The autotldr-bot only summarized the first page, so here are some more quotes. Basically, the performance was almost identical, with two notable exceptions.
Across a variety of demanding GPU benchmarks the NVIDIA R550 open kernel driver continued to perform on-par with the proprietary driver for these GeForce RTX 40 graphics cards.
While running Blender 4.0 the proprietary kernel driver seemed to yield slightly better performance. It was just fractions of a second but was rather consistently showing the proprietary driver having that slight advantage here, unlike in other workloads.
There was the small advantage too that during periods of brief downtime using the open kernel driver appeared to deliver slightly lower GPU power consumption than the proprietary driver.
Does anyone have an idea what’s the point with the proprietary driver now? Does it have any features missing in the open driver?
Does it have any features missing in the open driver?
You can see the differences in the official README here: https://us.download.nvidia.com/XFree86/Linux-x86_64/550.54.14/README/kernel_open.html
The main difference for me is the inability to preserve video memory during suspend and hibernate. Without it, sleep and hibernate will cause all sorts of weird graphical glitches upon resume.
You have to use the proprietary driver with GTX 1000 series cards and older to get good performance. 900 and 1000 series cards are cucked with no re-clocking on nouveau because they require signed firmware.
You’re right, the new open source driver does not support the 1000 series and older, only Turing (2000 series) and above.
And 16 series, but that is also Turing.
This is the best summary I could come up with:
Previously the open-source (out-of-tree) kernel modules were just certified for their data center GPUs while now they are basically acknowledging that they are in good shape too for GeForce and workstation products.
It was back in May 2022 that NVIDIA announced their open-source GPU kernel driver plans for Turing GPus and newer.
Since then they’ve continued advancing this out-of-tree code that is bundled as part of their distributed driver package.
NVIDIA’s open kernel modules option is NOT to be confused with the upstream open-source Nouveau driver effort.
But there are some known limitations in the open kernel driver with not yet having G-SYNC support on notebooks and vRAM is not preserved across power management yet.
During the benchmarking not only was the raw performance analyzed but also the GPU power consumption between these alternative kernel drivers.
The original article contains 440 words, the summary contains 136 words. Saved 69%. I’m a bot and I’m open source!
