CPU usage scales with download speed and reaches very high usage at high speed

[VictorVow]

Is it a bug?

• I know this is an issue with the app, and contacting Mullvad support is not relevant.

I have checked if others have reported this already

• I have checked the issue tracker to see if others have reported similar issues.

Current Behavior

When downloading at high speeds (e.g. Steam games) Mullvad is using a high amount of CPU (e.g. 85% according to btop).

Expected Behavior

It should be using a low amount of CPU even at high download speeds.

Steps to Reproduce

1. Download at high speed e.g. a Steam game
2. Check the CPU usage using a utility like btop

Failure Logs

Operating system version

Kubuntu 24.10

Mullvad VPN app version

2025.5

Additional Information

No response

[hulthe]

CPU usage scales with download speed

This is expected and unavoidable. More data = more time spend e.g. encrypting and decrypting it.

and reaches very high usage at high speed

There is always going to be some overhead to using a VPN. Whether it's noticeable depends on which features you're using, and how powerful your computer is. If you can, please post a list of the features you're using (wireguard, openvpn, shadowsocks, daita, quantum, etc...) as it will help us know where to direct our efforts.

[VictorVow]

Am using wireguard, DAITA, automatic obfuscation, automatic quantum.

[Serock3]

Thanks! Both DAITA and obfuscation could contribute to the increased CPU usage (quantum resistance should not have any effect after the tunnel is established). Could you try enabling DAITA, Shadowsock obfuscation and UDP-over-TCP Obfuscation one at a time to see if any in particular makes up the bulk of the CPU overhead?

[VictorVow]

Testing with an internet speed test site I found that:

Shadowsocks = minimal CPU usage
UDP-over-TCP = moderate
DAITA = high

[Serock3]

Thanks for taking the time to test. When you set obfuscation to "Automatic", then the app will prefer shadowsocks over UDP-over-TCP, so likely DAITA is the main offender for you issues. I suspected as much, but it's nice to have it confirmed.

We have projects in the pipeline that will hopefully improve the performance of DAITA, but nothing specific that I can share at the moment. I'll keep this issue open to track engagement of the DAITA performance issues.

[MarkusPettersson98]

There's not much more to discuss here, so I'll close this thread for now. Hopefully we'll bring some optimizations to DAITA specifically in the near-ish feature 😊