SimulaVR Compute Pack Progress

8 minute read

—

Calendar

2026-10-14

—

Person

George Singer

—

Pricetags

#update

and

#timeline

1 Progress on our compute pack

We've made substantial progress on Simula's VR/AR compute pack, which will eventually be shipped with Simula One headsets (and which we also want to sell as a standalone "VR/AR compatible laptop replacement computer", when we are able to afford to produce a larger batch of them; if you are interested in this, shoot me an email at georges@simulavr.com):

The compute pack itself comes with Simula pre-installed as a Linux VR distro (built over a NixOS flake), though it should be fairly easy to use whatever distro you want so long as you set Simula as its window manager.

It also has plug-and-play compatibility with some of the existing AR headsets on the market (besides our own). We've been using a Rokid Air (with monado support) as our main test headset:

In our opinion, one of the coolest aspects of VR computing isbeing able to compute with the full power of your entire (private) desktop wherever you want, including during long walks outside or elsewhere:

...so it's been very satisfying demoing Simula's portability in a bunch of different places:

In terms of the exterior of the headset itself, this particular one was SLS printed:

and came as a result of some height adjustments we made to better align ports:

The final iteration will be a bit shorter depth-wise than this one, depending upon the adjustments we make to the heatsink (see below).

The interiors of our compute pack is built around a Ryzen 8840U EEPD NUC, with the following specs:

Our own custom BMS Board (whose assets/code can be found https://github.com/SimulaVR/BMSBoard) carries two battery holders, for a total of four 18650 lithium-ion cells on its front side:

On its back side, it contains chips to monitor and protect the battery cells:

A TI BQ76922 (battery-management chip) monitors the individual cell voltages, current, and temperature while balancing the charge across the 4 battery cells.

A separate TI BQ771807 provides overvoltage protection (shutting off the charging path if a cell exceeds its voltage threshold), while two temperature sensors check whether the battery cells are getting too hot.

Our BMS board is still missing direct charger integration. Currently we've been using an off-the-shelf component for this:

With that said, the next BMS revision is ready for testing (it just hasn't been assembled yet):

It integrates a TI BQ25713 buck-boost charger and keeps the entire combined BMS + charging module at the current dimensions of our BMS board.

When our compute pack runs, it generates a lot of heat. The way this heat gets removed is a via a cooling system (or "thermal stack"), which transfers this heat up and out of the device into the ambient air. Our thermal stack has the following components:

An (AI generated) schematic of our stack looks as follows:

and here is what some of these components look like in real life:

The maximum temperature our Ryzen CPU is supposed to run at is 100℃. But running Simula under even a medium load (e.g. running Simula while compiling Simula), and we're able to hit this way too easily. For example, when we probe system temps using lm-sensors, we see something like the following:

Here we're showing an ~85℃ CPU temp (running at ~15W), but we're able to hit 100℃ quite a lot depending upon what's going on. When this happens, our Ryzen CPU throttles the wattage it runs at downwards, which in turn causes Simula's performance/frame rate to drop.

Another problem with running too hot is that the surface of the compute pack gets...well, too hot. In our case, parts of its surface are reaching as high as ~65℃ under load (see above). For comparison, the surface of an Android smartphone can get as hot as ~45℃ (which is a way more sensible target than where we're at now).

Since we're definitely running too hot, we had to figure out why this is. Generally, the problem could be at any layer of our thermal stack. But through heat testing, we were able to determine the two biggest sources of heat friction were (i) coming from the CPU die itself (which we can't control) and (ii) from the heatsink to to ambient air junction (which we can control).

So the way to fix this is basically to increase the surface area of the heatsink (by increasing the heatsink "fin" density), and possibly also by changing the material from aluminum to copper (which will add to the compute pack's weight as a tradeoff).

We've run simulations on different heatsink designs and have found a new one which solves the problem. So we're in a position to fabricate a new one for testing pretty soon.

The Simula One compute pack is of course part of our Simula One headset. So here's how it fits in with the other components, in broad strokes:

If you're interested in purchasing/demoing a compute pack, shoot me an email at georges@simulavr.com.