NVIDIA proprietary / Ampere: yserver-hw and MATE start successfully, but MATE control-center windows drag poorly

[ariel3259]

Summary

I tested yserver on Void Linux with an NVIDIA RTX 3060 Ti using the proprietary NVIDIA driver. The result is more functional than I expected from the current NVIDIA proprietary warning in the README.

yserver-hw starts successfully on real hardware, and yserver-mate-hw starts a usable MATE session. xterm, Caja and general desktop interaction work.

However, some MATE applications, especially MATE Control Center / central settings panels, have visibly poor interactive window dragging. The window eventually lands where the pointer stops, but intermediate movement is not smooth.

Enabling hardware cursor with YSERVER_V2_HW_CURSOR=1 improves the general feel, but it does not fully fix the poor dragging behavior for those MATE settings/control-center windows.

Environment

• Distribution: Void Linux

• GPU: NVIDIA GeForce RTX 3060 Ti

• GPU generation: Ampere

• Driver: NVIDIA proprietary

• NVIDIA driver version: 595.71.05

• nvidia-drm.modeset: enabled

• yserver commit: 0849769  

• Rust version: 1.96.0

• Cargo version: 1.96.0

• Just version: 1.96.0

yserver-env-info.txt

Validation

The project builds and validates successfully:

cargo check
cargo test
cargo clippy
cargo +nightly fmt --check

All passed.

Runtime commands tested

LIBSEAT_BACKEND=seatd YSERVER_V2_HW_CURSOR=1 RUST_LOG=warn just yserver-hw

LIBSEAT_BACKEND=seatd YSERVER_V2_HW_CURSOR=1 RUST_LOG=warn just yserver-mate-hw

What works

• yserver-hw starts on real hardware.
• xterm clients appear.
• yserver-mate-hw starts MATE.
• MATE is usable.
• Caja and xterm are usable.
• General behavior improves when using YSERVER_V2_HW_CURSOR=1.

Remaining issue

Some MATE applications, especially MATE Control Center / central settings panels, drag poorly. The visible window movement is not smooth, although the final position follows the pointer correctly.

This looks less like raw pointer lag and more like an interactive move / repaint / Damage / Present / frame pacing issue for specific MATE/GTK windows.

Relevant log excerpt

During window dragging, yserver/libinput logged event processing lag warnings such as:

client bug: event processing lagging behind by 21ms
client bug: event processing lagging behind by 24ms
client bug: event processing lagging behind by 32ms
client bug: event processing lagging behind by 23ms
client bug: event processing lagging behind by 25ms
WARNING: log rate limit exceeded

After correcting input permissions and running with:

LIBSEAT_BACKEND=seatd YSERVER_V2_HW_CURSOR=1

general behavior improved, but the MATE Control Center drag issue remains.

Notes

This report is not claiming full NVIDIA proprietary support. I have not validated GLX_EXT_texture_from_pixmap, Compiz, or full compositor-heavy behavior.

The narrower observation is:

• NVIDIA proprietary on Ampere can start yserver-hw and yserver-mate-hw.
• MATE can run on real hardware.
• Some MATE windows still expose an interactive drag/performance issue.

Void Linux / virtme note

The just yserver virtme target was not usable out of the box on Void Linux:

• Void does not appear to package a Venus / vulkan-virtio ICD equivalent.
• The virtme target failed Vulkan initialization because the virtual GPU did not expose the required Vulkan external memory/fence extensions.
• Therefore, these tests were run with the real hardware target.

[joske]

It's really surprising that on such a capable GPU yserver is lagging. I run on much slower hardware as you can see in my hardware section, and everything is very smooth. There IS an undiagnosed issue with adapta theme in mate. Are you using that one maybe? Try using bluementa.

Also some clients (gkrellm) are known to bring yserver to its knees, also not yet investigated. It seems to generate a damage storm somehow.

Please send logs at TRACE level (will be much slower, but we may be able to actually see the root cause that way).

[joske]

BTW, nvidia proprietary should work fine, it's only texture from pixmap that is impossible with that driver, so no compiz. MATE doesn't need that extension.

[joske]

@ariel3259 please try again, a lot of work was done since your report, lots of bugs fixed and performance optimizations done.

[ariel3259]

Hi, today i pulled the changes and i tested the server, i see an improvement over the first time i use it. I recorded a video with my phone about the test
yserver test video`

[joske]

can't access it

[ariel3259]

I'm going to send the video by this way <https://drive.google.com/file/d/10UaMkW_qSfDu27s2bjNVQOxFoIcMaI3Y/view?usp=drivesdk> VID_20260702_193313818.mp4 <https://drive.google.com/file/d/10UaMkW_qSfDu27s2bjNVQOxFoIcMaI3Y/view?usp=drivesdk> El jue, 2 de jul de 2026, 19:56, Jos Dehaes ***@***.***> escribió:

…

*joske* left a comment (joske/yserver#32) <#32 (comment)> can't access it — Reply to this email directly, view it on GitHub <#32?email_source=notifications&email_token=ASWDAEWSDXI23UWI2SZ2KTT5C3SAPA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIOBXGEYTEMJZGMZ2M4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4871121933>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ASWDAEQLL63VLB2SLHWDCJL5C3SAPAVCNFSNUABGKJSXA33TNF2G64TZHMYTEMRSGU3DONZXGE5US43TOVSTWNBWGYYDEOBTGAYDPILWAI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[joske]

I can see the lag in the video. I don't have this on any machine I have. Could you run just yserver-mate-hw-telemetry and send me the logs (yserver-hw-mate.log and yserver-mate.submit.tsv)? Drag a window for about 30 seconds, then zap.

[joske]

@ariel3259 could you do the above test please?

[ariel3259]

I pull the changes and I see that now it's an improvement. Here's the video with the test
<I'm going to try in a few moments

[joske]

to really fix this I need the telemetry logs.

[ariel3259]

I had a problem with the logs of that video, so i made a test again(ran just startx with mate-session in the .xinitrc)

startx.log
yserver-hw-startx.log

[joske]

There's nothing useful in that log. Don't use just startx, but use

just yserver-mate-hw-telemetry

Then send me yserver-hw-mate.log and yserver-mate.submit.tsv.

[ariel3259]

Made new logs with that comand

yserver-hw-mate.log
yserver-mate.submit.tsv

[ariel3259]

I made another test but without the mate compositor, looks like it works better than i though

yserver-hw-mate.log

yserver-mate.submit.tsv