18-Way GPU Linux Benchmarks, Including The Radeon RX 460 & RX 470 On Open-Source

[dungeon]

Because ask most AMD Linux users what driver they use... open-source.

Ask most NVIDIA Linux users what they use... proprietary driver.

Easy on those "asking", overhalming majority uses 1920x1080 resolution, but you mostly do something else

[dungeon]

In my opinion, all benchmarks (of any product) have to show the best performer in the field. Otherwise the benchmark data would be just about local maxima. Many people want to see the global maxima because they contain information about the current state of the art in the field.

There is no global maxima in the land of non default nVidia variabile games used and AMD profiles used... perf really goes there from the case to the case

[dungeon]

Basically what AMD proprietary drivers do, is that profiling... while their default is stable, flat or call it normal default performer. They mostly implement sporadic feature and enable it on particular game, while global improvement also happens sometimes but much, much less.

While with AMD opensource drivers as they do not do profiling of that kind, they try to optimize something while many times breaking something else 100 testers with different GPUs are needed on every commit

[dungeon]

So, if you are AMD proprietary driver user they don't need you much, user there can just request profile ... while AMD opensource drivers is... simple story is sort of uncle Sam story - I wants you to fill a bug

[Delgarde]

I still don't understand one thing.. Why are AMD cards tested with open source drivers and Nvidia with proprietary ones? This comparison is pointless - either I want to be using open drivers and I want to see how Nvidia cards perform under them.. OR I don't care, so I will be using proprietary ones and want to see how AMD performs with them!

Then go and find a benchmark that compares NVidia performance on open vs closed drivers. This is a benchmark of different cards on their most commonly used drivers - not a comparison of the drivers themselves.

[Kano]

AMDGPU-PRO seems to be a bit tricky right now. If you want to use both mesa and binary stack in parallel then AMD has to provide all source code updates like for libdrm. I really don't understand AMD to ignore that completely. In theory a little wrapper setting some env vars should be enough to switch on the fly between both but nobody tested this yet.

[haplo602]

I still don't understand one thing.. Why are AMD cards tested with open source drivers and Nvidia with proprietary ones? This comparison is pointless - either I want to be using open drivers and I want to see how Nvidia cards perform under them.. OR I don't care, so I will be using proprietary ones and want to see how AMD performs with them!

The comparison is very valid for most users.

[drSeehas]

... there are already benches with closed AMD closed driver vs NVIDIA driver in Phoronix.

Where can I find benches with closed AMD driver for RX 460 in Phoronix?

[pewspewpew]

to people who are wondering about mesa vs nvidia proprietary:

• no real reason to test nvidia on open source, as author wanted to compare experience with cards, not start a driver war
• no real reason to tet amd on proprietary, as amdgpu-pro (maybe due to beta) fares mostly same as mesa, some things are even slower now

I think this comparison is for the best for people that choose a card for gaming on linux, it gives the real-word results, not the honest ones

[dungeon]

Where can I find benches with closed AMD driver for RX 460 in Phoronix?

That link is probably lost somewhere in future