Today we have a very special episode of

the standup. We have with us Luke from

Linus Tech Tips. Say hi.

>> All right. There he is. Right there. We

have

>> I thought it was Luke Tech Tips. Is Was

I right? New

>> to not be confused with LT. He has LTT.

This is LT, not LT.

>> Okay. Different LT. Got it. LT. Okay,

that makes sense. Thank you.

>> There's some other guy who no one knows

who's like a co-host or something on

that. The main channel, I think. I don't

know. But Luke's the main dude.

>> Luke's does the heavy lifting.

>> That's real.

>> All right. Thanks for the intro

interruption. Yeah.

Uh, anyways, sorry. Also with us is Tee

and I interrupt things.

>> Thank you. Yep. TJ, you're wearing a a

sweater that can no longer be bought. It

must be very exclusive.

>> Arch, by the way,

>> Arch forever.

>> TJ already has children, so it's okay.

Um, all right. Well, anyways, today

we're going to be about you have Can I

just say though, you have to get married

and have kids before you install.

Otherwise, it's GG, boys. Your line,

your family line is finished. Okay, your

ancestors are disappointed. They hunted

and gathered for nothing. You ended the

line just because of your operating

system. Wait till you have kids before

you start Arch. Thanks.

>> That's just the PSA for today.

>> That's a good PSA. Honestly, I think

more people need to hear that. Um, all

right. So, uh, today we're going to be

talking about RAM prices. get some

thoughts where we think it's going along

with probably SSDs and some other

things. Is there any other unknown costs

coming in? And then after that, I'm

going to probably ask Luke a little bit

about running Linus Tech Tips, what it's

like hiring people, all the good stuff.

Uh what does he kind of experience in

today's age of AI and such, especially

in the YouTube scene? Uh just because

it's a much different set of technology

and how you develop stuff and kind of

the software you run for. Uh very

curious about all that stuff. So, let's

start off obviously with the maiden

flagship topic, which is RAM prices. I

don't know about you, but the moment I

heard that AI loves RAM, the first thing

I did is went out and bought MU, uh,

which is a stock, Micron, which

exclusively does, uh, RAM. And guess

what? It was my best investment I have

ever made in my entire lifetime. And so,

there's there must be something there.

If the stock market says there's

something there, there must be a there

there. And so,

>> number go up.

>> The number does go up. Now, there's only

one person in this call that I know for

a fact can not only build the PC, but

also knows how to order a GPU from the

internet, which is an impossible task

these days. I have no idea how to order

a GPU. So, I figured that we'd bring

Luke on and he could uh kind of give us

some of his thoughts about this whole

RAM price debacle and where he kind of

thinks it going, where he thinks it's

kind of going and maybe inform us uh

uneducated non-GPU buyers about the

hardware market. the permanent

underclass. Permanent underclass right

here.

>> Sure. Yeah. I um I'm not surprised by

certain moves like uh like Micron's

Crucials. Uh so it was Crucial is owned

by Micron. Uh Crucial stepped out of the

RAM market. That was not too surprising.

I think they've honestly wanted to do

that for a long time. If you look at how

honestly somewhat specifically DDR5 was

even designed, it was not designed from

a consumer standpoint. was designed from

a enterprise standpoint. The goal has

been enterprise for a long time. This is

mostly a convenient and high profit exit

from the consumer space for Crucial. Um,

and I don't personally suspect that

prices are going to come down from

supply being super high for a pretty

considerable amount of time. It takes a

really long time to get fabs online. Um

and they are also going to resist the

fall of the price because they want it

to be high. Some of the like most

collusion in effectively like any market

anywhere has been in RAM. Uh these

companies love working together to price

fix. They love working together to um

keep prices not too high to not increase

fab capacity too much. That's another

thing they do to um I don't know if you

can say artificially or not. There you

go. Nice. Uh

>> I'm listening. I'm listening.

>> This is super well documented if you're

if you're interested in it. Um it

there's been a lot of legal suits over

it. Um it's it's not even a secret at

this point, let alone an open secret. So

you can it's Yeah, you can dive into it

if you're interested.

>> Commish, you got to listen to me.

>> Oh no. Now one more comment from you.

I'm done rebasing your mistakes. You're

on junior CSS duty until further notice.

>> Commish, you can't do this to me.

>> Keep talking and you'll be doing store

procedures for a month now. Get out of

here. Take him with you.

>> Fun fact, CSS is actually touring

complete.

>> Larry, Gary, Tango, Mary, I'm just

pulling your request. It looks good to

me. You're clear to ship. Thanks. You're

welcome. Next.

It's an awfully big PR

>> for an intern.

>> Well, I just bumped some dependencies.

It's nothing major.

>> Hey, can I get a quick stamp on this?

>> Yeah, don't worry about it.

>> Quick approval. Oh, not on my watch. I'm

on your disc like a pee on slur. Not

this again. It's literally just a hex

code change. Just to prove it. Just to

prove it. Squish. Perf resisting review.

Oh, I know you're the tipler. I've seen

vibe coding, but that AIN'T IT.

>> MERGE COP. NO,

>> I hate merge cop. He always makes

reviewing take forever. We have Code

Rabbit. Oh, come on. I wasn't even

merging the prod. It was a hex code

change. We have Code Rabbit. We don't

need real people reviewing such simple

changes. Code Rabbit can do it for us.

Our engineers time is better spent

solving problems for customers. You can

try it too at codrabbit.ai.

>> Next week on Merge Cop.

>> Now my plan to merge a dip so big you're

the dipler and I always knew it.

>> They have also worked together to yeah

not increase fab capacity. Now right now

there is enough demand that yeah they're

looking into fabs. There's partnerships

from from at least one brand to work

with a much smaller fab company to try

to increase their capacity and and

throughput. There's one happening in

Taiwan. Um,

but it's going to take a while for us to

really see the benefits of that. I think

the thing that might happen first, and I

could be wrong here, is a

slowing down of investment in

um in data centers and the a need for

RAM. Um I don't see that coming anytime

soon. I don't personally necessarily see

a like massive market crash that some

people are predicting for AI happening

super soon. uh due to a lot of the

companies that are really driving this

forward being hugely profitable

regardless of AI. Um

but yeah, I guess that's my like my

quick thoughts. Quick question. You said

that DDR5 was designed more for

enterprise, less for consumer. I have no

idea what that could possibly mean. Can

you say a little bit more about that for

me? Like what does it have like parental

controls or what? Like

>> I also just to be fair I think DDR

stands for Dance Dance Revolution. So

like that's how much I know about RAM.

So I'm like I'm very far

>> number five. Sick. I thought they'd have

more editions out by now. But that makes

sense.

>> It's a very popular game.

>> It is it is disappointing that they

don't have more Dance Dance Revolution

editions. Um no there's a there's a

really good video from

sorry um from Wendell from Level One

Text. if you check out his YouTube

channel and scroll down a bit because

it's not very new. Where is it? Hold on.

Uh, it's not that old either. Um, it's

it's called your DDR5 memory could be at

risk. All about DDR5. Um, 20 minutes

long deep diving what the heck is going

on with DDR5. I think it's interesting

that he has that video about issues that

we're finding especially in the consumer

space but definitely on the enterprise

space with with uh DDR5 and also um Lus

Tvoltz was on our channel not that long

ago and he was mentioning that he thinks

a lot of the problems that users have

with Windows is actually users with bad

RAM. um which was which was very

interesting and part of these two

discussions in my opinion merged

together because of one of the points

that come from it being designed

enterprise first is it's not really

designed for what most of us have in our

like desktop chassis. um which is

honestly not a ton of air flow,

especially compared to a like server

environment where you don't care about

uh fan noise, you don't care about um

much to be honest. You just want the

performance and ideally low power draw,

but that's often a uh an afterthought.

Um, so a lot of desktop DDR5 is like

overheating or having various other

problems and Wendell and some of the

level one tech forum crew and whatnot

have designed these like I don't know if

you were into desktops back in the like

DDR3 Corsair Dominator era where they

had those metal fan brackets where you'd

have those little tiny fans that would

go over your RAM. Um, well those might

actually like matter now. Um, and

they're the the community is making 3D

printed uh shrouds for their RAM so that

they can then mount like small I think

it's 80 mm, but it's been a while since

I've looked into this. Um, little tiny

fans and point them directly at the RAM

up close um to get more air flow on

there, then they're getting um less

errors, less problems. Um because if

if it's doing error correction like on

the actual stick before it gets to the

CPU, all that type of stuff, and it's

overheating, that's not going to perform

as well, and you might have more issues

is my like fairly not amazing

understanding of what's going on in

there.

>> So, hold on. I just want to rewind that

for a second. Mhm.

>> You were telling me that some of the

Windows issues I have not

>> Before you say that though, it's the

stand up. Can you say we're going to

circle back on that? That's a little bit

more work appropriate. We don't rewind

here.

>> Please, I don't want to boil the ocean

right now. Okay.

>> Okay. Thank you.

>> But okay, because I I can't my uh my

ability to maximize Windows decided to

no longer work anymore. And so that's my

life as a Windows user right now is the

ability to only have Windows one size.

Uh, so that's RAM.

>> Yeah, that's all that's all RAM. That's

all RAM.

>> What?

>> Probably not RAM.

>> Yeah, this is crazy. I thought again, I

thought I was a good programmer. Wind

>> is just bad. Okay.

>> Okay. Okay, fair.

>> There's only so much we can do.

>> Yeah.

>> All right.

>> I do like that though. I I like I like

Microsoft coming out and saying,

"Actually, what you guys need is little

fans on your RAM. that's going to fix

it out yourself.

>> Yeah. No, I think it's more like

crashes, blue screens, um application

crashes, application errors. If you look

into stuff in event log and there's like

things going wrong, uh it might be

related to that, but it it shouldn't be

like a a random function not working

like be able to full screen windows

>> that obviously.

>> Have you asked Cortana and or Copilot to

full screen it? It might work that way

instead. I actually haven't, but I am

very curious about doing it. I did have

>> Please record it. Please record it. I

need to see I need to see the results.

>> Yesterday I actually did give my first

AI swing of like a of of a application

and uh I noticed that Gemini has been

added to all Google Chromes and uh

Windows. So I was like, "Oh my gosh, uh

I can't log into Frame.io and so I want

to delete this site's cookies." So I

clicked on Gemini and said, "Delete this

site's cookies." And it said, "I can't

do that." And that was like my that's

been my only experience thus far,

>> which is very disappointing.

>> I just wanted the cookies gone. So then

I had to go and ask Gemini on uh on

what's it called on Google. Truly,

>> it is a futuristic AI then though

because it's like it is denying your

requests. That's what the end game is

for AI, right? So they they achieved it.

>> They achieved it. Yeah. Lore accurate

for AI. So

>> that's funny. Uh, so we are like uh

we're year one effectively into the the

great ramming. Uh,

>> I'm not

>> That's a Oh my god, that's a great

title.

>> Thank you. I just made that up right

now. But anyways, the great ramming is

happening right now.

>> How long do you think this can go before

RAM prices level out? Because I was

looking at it and MU and Western

Digital, which is SSDs, they are up 300%

in their stock price. So like people are

obviously pricing in what they think

this you know this value is going to be

and it it has not looked like it's

slowing down at all on the stock market.

Is this reflective of what RAM prices

will be? Are we going to see a 3x cost?

>> Aren't they already pretty much at that?

>> As per usual, this is not financial

advice, but

>> is the stock market predictive of

anything these days? Um, I mean, you you

had a good guess with with investing in

MU early, which is why you own the

office building you're sitting in right

now.

>> Thank you. Thank you.

>> But but it's like

>> he's leasing. Okay, Luke, don't don't

let him get out of

>> smart with my money. Okay, I lease

>> commercial real estate's terrible.

>> Yeah. Yeah. Um,

yeah. I don't know. I I mean, I think I

think the demand and price is going to

be high for a long time. I think that's

one of the main things that I meant with

my like little intro thing there. um is

that I I don't think it's just going to

randomly come down. Um I think for the

last few years we've been seeing a

general all computer hardware shift

towards enterprise and when you do that

and and consumer becomes more and more

of an afterthought, uh consumer gear is

going to be more price performance

expensive for what you're getting. Uh we

saw Intel recently talk about this where

they're like Intel just openly was like

you know what we've been giving the

consumer market like too much attention.

We're going to focus more on

hyperscalers. Uh we're we're not

interested in that anymore. You're

seeing crucial step out of the market.

You're seeing Nvidia like

>> almost like begrudgingly still talk

about GeForce cards. There's like these

these companies are are seeing the bag

that is um working with enterprise. It

was actually I mean even for me like

being in the hardware space

predominantly it was it was fascinating

not that long ago to find out that like

a lot of these water cooling companies

that I thought of as these like scrappy

little consumer water cooling brands

sold the vast majority of their revenue

of products through enterprise doing

water cooling for servers and data

centers. Um, it's it's an interesting

world and there's a lot more money in it

than like um, you know, uh, us like

waiting for deals on freaking New Egg or

whatever. Uh it's it's yeah it's a it's

a totally different thing when you when

you're willing to buy like just some

incredible amount of uh of of money

worth of like if I don't know some of

these reservations for even during coin

mining some of these massive operations

were buying pallets and pallets and

pallets of GPUs and if they had issues

with them kind of like almost shrugging

it off. Um, whereas consumers, like, you

know, if you have an issue with it,

there's a there's a Reddit thread with

7,000 upvotes and you're screaming from

the mountains and all this kind of

stuff. Like, why why deal with us

annoying low people who lack money uh

when you can deal with the the

hyperscaler

boys with all of the money and and less

of the problems. Yeah. You know, it's

funny you should say that because if you

have looked at OpenAI's kind of oraoris

investing, uh, Nvidia has promised

they're going to pay them a progressive

amount up to one or hundred billion

dollars and OpenAI's apparently one of

their agreements. I was really trying

confused trying to find out the exact

number and the only number I found was

they're going to pay them back 10

gawatts worth of purchases. And I was

like, I don't know what that number

means cuz typically I use USD. I'm not

used to whatever this AI money currency

is. But a gigawatt appar again this is

Chad GPTI ask like what the hell's a

gigawatt what the hell's a gigawatt and

it said one nuclear power plant worth of

energy. So I was just like oh my gosh

that's how much they're going to be

buying is 10 nuclear power plants and

then I was like did they make this deal

elsewhere and then apparently with

Oracle their 300 billion will also

result in 6 gawatt of data center or six

nuclear and I like 16 nuclear power like

that's first off that sounds like a

metal band. Second off, the United

States has 39 total. Like that's half

that's like 50% of the entire power grid

worth of

>> new nuclear power plants. And I'm

probably not even saying nuclear.

Correct.

>> So

>> it's nuclear.

>> Nuclear.

>> It's nuclear.

>> Nuclear power. Yeah.

>> Nuclear.

>> Nuclear power.

>> Nucleular.

>> Nuclear.

>> There's a lot of

>> nuclear. Yeah.

>> I don't know if you want me to inject

some random technical stuff in here or

not. Yes, please. Certainly can. Okay.

So, I don't study any of this stuff.

Like, hardware is definitely not my

monkeys and not my zoo. I just look at

it and go like, look, the worse the

hardware is, the better is for people

like me who like to talk about

programming performance because it just

means you have to be better, right? So,

I'm fine with like go great. If there's

a if you have to start programming for a

10-year-old laptop, I'm happy about

that. But

>> yeah, actually a typical data center, my

understanding is that you're in the like

100 to 200 megawatt range for the data

center. Like like a typical modern data

center, that's what you would be looking

at for the total number of megawws.

Whereas AI data centers are like 10x

that for power consumption. So when

they're building those, they're looking

at things like gigawatt or, you know,

mult mult like up to over one gawatt

worth of inflow of power to this data

center, right? That's how much it

consumes. So when they talk about

gigawatts, they're literally talking

about like, okay, 10 gigawatts might be

five data centers or something like

that. Five new data center buildouts

worth or something like this, right? if

I'm just like ballparking those numbers

based on what I've seen. Uh so when you

think about that when they're

guaranteeing that kind of purchase, I

mean that's an ungodly number of GPUs,

right? You know, think think literally

five data centers or something or more

worth of these racks upon racks of

Nvidia Rubin like you know slotted

things or whatever you know they're on

at that point.

And so that's just I mean I don't know

we'd have to go break out a calculator

to even try to figure out what kind of a

outlay that is for Nvidia but it's

massive right like it's a massive

amount.

Um tying that back though to the thing

we're actually talking about which is

memory. My understanding was that OpenAI

actually signed some kind of nutso deal

where they were going to buy up to

900,000

wafers a month of DDR memory. What's a

wafer and how much memory's on a wave?

>> Uh, that's a Okay, so again,

>> not my monkeys, not my zoo. I'm doing my

best.

>> It's classified prime. It's classified.

>> And

it is if you actually want to know the

the yield per wafer,

>> like they have the number of like bits

per wafer. These are things you pay

analyst firms to get and they very they

they are constantly moving around

because people are trying to increase

their yield and so on.

>> A wafer is a piece of silicon. It's

circular. You've seen them.

>> Is that the circular one that we talk

about?

>> Yep.

>> So they're just buying circles.

>> Yep.

>> They're buying the output of the

circles.

>> Yeah.

>> Okay. I was about to say I don't know

how to take a circle and turn it into a

stick. Like that's very difficult.

Casey, I've never seen a GPU that's

round. So, I don't know if I believe

this.

>> Well, there are some uh the the uh

there's people who make wafer who make

wafer scale. They're kind of prototypes,

but they make wafer scale AI

accelerators, and they are circular.

>> That's pretty cool.

>> Yeah. But that's just nutso stuff. I

mean, I don't

>> They're designed after Sam Alman's orb.

>> Yeah.

>> So, it's like it's circular.

>> You can only do it.

>> I think technically uh technically

Prime, you're more right. You were

trying to make a joke, but you're more

right than you think you are. I believe

the Open AI deal is literally for uncut

wafers. They have been patterned, but

they are uncut and they are going to be

like controlling where that gets shipped

to and how it ends up getting stacked

and packaged was like the last time I

read it, they were uncut patterned

wafers, not cut and packaged, which is

very unusual. I could be wrong, but

don't say it. My recollection.

>> No, it sounds like you're right. I

didn't actually realize.

>> I believe I am. Yeah. Yeah.

>> I'm thinking about a Sam Alman as

Scarface right now.

>> Yep.

>> Just uncut wafers all over his desk.

>> Um, so

>> again to give a really bad explanation

because I'm the wrong guy to give this

expl you really want to get like, you

know, uh, Dev Patel.

>> What about the guy's name for Dylan

Patel? Dylan Patel from um semi analysis

or something. You want to get him, he

would know, right?

>> It depends. Yeah. Like we're we're more

on the consumer side of things. Like I

don't I don't get way into this too

often. I do know that like the the

difficult part is making the wafer. Um

like there's there's somebody online who

made their own stick of of computer

memory um using using chips they got

from you know somebody else. Um, so it's

not like it's not crazy surprising that

they bought just the wafers. I just

thought they would, you know, cut to the

chase and get them to actually hand them

over like functioning sticks, but um,

>> well, they're not sticks is the problem,

right? Because they are using HBM. So,

>> Oh, right. Yeah. I mean,

>> what happens? Yeah. So,

>> for us normies,

>> okay, so, uh,

>> this is again way out of my league. and

I both have to guess what bandwidth for.

>> Yeah. So, what it is,

>> okay,

>> what it is is like if you think about

normally how memory works, right, it's

it's on a stick and you basically have

these the the individual DRAM chips that

have been fabricated are like on a line

of the chip and then the the like, you

know, the connection is the you know the

little pins it slots in and that's how

it's talking to CPU. So, if you think

about it, you've got like essentially

the CPU is on a package, the DDR is on a

package on a little thing. You've got

those connections. They go through the

PCB, right, your motherboard. And

there's like some small number of

connections connecting them, right? So,

that there's like traces on the PCB that

are going to drive over to those like,

you know, places you slot them in. So

GDDDR,

right, which is the graphics graphics

DDR, not the kind you slot into your

thing, right? So the kind that goes on a

on a on a GPU

is a little bit different. That one is

like welded right on like it's like

soldered onto the motherboard and

directly connected. And the re and

closer and the reason for that is they

it's basically the same kind of memory

for all intents and purposes as far as I

know but the signaling is much much

faster. So they they drive the data rate

up the the band total bandwidth they

drive up by increasing the speed at

which it can transfer things back and

forth. And so it has like basically

higher quality sign physical signaling

to get things back and forth. But

otherwise, same basic idea as the kind

you would slot

>> into your motherboard. Otherwise, not a

huge difference. Beyond that,

>> some people were asking for visual

representation. And if you if you look

up if you if you wiki I don't know if I

can share my screen on this thing, but

uh if you wiki high bandwidth memory,

there is a very good photo if you scroll

down to the interface section um that

shows how it's like 3D stacked as well.

Um so you can

>> Well, we haven't gotten HBM yet.

>> Okay. totally different than both of

these. Totally different than these two

things.

>> So HBM is completely different from

those two things.

>> Completely different.

>> It's the same sort of memory. Like it's

still the idea is still that there's a

capacitor and a transistor per cell of

memory. Like so the actual thing you're

fabbing is somewhat similar, but it's

very different in two very important

ways. One, like Luke just said, it's

stacked. And in order to stack it, it

needs to be uh manufactured with this

sort of different kind of uh

connectivity. It's got these things

called TSVs or through silicon VAS.

They're like these connections that go

through the stacks so that you can kind

of like have each stack is talking to

the next stack. It like and and they

tunnel through, right? So it's wider,

right? It's a wider. It's like the

actual physical footprint for the same

amount of memory is a little bit bigger

because it's got to have space for this.

The normal DRM can be doesn't have to

have those, right? Those TSVs. So,

that's a thing. But the much bigger

thing, I mean, although that's that's uh

obviously a slight difference there, the

much bigger thing is the stacking and

packaging is just a way harder problem.

So, first you have to be able to stack

them and this reduces yield apparently

for I don't know the reasons why but

like

>> again can you explain yield?

>> I I forget just like a quick like

oneliner. Why do you not like

>> percentage of stuff coming out based on

what you put in?

>> Like just because some of it's bad. It

just comes out like broken. So they're

just like this wafer is broken, bro.

>> We're different levels quality.

>> Yeah.

>> Yeah. Computer chip small hard to make.

We mess up times sometimes bad. That's

>> okay. Okay. Okay.

>> The answer is that's what I needed.

That's what I needed. Okay. Thank you.

>> The answer is I don't know. Normally

when we talk about yield, I understood

it fairly well from the old like what's

the yield on a wafer? Cuz you figure you

make a wafer, there are defects on the

wafer. So you you know, you imagine

you're patterning this thing and you've

got defects. some of the defects or

maybe could be, you know, some kind of

impurity or something went wrong or

maybe when you were patterning the light

just hit something or there was one

little speck of dust in there or

whatever. So, you have certain things

that didn't quite work as well as they

should on this wafer. And so, some

number of the things that you put on the

wafer aren't are just going to fail. And

so normally when you talk about yield,

you talk about like all right, if

there's n number of defects on a wafer,

we expect those to fall fairly randomly.

Sometimes they're they're distributed

differently like more towards the

outside than the center because of the

way the reticles work or all these other

sorts of things. Who knows? But so you

only get a certain number of those

chips, right? Only a certain number of

them will work or a certain number of

them will perform better than others for

certain reasons. All those sorts of

things. And all that goes into what you

call your yield, right? And bin. So bin

is right the process where you say like

some of these chips work better than

others for whatever reasons, run at

higher clock rates, things like that.

Others are like they literally don't

work. There's a defect on them. We can't

get it working at all. And yet others

are designed to like have certain things

constructively turned off. think GPU

where it's like okay there's this many

processing units on the GPU and they're

designed to have some of those fail and

be turned off right so it's like okay

this just has less cores now and it gets

slotted as a different thing blah blah

blah that's the normal way that like I

have heard yield talked about for this

apparently when you you have to like

grind down the silicon and like stack it

on top of each other and all the and I

guess there's a separate yield loss that

happens. It has nothing to do with

whether the original chips were working

or not, maybe. But then there's also the

fact that I got the sense when I tried

to look at this stuff cuz um I knew we

were doing this podcast. So, I was like,

"Let me go see like what what are they

talking about with some of these memory

things?" And I wasn't really able to

understand it because I wasn't sure if

what they were talking about was yield

loss from we stack these wafers on top

of each other and that process produces

problems or if what they were talking

about is we can't test the memory prior

to the stacking for some reason. Let's

say I'm making that up. I don't know.

And so when we stack eight things on top

of each other, we just multiplied our

total failure rate by eight because if

one of them's bad, they're all bad,

right? Because the VAS won't work or who

knows what. So I don't know which of

those are talking about, but what I've

seen put out for numbers is when you're

making one of these HBM modules, your

your yield rate is awful compared to

normal. They said three times as much

wafer space is required for the same

amount of HBM memory as for regular DDR5

or something like that. So the same

number of bits stored if you're storing

them in HBM, you required three times

the input wafer size and patterning to

get that out. And that's a huge yield.

Like that's massive, right? I mean just

think about that number. It means you're

doing all the same work, but you end up

with three times more chips for DDR5 if

you were producing the R5 as if you did

HBM.

Make sense?

>> Okay, that does make sense. So, that

sucks, right? That that is really really

bad. And it means that the price of

those things has to be much higher

because you're doing a ton more work for

each one. You're doing three times more

work to get the same amount of bit

storage. So if you want a gigabyte of

HBM versus a gigabyte of just the random

GDR or DDR that we're sticking in the

computer,

>> it's it's right.

>> Sorry just am. Okay,

>> just to finish this ridiculous tangent

on how it works.

>> So the thing about HBM is what I tried

to get to at the the beginning before we

went into that yield thing is it doesn't

go outside the chip. It's not connected

to the chip in this sort of way that we

think of like with GDDR that's like SL

on the PCB or DRM which is slotted in.

It's on the package. So the way that it

works is you literally, you know, you

think about I don't know if you guys

have ever seen like die shots of

something like a modern AMD processor

where there's chiplets. There's like

little chiplets in there. So it's like

you delid the processor. There's like

two, you know, CPU dyes and something on

there, right?

That's how HBM works. It's on the

package. So, the CPU and the memory are

all together in one module. They're not

out. They don't they're not things you

plug in or put on a PCB. Does that make

sense? Is this uh like

>> this could be way off, but some of the

stuff with like Mac minis and all these

some of the new Mac stuff where they

have like the unified memory, which is

why a bunch of people are using them for

running local LLM stuff because they

have this higher bandwidth connection to

CPU that actually lets you run like

local models at reasonable speeds. Is is

this related at all? I don't know.

>> I'm just asking.

>> I am not sure. I I want to say that

there may be I thought there was a uh

>> I'm not I'm not sure. I don't want to

say because I'm not sure about that. I

know that there's uh there are some like

A series uh chips I think that use on

chip memory which is very different and

super duper duper fast, way faster than

even HBM. Um but that's a different

thing which we could go into later but

it's not relevant to a DRM shortage at

all.

>> Yeah.

>> Yeah. They're they're expanding cash in

in a lot of ways, but like you said,

that's not really related.

>> Yeah, it's a different type of it's a

completely different it's it's SRAM.

It's not even the same kind of

>> Can I Can I ask a dumb question?

>> Yes, please.

>> I have been asking dumb questions the

entire time because this stuff, like I

said, not my monkeys, not my zoo. And

you look into it and you're like, Jesus

Christ. Like it's just like this huge

morass of stuff where you're like, oh

god. Okay,

>> so

>> me the consumer, I don't just want a

wafer. I want it like nicely packaged

and then have like little LEDs on top of

it. And then I want NZXT or something

LEDs.

>> You're not an LED. Come on.

>> Dude, I have LEDs right now, baby, with

my with my little rings.

>> Oh, no.

>> Just like, you know, the whole nine

yards, right? And that's me as a

consumer. So, why in the world as a

producer would I ever sell to consumers

ever for any reason if Samuel Jippity

Alman is just like, "Give me the

circles. I don't even want like I don't

even want you to put LEDs on them. I

just want I just want the circles.

>> And

>> that's the that's that's the short. You

just described the shortage. Yes.

>> Yeah. That's what's happening, right?

You nailed it, Prime.

>> Good job.

>> Very done.

>> Episode's over. We can all go home. So,

we're like literally never

>> crucial was the the consumer,

specifically consumer side of Micron.

And they were just like, "Yeah, screw

it. Why do I want to deal with you guys?

I can just sell to the to the

hyperscalers. Um there is if you look at

the amount of companies that are

actually making wafers um versus the

amount of companies that are selling in

in you know the most consumer side of

things, the sticks of RAM. There's an

incredible amount of companies selling

sticks of RAM. Uh there are very very

very few actually making wafers. Um the

the hard part is the the making of the

wafers, not the making of the sticks. um

or in the in the HBM sense, uh it

continues to be hard the whole way

through. Um we're we're we're

specifically talking in this case about

about the sticks. You were talking about

putting LEDs. You're not putting LEDs on

HBM. Um so I'm I'm talking about sticks,

but

>> they're missing out.

>> They're missing it out. Data center

them things that like, you know,

primarily coolers that go on top of of

HBM or something like that. But uh

that's probably what you might see. But

>> when Razer gets into the HBM market and

there we go.

>> I'm I'm imagining

>> when I get my CHBT subscription, there's

an additional tier and I get like a live

video feed into the data center and I

have like my rack and it has my cool

LEDs on it for an additional price. Like

they're missing out on these secondary

effects that they could be selling, you

know?

>> I like that. Yeah. So that means the,

you know, the AI future always shows

these data centers that are like glowing

with bright lights. So Sam's not going

for that style of LED future. He's going

for like the dark scary one.

>> Yeah. Yeah.

>> Okay. And we don't need light where

we're going. Okay. That's pretty

interesting. So does this is there is

there any computer part that's protected

from this AI revolution? Like is there

anything that's going to remain normal

priced or am I just screwed? Because I

have to buy a new computer here soon.

Um, I don't I should probably buy it

sooner than later. Is it It seems like

But also, is there like anything that's

going to be cheap or is it all Is it all

gone?

>> Well, I mean, one kind of sort of nice

thing is that the whole RAM situation is

sort of creating this nice bottleneck.

So it's unclear like

it's unclear how much CPU prices would

be affected long term because TSMC so

TSMC okay so there is a way in which the

production capacity for things like CPUs

is implicated by this but I don't fully

know how to what extent so when you do

these HBM modules the stack that stack

stack that we were talking about. The

bottom of that stack is not memory. So

the things that are like up the whole

way, that's memory, but the bottom is

actually a logic die. So it's kind of

like one logic layer and then a bunch of

like memory cell layers. And that bottom

die supposedly is actually like advanced

process logic process. So, my

understanding is that TSMC actually is

going to be fabbing a lot of like the

for HBM4 or something. They're actually

going to be fabbing the bottom layer.

So, it's like TSMC fabs the bottom layer

of the stack. SKH Highex, Micron, and

Samsung, if their stuff ever works, will

be fabbing the other layers, and then

they get like packaged together and then

stuck on the the, you know, chip, the

the wafer, the COS

packaging thing or whatever they're

going to be using at that point. And so

I suppose that could have some negative

consequences and that that adds more

stress to the TSMC side, but I don't

know like to what extent that competes

with things like you know the logic dice

for CPUs and stuff. So it seems like if

you're sitting around waiting for memory

all the time because you're memory

supply constrained then that would mean

that like the the fab processes that are

used for CPU and GPU dies that aren't

the parts that aren't the memory. That

part seems like it wouldn't necessarily

go up, right? Cuz they're they just have

extra capacity at that point because you

can't make new accelerators without the

memory and you need a lot of that memory

and they just don't have the fabulous.

So that could be someone out there has

done that analysis but I again it's

probably something you have to pay for

the research.

>> Yeah, there's a lot of analysts. I think

like theoretically pretty much

everything could be impacted to a

certain degree. Like if we're if we're

building all this stuff, basically

everything in a consumer computer, those

brands, a lot of them, not all of them,

uh are are also in the hyperscaler

space. But the amount of impact on like

a company that makes uh you know, server

chassis and desktop computer cases is

going to be effectively nothing because

you can scale that up no problem. Um,

and there's a lot of potential

manufacturing in that space. So, it's

just kind of whatever. Um, the amount of

things impacted is going to be like a a

computer fan, for example. Uh, they

might have lots of fans, but who cares?

You can make tons of fans all the time

basically everywhere. There's there's

there's not really going to be any

actual real impact there. Um, so it'll

it'll probably stay mostly with um, you

know, the things on or attached to or

very close to the board. Um, which is

mostly impact that we've already seen.

But I I do agree. I think it's

bottlenecked.

Casey, so are you saying since we have

to pay a lot of money for memory and our

CPUs are just going to be waiting all

the time, JavaScript is so back. Is that

like CPU doesn't even matter? We're just

waiting around.

We're just waiting around. It doesn't

even matter how long it takes anymore.

It's just chilling out in memory.

>> Or other way around. You'll only be able

to like new machines will only have 512

megabytes of memory cuz that's all

you'll be able to afford. Like each year

the memory will go down and down and

down until eventually you're running

like the Commodore 64 is like roughly

what you get. You 64K. That's all you

you know you plebs can afford.

>> I've trained normies are not allowed to

use graphical gigabyte. You don't get a

gigabyte of memory. What are you

talking? What are you? You think you're,

you know, some kind of uh the president

or something? You don't get a gigabyte.

>> Maybe uh maybe Bill Gates's old fake

quote was actually right, but it wasn't

right because it's most of what we need.

It'll be right because it's most what we

can actually get.

>> Yeah. It's like that's what you get.

>> I like this.

>> 640K will be enough

>> for anybody because it is all you will

get. Yeah. Yeah.

>> All right. So, the final question,

>> own almost nothing and be happy.

All right. So, the followup question I

think is um is pretty interesting

because uh Google Stadia obviously they

they killed Google Stadia. Uh Netflix

actually went into this whole idea of

doing um games on the server and they're

they're doing it quite successfully. Uh

a lot of these places it kind of seems

like we're moving more and more to the

server. Uh does this mean that within

the next 5 years we could see something

along the lines where consumers are

getting so priced out that it's better

just to rent uh your workstation. It's

better just to rent your time on some

sort of server computer cuz I know

there's a lot of this being pushed with

like VS Code and GitHub and everything

where you can just you rent out your

development environment as opposed to

you actually owning your development

environment. You're just like I can

program anywhere now on any you know I

just need a terminal and boom you're up

and running. And so, um, is this like a

future that we're kind of forcing people

into being like, "Oh, you want a

computer, brother? That's $20,000. Like,

you you better just go rent one for

life."

>> I think there's some desire there from a

variety of companies. But I I think when

we're looking at um at least some of the

comms that I'm seeing from these like

RAM wafer companies and stuff like that

is that they they want to scale up

because of this, like they would like

both markets. Um I but I don't know. I

mean there's a there's a Twitch chatter

mentioning Chinese manufacturers

covering to the the memory market. Like

there we might also start seeing

competition like that. Um I wouldn't be

too surprised if the gap between what

the enterprise has available to them and

what the consumers have available to

them or at least what the consumers

reasonably have available to them

continues widening. um for for a while

there the unless you're looking at like

what was that the name of that CPU like

knights able or whatever um unless

you're looking at like the extreme end

of of server specific compute hardware

uh consumers were pretty interested in

that stuff for a long time and now we

have GPUs that are just mind-blowingly

expensive and CPUs that are

mind-blowingly expensive on the extreme

high end um that just in a lot of cases

while those things are relevant isn't

going to be super necessary on the

consumer side of things. Um, so

yeah, I I don't think that's necessarily

going to happen because

there will always be a section of the

market where someone could make money

selling to those people. Um, and I think

the the Chinese who are both getting

into CPU, not both, but uh all getting

into CPU, memory, GPU, all that type of

stuff, uh, they might come for that

market. Someone else might, not sure.

But yeah, I don't I don't think that's

going to happen.

CXMT is supposedly not far behind modern

memory standards like right now. Um, and

so they're kind of predicted to be able

to ship at least like DDR4.

>> Yes.

>> Style memory, right? Or something like

this. So they're they're not

>> Chinese memory manufacturers aren't like

way behind. And so it's very possible

that especially if you start talking

about well you know no one can afford

DDR5 so you know maybe you could ship a

cheaper thing uh or a less good thing

and that would be fine right

>> and maybe you're you're you're not able

to buy as high performant computers or

relatively as now or whatever but maybe

Casey's stuff gets more popular and

programming goes more more uh you know

efficiency and performance and stuff

like that we start making Roller Coaster

Tycoon again.

>> Yeah.

>> Um

>> there we go. Yeah,

>> it really consumer desktop is just like

gone. Uh or consumer compute, I should

say.

>> I was going to say it's actually really

uh this this whole kind of crunch on

everything and and not being able to buy

the next greatest MacBook uh MacBook Pro

every single six months from the Silicon

Valley. Uh it is actually kind of a

positive thing because people might

actually start experiencing the uh the

crap they make on devices that aren't

the premium of the premium. followup

question, which is will when do you

think or do you think this is going to

hit phones soon? Because I kind of

realize like people buy phones fairly

frequently in America.

>> Mhm. Like at what point does Verizon

say, "Yeah, sorry. We're not giving you

a phone with a 2-year contract anymore.

It's just like too damn expensive now."

Like is there is there a world where

this is coming soon? Because I assume

all those parts are equally susceptible

to whatever is happening. The rammoning.

Yeah,

>> it would make sense that it would. Um, I

haven't really seen it happen yet. Uh,

you you've seen it a lot of other

worlds. I don't know if that's because

like they have contracts far enough out

that it hasn't really impacted their

their available stock at this time. Um,

so I I don't know when that would

happen, but it it Yeah, it would

absolutely make sense to me that it

would.

There's also like supposedly and this is

kind of I I don't know to what extent

they're talking about this but I guess

the AI buildout doesn't only affect DRAM

it also affects like NAND flash storage

and things like this like apparently

like you know these machines that

they're prepping I don't know what a

typical AI data center machine looks

like but it's not like oh it just has

high bandwidth memory it also has DDR5

regular And it also has nan flash like

SSD. So they're taking capacity for all

of those things. And I think the memory

side is what we talk about probably most

because that HBM stuff has such the bad

yield stuff and all that. So probably

like that hit first. But the other qu

the other sort of storage technologies

are also just facing limited supply for

the same reason. Like AI just wants to

buy a lot of it and there weren't enough

people. there wasn't enough slack like

to pick that up. So, I would imagine

that phones will have all of these same

problems. Like they'll in a year or two

when their pre, you know, when their

aotments have run out, uh, they will be

like, "Oh crap." Like the flash storage

is more expensive. The memory is more

expensive. About the only thing that

won't get more expensive probably is

that SRAMM part that's on chip memory

like in a in a Bionic like in a um Apple

A series chip or whatever, right? Um but

they, you know, that's just part of

their memory architect. They still have

the other parts, so they're still going

to have problems with that.

>> Yeah.

>> The Western Digital, uh, which does a

lot of, I assume that Nan Flash, they're

also up just an absurd amount in the

last 6 months, they're up 300 plus

percent. Uh, just today they're up over

10%. Like they're obvious they're

actually they actually have risen faster

than Micron as far as

>> That was because of the podcast, by the

way.

>> Huh.

>> That's because of the podcast, by the

way.

>> We got the motion. Uh, all right. Well,

I mean, I this could be a good time to

kind of jump off some of the hardware

stuff, unless if Luke, you have anything

else you want to talk about because I I

I do have some questions about YouTube

and tech and all that kind of stuff for

you as well.

>> Can I make two stupid jokes before we go

to the next one? Because I had some, but

I just didn't get a chance to put them

in. People were saying smart things.

When we were talking about yield, I

really wanted to ask Luke what his

thoughts were about roundabouts.

>> That's pretty good. Cuz you have to

>> Oh, round. I like roundabouts.

>> Okay, cool. Nice.

>> That didn't That didn't go as well as I

thought.

>> You don't have to yield. You don't have

to

>> yield. You have to yield at them instead

of stop. That's

>> I mean, if you're just confident enough,

you don't

>> No, I like I love roundabouts.

Personally, that's a great point.

Self-driving cars, baby. They just all

They just do the thing. It's fine.

>> In my experience, people stop for me.

>> It's done.

>> That's just cuz you put the LTT thing on

the side and it's coming through.

Yeah. Yeah.

>> I wasn't going to be able to focus for

the rest of the episode. Okay. It was

just in my head the roundabout joke

mostly. So, you had to be there.

>> Yeah.

>> Yeah.

Had to be there. Well, looks like we all

got to be there. That was very

fantastic. Thanks.

>> Brian, ask your next question. That'd be

sweet.

>> All right. All right. we should shift a

little bit of gears and talk more about

kind of uh LT cuz you you guys obviously

have uh some developers on board and all

that and you're doing I assume the tech

and the things that you're building is

much different experience than say

something like Netflix where they've

just been working on one product and

ancillary products for 20 years or

Google 20 right like all these companies

have been just working on the thing

they're doing well Google kills a bunch

of projects too but imagine a company

that doesn't kill a bunch of things

they've been doing one thing and that's

it uh I kind of like even though AI's

had a uh there's a lot of negative

impacts to things. How has it impacted

kind of this hiring what work looks

like? Do you guys feel like you're doing

more things? Like the positives, the

negatives? I'm I'm curious.

>> I don't feel like it's changed like

I don't know. Hiring has gotten really

weird. Um hiring has gotten extremely

strange. Uh, I mean, being like as

public as we are, I guess, um, the

hiring has always been a little bit odd.

Uh, I've I've always been fairly open to

hiring remote devs. Um, so even like I

don't know, six years ago, putting up a

position for remote dev, we'd get 2,000

applicants. Um, and and a lot of them

are, you know, um,

people who are not prepared and are just

like, I want to put an application. and

I'll mop your floors. I don't care. Blah

blah blah blah.

>> It's like, okay, but I actually just

need a an experienced developer. I'm

sorry. Like, I don't know. Um,

>> the camera pans around and there's like

13 people mopping this one tiny piece of

floor. It's like, guys, we don't need

anymore.

>> Come on, write code. Come on. U so you

can sift through a lot of them pretty

fast but like it's it has really taken

like I think especially over the last

year um the influx of what are like

obviously just garbage like crapped out

AI written um AI submitted applications

has gotten crazy u and then sifting

through that it was already difficult

because we had so many applicants for

certain certain positions um if if it's

like hyper local and specialized it's we

don't get that many. Um Canada doesn't

have all that many people. Uh and then

this area is quite expensive to live in.

So it's it's a little bit, you know, if

it's if it's a local position, it's not

that much. If it's global, it's it's a

ton. Um but yeah, sipping through them

trying to find the stuff that's actually

real has gotten more difficult. And the

temptation to like counter sort

basically by okay, well, you're spamming

me with AI applications. Let me sort

through it with AI is high. Um, but then

the error rate there would suck if you

lose someone who's really good because

they got AI sifted out. Um, that would

be very unfortunate. So, there's some

tension there. Um, we have a a coding

test that we do for incoming applicants

of of every type of developer that we've

hired, which is a decent range. Uh,

we've had to make different ones. uh but

they're made in-house and we've ran that

coding test in a bunch of different

models to try to see what that output

looks like. Um and we try to

kind of detect how hard someone leaned

on it. I don't personally care like if

you look at the coding test and you get

AI assistance and you make the best

result out of everyone.

>> I think you made the result out of

everyone. um it doesn't really matter to

me too much, but there is a little bit

of sussing that I want to do to figure

out like did did you just take its

output and just slap it in and not even

uh in in Tea's situation review it

properly? Um did you just did you just

throw it through? Um, and if that's

true, then you know, I'm not that

interested because the even if it did

even if it did one shot super

effectively. Um, I'm concerned about

your ability to maintain things and keep

them good over time. Um, so yeah, hiring

is definitely changed. It was difficult

and now it's way harder.

So I I've been reliably informed by Sam

Alman that uh the problems of AI will in

fact be solved by AI. And so you're

telling me that you you aren't solving

your AI problems with AI?

>> No. Yeah, not so much. Not right now. Um

you know, we'll see if that happens

eventually, but I kind of doubt it.

There's there's there like there's the

basically inherent error rate and for

certain things that's uh frustrating to

deal with. All right. So, I actually do

want to ask a real question here, which

is that Okay, so I think

>> Wait, what were all Wait, what were all

the other questions?

>> Fake questions all those are all those

answers to get to get riled up and go on

a graphic 10-minute rant.

>> Oh god.

>> All right,

>> get wrecked, by the way. Okay, so the I

think the obvious elephant now in the

room, and I'm sure a lot of people would

appreciate this in the audience, which

is okay, I am stuck in effectively elo

hell, right? like you're playing League

of Legends except for it's with your

career and you're a junior and you're

trying to figure out how to get out of

ELO hell because everybody is just dog

piling in and quitting Midame. So, how

do I stand out at least in your process

without giving away too many? Because

then the AI will be like, "Got it." But

like, how would you make yourself stand

out in a sense that when reviewing you

go, "Oh, this is actually really good."

Um, the application I guess is how I'll

take this instead of the the code.

>> I assume the application is the

beginning of ELO hell, right? That

that's the bronze level right there.

>> Yeah, that's how I took that.

>> Would would it be possible just for

those of us who don't who aren't that

familiar with it? Also, what exactly

like what kind of people are you trying

to hire? Exactly. These are programming.

These are primarily programmers and

programmers for

what kind of stuff?

>> A pretty fairly surprising to most

people variety of things. So, so we have

multiple websites. Um, so we have

flplane.com which is the the the first

like thing that we made which is a

>> uh like patreones but video first um you

know creators support platform uh where

people can upload videos have their own

channel and get direct funding from

fans. We also have ltlabs.com

which is where a lot of our

I don't even know if I can say a lot.

Some of our testing data from our local

uh testing lab gets put up on there and

you can do comparison runs and and all

these other types of things which is

kind of neat. Uh then we have somebody

in chat pointed out LGT store. That

one's a lot more basic because it's on

top of Shopify. It's not just like our

own thing. Um but then we have tied that

in with like a system that we have which

is currently called merch messages but

it might be renamed where if we are on a

live stream and you are buying something

on the store as you're checking out you

can leave a a message and that message

will go to the stream to be go to a

stream dashboard. So, it can be curated

and then selected and then um the

producer can just respond to them in

line and then they'll show up in like

the the lower third banner at the bottom

>> or they can curate them and then Linus

or myself or whoever is hosting the

stream um can respond to them directly

through like voice. So, we have that tie

in. We have a couple other things like

that. Flip plane and the LT store link

together for like exclusive sales and

blah blah blah blah blah. Um that's most

of our web development. We also have

internal tools development um which

ranges through a bunch of things. We've

done a bunch of uh you know

internal development on top of snipet

for inventory stuff. We we wrote a guey

for whisper back in the day because our

editors didn't like doing command line

stuff. Um we helped automate things

around the office. Just little things

like that. And then the lab itself also

has a pile of developers um that work on

like data ingest uh both for handling

data from um well it's entirely for

handling data from like benchmarking and

testing of computer components but then

also making sure that like the website

receives the right stuff and not the

embargoed things and whatnot. Um, and

also making a variety of benchmarking

tools like Markbench is the the name of

one of the the tools that we have which

does a bunch of automations for testing

hardware. So like you you put a new

graphics card on a test bench. You run

Markbench. Markbench will kind of try to

verify that everything is set up

correctly and then automatically go

through run all the benchmarks. It'll

it'll set up the gain, make sure the

settings are correct, scroll through the

settings. um and take screenshots so

that if something is strange, if we're

looking at the results, we're like,

"This result seems really weird." We can

go back through and check, okay, was

G-Sync enabled on the on the computer?

It's a fairly common um you know, when

you're trying to test things um error

vector. Um is is this random setting in

the game wrong? Um like is this a game

where we had to manually select the

settings and and the person clicked

beside the drop down instead of on it?

Uh so it didn't actually select the

thing that they wanted. uh or whatever

else because the most frustrating thing

is like, okay, we have five days to like

fully test this suite of new graphics

cards. We're on day four, everything is

done. We notice this one result is super

weird.

>> Why?

>> Right?

>> Like, is it that one result is super

weird or is it all the other ones are

super weird and this one result is the

real one? So being able to go through

and figure out like dive through all the

information make sure everything's all

correct is super valuable. We also do um

some amount of like AI checking to

verify results. We only use that as an

early warning system. So that's running

as test results are coming in based on

our expectations. Um and that will try

to tell us if it thinks something is

wrong early. But that hasn't removed any

human checks. um to to verify all the

data as well. U so we have developers

working on on that tube. So we we have a

a fair bit of range of of people that

we're hiring in the development space.

Um and you asked uh what stands out for

me. It's the same thing personally. It's

the same thing that has kind of always

stood out. Uh my focal point keeps

moving. Sorry about that. Um I changed

my camera setup last night. Very smart.

Um but it it it it's been portfolio

stuff for me always. Like my my

first major specifically developer hire

was very very largely off of portfolio.

They had no post-secary experience at

all. They had no work experience, but

their portfolio was just insane. And I

gave them a way too difficult project.

We've we've drawn this back a lot since

back then, but we gave them a way too

difficult take-home project and they

just nailed it. My Oh man, I am evil for

this. Uh, but my first like take-home

project was make a game um and and

submit it. And I had I had some more

details than that, but it was brutal. It

was very brutal. And he made a um

>> it was actually really fun going through

those submissions. It was very fun. But

it was it was if I remember correctly,

it was it was I attempted to tone it

down so it wasn't like way too

>> AAA game 1 million in sales.

>> No back.

>> It was supposed to be web boost. I think

it was supposed to be Amal. Oh man,

>> roughly GTA,

buddy.

>> I know.

>> Yeah. Speaking of wanted levels in GTA,

you really didn't try hard enough.

>> Yeah. Well, it's I I don't remember the

exact details, but it was web based and

there was like some some stuff that

tried to keep it relatively relaxed, but

my god, that was way too crazy of a

scope. So that Yeah, that never happened

again. But he made a uh competitive

asteroids game. Um, and he set up a a

little bot that set off an alarm in his

house when I joined the lobby. Um, so I

I like loaded the website and joined the

lobby and it like went off and he ran to

his computer and then like fought

against me and there was chat in the

sidebar. So we like talked about the

game and I was just like this is insane.

This has to be the the guy. This is

crazy. Um, so

>> the alarm that is so brilliant of him to

actually create a a physical alarm. Like

that alone is like okay yeah you're

hired. That was a great That's a great

>> The whole thing is just nuts. I'm like I

already knew from his portfolio that I

was like very sold on this person and

then that happened. I was just like, "Oh

my god,

>> you already knew." And then you said,

"Go make a game." I mean, look.

>> I was, you know, I have changed. I've

reformed aware of this. Did you

>> Is he about to get a new lore drop? When

this thing comes out, it's going to be

really

>> Yeah.

>> Do you So, I was strictly out of

curious. Do you pay? Well, can we poach

this guy? Sounds pretty good.

>> I'm not I'm not going to disclose who it

was. We

>> We're not Yeah,

>> we try to do what we can. We're not

super competitive with It's It's tough,

man. The YouTube space is very different

than like Silicon Valley. If uh if

people want to like

>> really really push and be in like hyper

competitive environments, there's more

money out there. Um

>> Yeah, I was joking.

>> Yeah. Yeah.

>> It's just I've been joking, but also

seriously, what's his name? But also, if

you happen to tell me who it was, I

might pass.

>> But also, I have a game.

>> I might pass him along to some people.

I'm not saying

>> he's still on our team. I'm happy about

that.

>> Nice. All right. So, uh, a question I

really am actually very curious about,

which is I I don't know how to explain

this. I sent TJ a frantic message last

night and I've just been feeling it for

months on end about this, which is I

feel a change coming in the specifically

more in the tech YouTube space, but I

don't know what that change is, but I

know it's happening.

>> Wait, what?

>> Well, mean

the means in which people uh consume

content, what their expectations of what

content should look like,

>> what like all those things coming in. I

just feel like there's a change coming,

but I just don't know what it is more in

the tech space. just because there's a

big influx of people. Uh, you know, I

think there's

>> who knows what's going to happen. And so

I'm I'm actually very curious on your

thoughts and really I don't know how

much of the T you can spill on LT, but

if you do I think the second T does span

stand for T, but if like what is your

guys' thoughts about this next year or

these next two years? Is it business as

usual or do you think that there is like

some pivots you may have to make or

changes you have to make to remain more

competitive?

>> Yeah, I can talk pretty openly about

that. Um, one thing question that I

would throw back at you real quick is

that I often hear the the the software

dev space referred to as tech YouTube.

Um, because of you guys talking about

being in tech um for for tech YouTube is

usually like hardware stuff. Um, so when

you're talking about the change you're

feeling in tech YouTube, are you talking

more on the like development side or the

like uh consumer electronics and and

server stuff and whatnot side?

>> I would say either or honestly because I

think I think there is a just a general

uh going to be something is going to

happen here in this space that's just

going to kind of change maybe how people

consume information. I'm not really sure

yet. Yeah, I mean I can talk about some

of the like history of of LTT and and

the the change that we are still going

through uh which has been kind of from

the beginning. So when when Lattis and I

first started working together, we were

part of a computer shop up in Canada

called NCX. Um we would show up to to

like a filming day and in this in the

studio in a corner would just be a a

mountain of products. like it would it

would they the the logistics guys would

just throw them in the corner and they

would literally like pile up sometimes

up like to near the top of the windows.

Like it was it was a lot and we could

just like old school Linusectives videos

and even some of the old school NCX tech

videos um were just like

okay pull something out of the mountain

talk about it while taking it out of the

box end video upload. Uh, and back then

we've kind of retroactively defined that

as like productdriven video where there

was so much product is coming out so

often you're getting a new major

graphics card launch every nine months,

something like that. Um, you could

really just lean on the products are

interesting. Let's talk about the

products. Let's show you the products.

and that that is your content. Around

2015ish

maybe, um it started we started having

these conversations internally about

like we kind of have to

make the story now. Um that's when we

started filming things like Scrapard

Wars, which is like a series Scrapard

Wars like 2013, I think. Um, but we we

kind of saw the writing on the wall a

little bit early, but that Scrapped Wars

is a series where you have a fixed

budget and you go and you try to buy

used hardware and you compete against

each other to who can have the better

system. And that was us trying to like,

okay, let's create a story, let's create

an interesting thing instead of the

interesting thing being the name and the

title that is the new product that's

coming. Um, and

>> was that shift driven by necessity or

was that just you guys saying we could

expand or were you saying like no, we're

like the products aren't that

interesting anymore or there's too many

people doing this? Like how did why did

you decide to do that?

>> Yeah, it's a good question. It's it was

definitely both. Um, so we were you were

detecting a you know a wider release

cadence and kind of less interesting

product releases. Both of those things

kind of happening at the same time. And

also, you know, this is our full-time

thing now. Originally, when we were part

of NCX, this was I was going to school.

Uh, Lionus had a a job as a a a product

manager um or a guy who buys product. I

don't remember what his uh that's a

little different than a product badger.

>> Yeah. Yeah. Yeah.

>> At least on tech, you would buy like

SSDs and sell them through the store or

what.

>> Got it. Yeah.

So, this was our part-time thing. So,

there's only so much we could put into

it. And unboxing stuff was was, you

know, approachable. Now, when we're

doing this full-time, that's that was

probably the main starter for for

something like Scrapard Wars was like,

"Oh, we have time. We could do this

interesting thing." And we always wanted

to promote uh you know, you can buy

stuff used. You don't have to buy the

super flashy things all the time. And

it's been a little bit frustrating for

us forever that like the the you know

the 5090 video is going to get all of

the views and the card that people are

actually buying is not going to get

branded at all because people want to

see the we call it Lamborghini content.

People want to see the the big expensive

Halo products. Um and from there they'll

go like well I this card is really

interesting therefore I want Nvidia and

then they'll go buy the thing that they

can actually afford. Um,

so we wanted to find a way to make it

more entertaining

to see content that is about not just

blowing your wallet on everything. Um,

so that was the main reason for that.

But then the reason why we really

continued doing it, not only did it get

a ton of views, but was because there

wasn't a lot of other content we could

as easily make. Um, GPU timelines for

example.

>> Yeah,

>> I said nine months. That was around when

I first started doing content. Now it's

like two years. Um, so that has been a

sliding scale the whole time. Um, and

also it's

I I find the releases are getting a lot

less interesting. Um, like when the 8800

series GPUs came out, like oh the the

like performance expansion that happened

was mind-blowing and then now it's like

uh well you're paying 50% more dollar

and you're getting 50% more performance.

Hooray. Um

>> if even

>> Yeah, exactly. So it's like it's it's

it's not nearly as interesting. Um so

now we have to try to yeah create

stories, do weird projects, do stuff

like that. Um and I the the

competitiveness of it

maybe a bit. Um Lionus like notoriously

just like doesn't watch YouTube of

almost any form. Um, so like the seeing

what other people are doing and then

doing is

>> not so much of a a thing. Um, he'll read

comments on videos but like doesn't

really watch them which has gotten him

in trouble a couple times but uh you

know um it's I mean I watch a lot of

YouTube but um

>> yeah I don't think like what other

creators in the tech space are doing has

really driven a lot of

a lot of what we do. Um, there are plans

for the future. We want to resurrect

some channels that we have on hiatus.

Uh, we want to start making more videos

again. Um, but a lot of it's like, man,

if the products aren't there, we're

going to have to make something. Um,

that just is what it is. And if you're

trying to release, like if we want to

get back to say six videos a week, for

example, which is what we did for a

decade, um,

>> the vast majority of those are going to

have to be not, uh, you know, new

product review. maybe is product focus

but not new product review type videos

because there just isn't enough

interesting stuff.

>> Very good. Sorry, I was just thinking

about it. I was

>> like waiting for a response or did he

like the answer? Did he hate it? I don't

know. I can't tell.

>> Well, the mustache did not move. It

stayed completely stationary.

>> I thought maybe I I identified with uh

Lionus quite a bit cuz I I too

notoriously have been known for not

watching YouTube. Uh, true. It is a It

is a thing that

>> And I And I like watching YouTube. I can

relate a lot to you, Luke. I can relate

a lot.

>> Yeah, I know. TJ's my Luke. Uh,

>> there you go. Nice.

>> Also, the the important one in the

group, too. Uh, so uh

>> Okay. Okay. That's that's interesting.

Um, I'm just I was just curious about it

just because there's so much content now

that's getting pumped out and things are

even like you know the so much Jeypity

content that's getting uh put out right

now that's even getting millions of