Studio GPU acceleration

[zkeyg]

Hi,
I would like to understand more about how to enable GPU acceleration for Foxglove. It appears that when we run foxglove on our machine (x86 linux ubuntu22.04, Nvidia RTX2000 series, Driver Version: 535), it has a high CPU usage whereas GPU usage remains unchanged.
We followed the doc to enable WebGL on Chrome but it seems it did not improve the performance. Is there a way to tell if the application is using GPU? And is there another other flag I can set for GPU acceleration?

Thanks!

[defunctzombie]

What do you see in chrome when you visit chrome://gpu/ ?

[zkeyg]

Hi @defunctzombie
Thanks for the reply! We used the official document and configure the chrome://gpu. It listed WebGL as hardware accelerated.

• Canvas: Hardware accelerated
• Canvas out-of-process rasterization: Enabled
• Direct Rendering Display Compositor: Disabled
• Compositing: Hardware accelerated
• Multiple Raster Threads: Enabled
• OpenGL: Enabled
• Rasterization: Hardware accelerated
• Raw Draw: Disabled
• Skia Graphite: Disabled
• Video Decode: Hardware accelerated
• Video Encode: Software only. Hardware acceleration disabled
• Vulkan: Disabled
• WebGL: Hardware accelerated
• WebGL2: Hardware accelerated
• WebGPU: Disabled
• WebNN: Disabled

[defunctzombie]

So that looks like the right things are enabled. Do you experience lag using the app when trying to visualize items in the 3d panel? Is that why you are wondering whether GPU is being used? CPU usage is going to depend on what else you're visualizing in the app.

[zkeyg]

Yes @defunctzombie We experienced lagging while visulizing 3D Pointclouds and wanted to see if there is a way to make sure Foxglove is using GPU.

[tonynajjar]

Hello, I'm also experiencing performance issues whenever I'm publishing e.g 2000 Sphere/Arrow Primitive + 2000 TextPrimitive.

Both WebGL and WebGL2 are hardware accelerated as mentioned here.

I can see that the second process (PID 49736) shows up in nvidia-smi but not the first one, which uses the most CPU. Is this expected?

My GPU is RTX 4060

[defunctzombie]

Thanks - I've ticketed this for us to investigate. If you can share this sample data it would be helpful for us to understand if its specific to a kind of data. We can make sample data but its not always an exact match for the data you've produced.

[nthiad]

I've attempted to replicate this performance issue and it could be related to how updates are written in the mcap.

Here are some scripts I used to generate mcap data of a similar scale (2000 arrow and text labels) where the fast version contains all the arrow and text primitives in a single SceneUpdate message per update and the slow version wraps each arrow and text primitive individually in a SceneUpdate:

https://gist.github.com/nthiad/2a13d2e1003814bca5e7e3052192e51d

If you likewise notice that fast.mcap performs well and slow.mcap performs poorly, then the problem may lie with how the data for SceneUpdates is formatted across scene primitives. If fast.mcap performs poorly, we can perhaps narrow in on some other cause but using a common benchmark.

Here are download links for the mcap files:

• fast.mcap (70M)
• slow.mcap (100M)