Valve Steam Frame Engineering Deep-Dive: Water Cooling, Thermals, Power, Acoustics

Valve announced the steam frame steam machine on steam control yesterday.

We covered all that in detail, but we have more.

This video goes deeper

on the engineering of the Steam Frame VR headset specifically.

Normally, we don't record every minute of our time with the company,

but valve was cool with it, and we had two separate deep dive

discussions with Valve's engineers about the frame VR headset.

We didn't want to just let that great, highly technical content go unseen.

For example,

All of those vias are actually working

as a thermal transfer through the PCB itself.

So in this video we'll talk with Valve's engineers

about some of the acoustic considerations.

some of the thermal solution design

So when we told Michael that we wanted to flip the maps inside it,

he immediately started thinking about those power management ICS.

And it's like, oh man, you're killing me. This is the worst thing ever.

and some of the PCB design and layout challenges.

We're not going to heavily edit and produce this, so you're going

to get basically the raw conversations that were just us talking.

Not intended for camera,

but we captured it anyway just in case it was useful.

And it was.

Valve made sure to have engineers from basically

every aspect of the hardware available, which is awesome.

And so we were able to get pretty much all of our questions answered.

These are fairly casual conversations, a lot of detail

to learn about the relevant sciences and Valve's hardware engineering process.

So let's get into it. Before that.

This video is brought to you by Haven and their new Bf 360 flow case.

The Haven Bf 360 flow is a higher flow case with a focus on performance,

with the model that includes fans using 200 millimeter front intake fans

at 40mm thick in combination with a one 20 millimeter rear exhaust fan,

and our previous testing of the case, we found its thermal performance

to be among the best performers on our charts.

The BF 360 uses a scooped trout near the GPU that our smoke test

found deflects the air into the left most GPU fan for more targeted GPU intake.

It also has marked cable management channels, rubber, graviton, most

metal contact points and what we think is overall good build quality.

Learn more about The Haven BF3 64 at the link in the description below.

maybe a year ago how basically we sat down and we said, okay, how can we really,

really squeeze the maximum, thermal thermal output from this guy

and so we literally took the board, mounted it up to a water cooler.

Okay.

Looked at actually the thermal performance of the heat

going out of the backside of the board, as well as actually directly off the.

So it's pop memory.

So you essentially have the main

SoC plus the memory stacked on top of it.

That's actually going to impede the, the, the heat.

Is it vertically stacked.

Yeah. On the silicon. Yeah. So this is a package.

Package. Yeah. Memory chips. Orders on top.

Okay. Cool.

Yeah.

So essentially you can think of it as, you know, two direct paths

where he could go either through the, the pop memory

up into a heat sink, or you could actually go straight backside.

Yeah yeah.

And and if you think about it, like in the case of C

with a ton of, vias, right?

All of those vias are actually working as a thermal transfer

through the PCB itself.

so essentially what we did is we said, okay, let's just throw out this idea here.

Maybe we can actually pull more heat out of the the backside of the board.

Yeah.

So we moved a number of components around so that instead of there

being a number of capacitors, inductors, all of those components,

let's kind of shove them all to the side and then create these big pads

that we can actually clamp an additional piece of metal to.

Right.

And then that would be where we'd actually connect a second heat's,

heat pipe to on the backside of the PCB is, is it relatively bare there.

Well on the back this is actually the backside of the PCB.

That you're looking at right there.

Is this,

thermal putty under the, yeah, the cold plate.

Okay.

And then I guess,

eight millimeter flat heat pipe, 6 or 8.

Sounds like a roughly correct number.

I can I can check on that. Anything.

Is it just a thin stack up here?

Is there any any, opponents, you know, no components.

Okay.

Just a thin bank that's been, you know, welded essentially to that, All right.

Okay. Are you solder into the heat pipe for welding? Yeah.

Okay. It's directly connected. Oh, okay.

the most part, the, the, aluminum side,

the other heat pipe, there's a cutout that goes down and touches that.

Oh okay.

And then that heat pipe is bonded to the, got it to the to the middle.

To the minimal. Yeah, yeah. Cool.

Very cool.

Yeah. Yeah.

So basically, you know, you can imagine a scenario of different,

kind of ways that people would connect the thermal solution.

Right?

These types of socks, one would be no thermal solution at all.

Then maybe they would have a heat spreader

and then maybe they would have maybe a heat pipe that would actually

go to some other place, you know, whether it's a phone.

Right.

Something like that.

In this case we're basically saying dual heat pipes.

We're saying, you know, a big fan that's going to be moving a ton of air.

That fan can run, like, get you some more precise numbers or whatever.

But it's, it's it gets loud, right?

Like if you're running it, if you really run it. Yeah.

We're going to have a sweet spot that's going to be

you're going to be happy to, you know, running it as is.

But if for power users that are like, I want to just max this thing out

so they can do that, that'll be an option the question

I'm gonna have at the end of this is, how much of the air flowing through

other location on the device

is to cool sort of periphery, components on the PCB.

But, yeah, the two considerations would be you're

obviously limited on sort of total porous area.

You can even pull air through to begin with. True.

And so puncturing everywhere lets you pull more in just in total.

But then yeah, secondarily with the, the deck OLED in particular,

I remember, talking about how the OLED had moved

some components around to better guide air as it comes through the, the inlet.

Yeah, across those components.

that's definitely a consideration. Yes.

So is there anything a heat sensitive particularly beyond the SoC.

Yeah. Yeah, definitely. So there's there's a Wi-Fi chip.

That's, depending upon what,

kind of mode you're operating in, it's kind of put off some power.

There's also,

when you're looking at the power delivery network that actually goes,

you know, from the battery ultimately, through the various,

regulators that are taking that battery voltage and dropping it down

and delivering the power into the, SoC, those are not 100% efficient.

And so some of the components that they're actually we're touching off, right.

There are also the, power management ICS that are,

you know, this is stuff that Michael spent an ornament inordinate amount of time.

So is there any trying to pull heat out of those as well?

I'm glad you mentioned because I wasn't sure

if I should even bother asking or not, but, you know, VRM is often like,

I it's in a lot of devices.

It's almost a commodity where the VRM is

the voltage regulator is just kind of like a given.

And it's not super interesting.

But is there anything, you know, particularly special or that required,

more creative engineering, I guess on the VRM side,

it was definitely a packaging challenge.

Okay.

It's, a lot of power that we're able to, to generate.

So, get everything tight.

There's, a lot of, impedance requirements.

Okay.

So everything wants to be really close to where it has to go.

But you can't really concentrate all of that heat.

So it's definitely a packaging exercise okay.

Sure.

So when we told Michael that we wanted to flip the maps inside it,

he immediately started thinking about those power management ICS.

And it's like, oh man, you're killing me.

This is the worst thing ever.

You're going to have to move on.

Yeah, yeah, you killed my dog.

So what is that?

How does that, what does your job look like?

I guess when you are, trying to figure out

how to meet the packaging or the density considerations.

Are you in software?

Just moving things to see what fits, you know, are you running calculations?

A little bit of, there's a lot of simulation.

Okay.

We're sure, especially for, impedance requirements.

The temperature pieces are a little bit

more challenging to simulate for, for some of that.

Right. Because

it's only as good from a sim as the,

the whole system, which is relatively difficult to simulate.

But yeah, kind of all of the above it.

Yeah.

So along the way, what we did, you know, because we're just like, hey, you know,

we're an engineering company, right?

We're just like, this is the kind of stuff we really want to, like, nerd out on.

It's like for like, how many temperature sensors can we add in them?

You know, can we ship with those?

You know, how many power, measurements can we get in line?

And the answer is we put a whole bunch of them in there.

Michael can probably give you exact details,

but the general idea is, you know,

we're we're able to measure the power, on these Qualcomm SoCs

you typically have, there's a series of temperature

measurements that are in the SoC itself a ton.

And those are really accurate.

And you basically need those for everything.

And then there's also the ability to, breakout power rails

that are dedicated to different, CPU clusters as well as, okay.

So you can imagine you've got, you know, one CPU cluster that would be,

the cores zero and one,

and there's a different, CPU cluster, four cores,

two, three and four, and then a different one for five, six and seven.

And those three different cores have dedicated

power rails to actually put inline power music.

Okay. For each of those, for the CPU.

And then also the same thing for the, for the GPU.

And then I kept telling Michael, like, hey, how many more can we get?

Yeah.

So we have a number of other ones as well that we've broken out.

So we really have a pretty good map

of the power flow in the device as far as like efficiency.

You know,

how much power is actually making it into each of the different CPU clusters.

And so when we're

playing a game or testing things out, we can say, okay, this game is CPU.

You know, intensive.

This game is GPU intensive.

Here's what the curve looks like when we're tweaking the GPU or CPU frequencies.

Here's what it looks like from device to device, you know?

So yeah, just getting tons and tons of data.

And then we're also getting all the temperature data on top of that.

Yeah.

How many sensors would you say we have for the for the thermistor.

So placed externally

not directly on the SSD itself.

I don't remember the exact number.

Like ballpark.

It's like ten thermistor and probably like 30 di temperature.

Okay. Like scattered throughout.

Yeah. But taller.

Yeah, a pretty good map.

is the the fan basically model.

Yes. And responding to those thermistor.

Yes. Okay. We have a totally custom fan curve.

You know, one of the great things with, the open source model is,

you know, this is running Linux.

It's less, you know, we write all of the,

the kernel drivers, you know, we're acting with the folks upstream.

Yeah.

We're we're benefiting from what we've learned on the Steam Deck

as well as Steam Machine.

So yeah.

So we have algorithms that are running that are essentially reading all of those

tall data points and saying, okay, which one is going to actually

give us the best data for this type of a load?

In other cases, we might say, you know, we could have different profiles,

you could have a variety of different things.

So that's stuff we're actually working on right now.

Come on.

We've even been doing

some airflow studies, you know, flowing through the device.

We, I'm not sure we want to show on camera exactly yet, but we do have,

some stuff that we could just show you to kind of give you an idea, right?

Of, the types of stuff that we've been doing.

The smoke testing was mentioned earlier.

Yeah, yeah. That's cool.

any other special challenges, things that you want to draw attention to or.

Yeah, things that you overcame that you want to point out.

I mean,

that one's probably like the most complicated,

I think, in any device that, you know, has this many things coming

together, we've really had to optimize the,

the data flow essentially from the cameras, from the displays

or going obviously to the displays, the wireless link.

And really all of this is there's a lot of demands on the memory.

Right right.

You have it's.

Yeah, like I said, Intel PDR five x, it's,

you know, 40 to 66 as we were discussing,

and, you know, so you really have to optimize all of that.

So you, you know, if we're going to be,

running the cameras at the same time we're running the GPU and also running,

the CPU as well as the displays, all of that puts a demand on the memory.

So that's been a big thing where we've had to really,

do an awesome job there I guess you've kind of got almost two thresholds

thermally, you have the threshold where the SoC and the other components

cannot operate above a certain temperature.

Right.

And then you have whatever the human like, the user's, sensitivity

is to the device on their face.

So do you, do you, do you trip the SoC sensory

threshold before the user threshold is a problem?

Meaning? Yeah.

All right. Interesting question. You get where I'm going with that.

Yeah.

Yeah, yeah I would say that, I mean, we're using these devices every day,

pushing the limits of these all the time. So,

it's hard to say where a

particular user would be saying, okay, this is uncomfortable to wear, right?

If we were running it at the absolute max, level,

but we're doing a really good job of decoupling thermal,

interface that's up here from the user.

Right.

So and that's the type of stuff that, you know,

if we were putting it on our head and saying, yeah, this is uncomfortable,

then that would definitely be, totally unacceptable.

So, you know, that's that's the type of thing where,

you know, we don't believe that that's going to be a common

complaint from users, because of the way that we're.

Yeah, but if you weren't controlling for that, for sure, that could be,

problematic. Right?

Yeah.

So we're definitely I mean, it's going to get warm

just from put in your face and goggles, you know,

and so that's that's kind of what I was curious about is like, does

the rest of the device contribute meaningfully?

Yeah. I mean, there definitely is a contribution.

You know,

I think that that's a little bit TBD as we're right, locking things down.

But definitely that's a huge consideration to us as comfort.

Right.

So we're definitely going to be saying

this is going to be something you're going to be happy

to have on your face for a long time, right?

So cool.

but I would say that one thing that, you know, we worked on pretty extensively,

is if you think about all the power that we're trying to, to put into this

for the SoC, that power delivering network from the battery all the way through to,

you know, driving the displays and driving the AC, you know, as you can

imagine, with anything where you're trying to get a lot of performance out of it,

you know, you look at a motherboard and it has all of these crazy capacitors.

Yeah, you know, all of the stuff that's really closely located

putting it into a, you know, head mounted form factor.

Yeah.

And having it be a battery, you know, power device a lot of challenges there.

So this is the kind of stuff

that Michael here, you know, definitely I would guess, you know, had to look at.

Yeah.

All that kind of stuff, you know,

on some of the denser boards we've looked at like, the RTX Pro 6000.

Blackwell Ford is a really good example of density of power components.

And, I know for them

the unfortunately, I guess in one way,

the solution to solving the we don't have enough space problem

was unfortunately spend more money on better parts

because that was the way to get the density of any of the, you know,

smart, smaller inductors that do the same work, things like that.

So yeah.

And we definitely, have been this is on the premium side

of those types of components especially.

It's like this is, you know, the type of thing

where we're saying, hey, we want to be able to run this thing

flat out, you know, and we don't want any limitations in there.

And ultimately, with it being a head mounted device,

there are going to be some thermal challenges, of course.

But, you know, that's been, you know, top of mind from the beginning.

Right.

What, what how long have each of you been working on this.

So I was, I worked on the index.

You know, one of the reasons, I work a lot more now on steam OS type

stuff, but, I worked on the main board a lot on, on index,

and then immediately started working on this, so.

Okay.

A while.

Yeah. Six years space. Wow. Okay. Yeah.

Just over two years.

Okay cool.

So. Right. Cool. Yeah.

Yeah. But, yeah.

Kind of typical thing at valve.

Working on things for a long time.

And, when it's ready, then it's ready when it's ready.

Yeah yeah.

Cool.

That'll be a good one for, We'll close that out here, I guess.

I appreciate it.

Thanks for thanks for. All right. Thanks a lot. Yeah, yeah. Appreciate it.

Thank you.

So that covers it for now.

For the Steam Frame VR headset engineering discussion.

We have a lot more from this trip though.

So we met with one of Valve's engineers who worked on the Steam Deck,

and he has some really cool discussion about the deck versus the deck

OLED, about PCB layout and engineering and design, things like that.

Challenges they ran into.

It's not about a deck two, but it is super interesting

information, really fun discussion that'll go up separately.

And we'll look around and see if we have any other interesting

clips to put online too.

But for now, that pretty much covers the Steam Frame

VR solution, and we'll have plenty more to talk about with the Steam Machine,

and probably with the Steam Controller as we approach the actual launch in 2026.

But in the meantime, make sure you check back for the Steam Deck conversation.

Our policy is to pay for all of the travel and hotels ourselves

for all these types of events.

So if you'd like to support us in doing that,

to maintain our independence when we're doing reporting in the field,

you can go to stalk arrows, access

dot net And grab our new retro IO themed coaster packs, complete

with 5.25in drives depicted on one of them in a floppy disk on the slot.

and old school controller with the original colorways.

An IBM model M inspired keyboard. Numpad.

And a retro style GM logo.

All of these are coasters. In the coaster pack.

You get one of each of these,

Or you can go to basically what's, self-funded Kickstarter style campaign

we have for our AI surveillance dystopia coverage coming up, where we have a whole

deep dive series digging into fans like the Nvidia Palantir relationship.

A raid of a normal person's home over a development console.

and multiple other investigations like GPU export control violations

and things of that nature.

So head over to the store, store our cameras, Nexus dot net

and check out our AI Dystopia Kickstarter style

fundraiser we're doing to help produce all of this great content

and start putting together the travel, especially for our road trip around the US

for several of the data center site visits.

That's it for this one.

Thanks for watching.

Subscribe for more like the Catch the Duck footage

and we will see you all next time!