING'
source: 'https://youtube.com/watch?v=GyjOOoboT1c'
video_id: 'GyjOOoboT1c'
date: 2026-06-28
duration_sec: 0
---

# I built a 2500W LLM monster... it DESTROYS EVERYTHING

> Source: [I built a 2500W LLM monster... it DESTROYS EVERYTHING](https://youtube.com/watch?v=GyjOOoboT1c)

## Summary

The video follows Alex as he builds an ultra-high-end AI workstation with a Threadripper CPU and dual RTX Pro 6000 GPUs. He visits MicroCenter to select a case, power supplies, fans, and a UPS, then assembles the system and tests it with a massive 235-billion-parameter LLM. The build is designed to handle demanding AI workloads that no single consumer GPU can manage.

### Key Points

- **Need for upgrade** [0:00] — Alex's existing AI rig can't fit a second RTX Pro 6000, so he plans to expand with a new build.
- **Visit to MicroCenter** [0:19] — Alex meets Dan at MicroCenter, bringing his 2500W power supply and explaining the specs: Threadripper TRX board, 9970X CPU, two RTX Pro 6000 GPUs.
- **Dual PSU solution** [1:18] — A single 2500W unit needs a special high-voltage plug; US outlets max at 1800W, so they lean toward a dual PSU setup (CPU+one GPU on one PSU, second GPU on another).
- **Choosing a case** [2:40] — Dan notes that dual PSUs require a full tower case; they explore options like a limited-edition Doom case and one with an OLED screen.
- **Power supply recommendations** [3:34] — Dan suggests a Taichi 1650W PSU for one GPU and an ROG Loki SFXL 1000W for the second, discussing efficiency ratings from bronze to titanium (95% efficiency for titanium).
- **Cost comparison** [5:23] — Dual PSU costs around $680 (1300W + 1200W) vs $830 for two 1650W units; they choose the cheaper option.
- **Synchronization adapter** [6:34] — An $11 adapter connects the two PSUs via motherboard cables to sync them.
- **UPS requirement** [6:57] — A massive 1500W+ UPS (about $800) is needed to protect against power blips during LLM workflows.
- **Fans selection** [8:15] — Noctua fans are chosen for quiet, efficient airflow; the case allows assembling fan arrays outside and sliding them in.
- **Build completion and test** [10:58] — Alex assembles the system with two RTX Pro 6000s and runs Qwen 3 235B model (142GB on disk) using Ollama, achieving 68 tokens per second with both GPUs active.

### Conclusion

Alex successfully builds a powerhouse AI workstation capable of running the largest open-source LLMs, proving that multiple RTX Pro 6000s are necessary for models beyond consumer GPU memory limits. The build demonstrates practical decisions around power, cooling, and component selection for extreme AI hardware.

## Transcript

This little box was my all-in-one AI rig
until I decided I needed another RTX Pro
6000, and there's absolutely no way this
is fitting in here. So, today I'm
planning to expand operations. This
build is going to take a lot, so I'm
going to go over here to MicroEnter, my
favorite spot to go shopping for
computer parts and talk to my buddy Dan.
He'll hook us up.
>> Hey Dan, what's up?
>> Hey, Alex. Welcome back to Good to see
you too.
>> I got a gift for you. Not a gift, it's
it's mine. I'm keeping it. But this is
the new power supply that I have. And I
don't think this is going to work with
my setup. So, we're going to look at
some power supplies, a case, and some
fans for this Thread Ripper build.
>> Okay. Well, like what exactly are we
doing?
>> I've got a Thread Ripper TRX board, a
Monster Thread Ripper 9970X, and two RTX
Pro 6000 GPUs waiting for a case. That
combo can pull an absurd amount of
power, which is why I bought this 2500 W
power supply. Now I'm at MicroEnter so
we can walk around and see which case
can actually fit this brick.
>> Feel that. Dang. It's almost as heavy as
my graphics card. It's crazy. You do
your push-ups today?
>> Yeah. Well, not yet. I guess we're going
to have to today, right?
>> This thing is crazy.
We basically have two options. split the
load so one power supply can run the CPU
plus one GPU and a second power supply
feeds the other GPU or try to shove
everything into this 2500 watt monster
to use all that power the big unit needs
a special high voltage plug and a normal
US outlet tops out around 1,800 watts so
before I accidentally discover the limit
of my office wiring we're leaning
towards the dual PSU setup right now
>> yeah well we're doing two GPUs knowing
you they're probably something very fun
pro 6000s All right. Pro 6000s. Two Pro
6000s and me and my measly 5090.
>> With that much hardware, small cases are
officially cancelled. Dan tells me I'm
stuck shopping in the skyscraper sized
full tower section if I want dual RTX
6000s and giant power supplies. What is
up with this? Oh, yeah. So, that one's
really cool. Bethesda teamed up with uh
Haven to create this case.
>> Wow.
>> So, for any Doom fans out there, these
are limited run. There's only 1,666 of
these in the world.
>> Really?
>> Oh, yeah. This is sick. And I am a Doom
fan, by the way.
>> We go full kid in a candy store on
cases, limited edition Doom art, fancy
glass, even one with a tiny panoramic
OLED screen.
>> I would, of course, have probably some
kind of a GPU usage readout on there.
Memory usage readout. I love how you
have some of these on and just like
showing what they're going to look like
when they're running.
>> For the record, if you have to do two
power supplies, that's got to be your
case behind our wonderful cameraman.
Yeah.
>> Is that the case? The That's the box for
the case. The case itself is just a
little bigger than than the half 700
enforcer.
>> What?
>> This is the case.
>> This might be your case.
>> Can I lift this?
>> Probably. Yeah, you're a strong guy. He
can lift it. But how am I going to put
this in my office
with a lot of elbow grease? I'm scared.
I'm scared. This is crazy.
>> If people like how this looks in your
office, I'll do one, too, just for fun.
>> You're going to do one?
>> If you end up going with this guy and
the people in the video like it, I'll do
one, too. I'm a sucker for peer
pressure.
>> We kick around other ideas like an open
air test bench or just parking a second
PSU on the floor next to the main case
like a Cyberpunk space heater.
>> It's doable. It's just uh you know how
how neat do you want your office to
look?
>> How much office do I want to have left
is the question after putting this case
in there.
>> Yeah, the huge enclosed case is starting
to look less like overkill and more like
the only sane option.
>> So, this would be one of the power
supplies we put in I was a fan of
recommending the Taichi power supply
here. It's a 1650 W power supply, which
is just enough to work. It has two 16
pin outs, so you can use that for your
two Pro 6000s. Yep.
>> Another cool feature about this power
supply is for all those people who are
still scared about melting cables, uh,
which if you know how to plug in your
phone at night, you shouldn't be worried
about that. It has an extra sensor to
detect temperature as well, so that if
anything goes wrong, it's done. It's
also titanium rated.
>> Nice. Well, let's go look at the power
supplies.
>> Yeah, let's go look at power supplies
while I talk about the rating.
>> This Vertex 1200 is the one you
suggested to me last year. And I love
this power supply. I got like three of
them now, and it's really good. It's
12,200 W, but um this is also gold. I'm
seeing now. There's bronze, gold,
platinum, and titanium.
>> Like, you know, if one of you guys is
just here to build a gaming PC and you
want to get a titanium watt, the
question on my mind is why? But if you
do a workstation, then yeah, that can
mean something for you,
>> right? So for us, we got two GPUs in
there. I want to keep the heat low. And
when you don't have an efficient PSU, it
gets rid of some of that wattage that's
coming into it as heat so that it's
going to go away as heat into the
machine. We don't want that. We want as
high as possible efficiency so that that
power is converted into just straight
wattage going into the system instead of
heat. Did I explain that right?
>> Yeah. I don't I mean I don't disagree
with it. So titanium has about 95% up to
95% efficiency, which is really nice.
And if you go with something like
bronze, that's low. That's pretty low.
It's like 80 to 82% efficiency. Yeah.
And honestly, for the purposes of this
machine, I don't think we want to go
lower than platinum. If we're doing one
of those dual GPU cases, while your
Vertex might be fine for like one of the
PSUs, the second PSU has to be SFX or
SFXL, which are just like size standard
power supplies. So, you could look at
something like the ROG Loki. Um, I'd
probably give you the 1000 watt just
because then you have like the 600 watt
rated cable, but it's SFXL. It's
platinum rated and it would sit
comfortably in that case.
>> I have one of these in my current 6000
build.
>> Silver Stones is a good one, too.
Although it's like double the price for
an extra 200 W, which people can do what
they want with their money. financially
the better option is CPU GPU on one and
other GPU on the other because uh
talking about that Taichi that's a 550 W
power supply if we did that for two GPUs
and then you add um 280 for the
extra,000 watt for the CPU and anything
else going on there that's going to put
you at $830 right
>> so
>> it's pretty quick
>> uh I try uh you do that versus you do
maybe you could do the 1300 watt version
of that power supply which runs you
about4 400 bucks and you're still going
to want one of these obviously. So then
that puts you at 680 bucks. So you're
going to be aboutundred
$150 less than the other.
>> I also need to figure out how to connect
the two power supplies together so
they're synchronized. Yes,
>> I haven't done that before.
>> We have an adapter. So what this adapter
lets you do is you plug in the two
motherbolt cables into it from each of
the uh power supplies. This adapter.
>> Okay.
>> So that's the easy part and the least
expensive.
>> Yeah, this is good. I like that. $11.
the cheapest part but important. It's
how we get everything firing together.
>> That sorts out the power inside the PC,
but now we need a UPS or a giant battery
to survive power blips while it's
running LLM workflows.
>> These are about 180. The challenge for
us though is we can get away with one of
them being up to 900 watts if it's for
one GPU, but we need we absolutely need
that second one to be much higher. So, I
think that that super expensive one is
going to be unavoidable. And that one is
uh
>> 800 bucks. Yeah. 800 bucks.
>> And that's big, too.
>> That one is uh probably the heaviest
thing you would carry today if you if we
had to go with it.
>> It's heavier than the case.
>> Um I'd say they're about the same. That
thing is like a brick. And it's it's
bigger than the picture would make you
believe.
>> Okay. You're killing me, Dan. You're
killing me.
>> Oh, I mean, we both have to get our
workout today, right?
>> Two GPUs, man. That's all I want to
power.
>> Yeah. Well, not everyone has Pro 6000s
that they're playing around with.
>> How loud is this thing going to It's not
going to be louder than a jet plane.
>> Is that a microenter guarantee?
>> Unofficial.
>> I don't make promises.
>> It's a Dan guarantee.
>> It's a Yeah, it's a Dan guarantee. There
we go. There you go.
>> Let's do it.
>> I guess we're going to do that big case,
huh?
>> We're going to do the big case. We're
going to do the big UPS.
>> We're going to go big. Let's do it.
>> Good thing there's three of us, right?
>> You know, we forgot one thing.
>> Case fans.
>> We don't need this thing to be a rainbow
show, do we?
>> I don't want any rainbow show.
>> There we go. As far as quietness,
there's Be Quiet and there's Nachua.
Naka is famous for being quiet.
>> Noctua is the quietest. They definitely
earn their price tag with that because
no matter how you spin it, Noctua
>> spins.
>> Get it?
>> Uh, but they they do it quietly. They do
it efficiently. They, as far as air flow
goes, they're still the kings. So, we
can look at that option, too. I mean,
it's already an expensive machine. Why
not make it quiet or better than a jet
engine? I want to have as quiet as we
possibly can. So, let's do that. Several
knockup fans. Holy cow, this is a huge
fan. The UPS is in our warehouse, which,
as you guys remember, I have to
disappear into and then reappear with
the magic item.
>> Let's see.
>> I'll be right back.
>> Maybe we can follow him this time. What
do you think? We can sneak in there.
>> I don't think it'd go too well.
Microenter wouldn't like it. Yo. Yeah.
There you go. The sign is just going to
prevent all entry.
>> All right.
Where's Alex? I'm going to ask him if he
wants to lift it. Wait, lost track of
him already. Hey, there you are. You
want to try to lift it?
>> Are you okay, Dan?
>> Uh, is your arm getting ripped off right
now?
>> Not yet. Maybe after like 30 more
seconds. All right, let's see.
>> Lift with your back, kids.
>> Told him he was going to get his work
out.
>> I'm just kidding. It's not that heavy.
>> No, it's not that bad. It's not too bad.
>> It's It's only about this. The case is
where
>> it's not actually that big.
>> No, no, no.
>> So, this is rack mountable and I can put
it on my floor.
>> Yeah, you should be able to do it either
way. For now, on the floor should be
fine. Maybe like standing it up a bit.
>> This is not a cheap UPS. Once you're
going over 900 W, these things just
start jumping up more and more and more
and more. So, that's it. That's all
we're getting.
>> One more UPS and that's it.
>> Oh, just one more UPS.
This one's not so bad. So, you have to
get to a certain point for microenter
employees to take this to the car.
>> If you want,
>> what if I was like an a little old lady?
What would it take?
>> Just ask for carry out and they'll say,
"Yeah, we got you. You need help getting
something out to your car? We'll be more
than happy to assist."
>> What if Arnold Schwarzenegger came?
Would you make him carry all his own
stuff?
>> No. And that's just because I'd probably
take the opportunity to ask him for like
workout tips and stuff. I feel like
getting on his good side would be good
because uh if anyone knows how to train,
it's it's that guy.
>> Definitely Arnold.
>> Definitely Arnold. Yeah.
>> Thanks, man.
>> Can't wait for the peer pressure to make
me do the same thing. I'll see you guys.
>> Can I just put this in here and call it
a day? I was super stoked to get this
thing to the office and set it up, even
though I was kind of nervous cuz this is
my first Thread Ripper build, and it'll
be my first time running multiple
$10,000 GPUs in there. I also wasn't
sure about the power requirements, but I
wanted to start off with a huge 2500
watt power supply before I upgrade to
more wattage and more power, which I
feel I'm going to need. So, I put
everything together and threw it into
this gigantic case, and it looked pretty
good. This case is really smartly
designed where fan assemblies can be
built outside of the case and then just
be slid in. So much easier to work in
this case than that small factor case.
This is a quality case. I decided to
throw in a 50/60 in there just as a test
if things explode. Uh it would be a sad
but not $10,000 ad. And after everything
turned on just fine.
>> All right.
>> And I installed the drivers. I went with
one RTX Pro 6000 and finally two. And
this is how it looks now. Oh,
don't put your fingers on the fan. Now,
somebody already yelled at me on Twitter
after I posted a picture of this saying,
"Why are they so close together?" This
machine is still in development. Plus,
I'm going to need some space for some
extra boards that might also go in
there. A package I just got in the mail.
Yeah, I'm glad I got the big case after
all. Stay tuned on the channel for more.
This case will be back. Oh, you want to
see a model running on this? Is that
what you want to see? Now, I did throw
VLM on there. I tried it with a couple
of models, but here is the first larger
model that I tried, which is called Quen
3 235 billion. And this model is 142 GB
on disk. So, there's no freaking way
that a 5090 or a 5080 or a 4090 or a
couple of those 4090s or a couple of
5090s will ever be able to run that. You
need two RTX Pro 6000s to be able to run
that. Unless you're using a totally
smashed quantization. And I'm going to
do Olama here because it's easy to
install. I made videos about all the
whole process of installation and
setting things up before. Lama run quen
235 billion. Watch NV top memory is
getting filled up. Not all the way.
We're about 75% on each of the GPUs, but
it's being split up nicely because Olama
when it detects that the model is larger
than would fit on one GPU, it starts to
use multiple GPUs, which is kind of
cool. Write a story. Yeah, I know. I
know you're going to complain. What kind
of prompt is that? Right, Alex? More
testing to come. Okay, have some
patience. Oh, do you hear that? Do you
hear that? And look how fast it's going.
This is thinking and now it's
generating. Wow, look at that activity
going on over there on the two GPUs.
Both of them being used. And we're done.
68 tokens per second on the generation
side. And this is a 235 billion
parameter model. This is freaking cool.
I'm really happy about this. Thanks to
MicroEnter for setting me up with a case
and the fans and check out the new
Phoenix store which is already open.
Maybe I'll run into you. Stop by and say
hi. I'll be there on the 10th of
December. Now, if you want to see a more
in-depth test of the RTX Pro 6000, watch
this video right here. Thanks for
watching and I'll see you next time.