Windows 10 vs. Linux 4.15 + Mesa 17.4-dev Radeon Gaming Performance

[papajo]

if you value speed, what are you doing on linux anyway? speed takes bigger hit from switching from windows to linux than from nvidia to radeonsi

There are things an OS is useful other than gaming and I also have personal reasons to want to support linux but that doesnt mean that when I see BS I am not going to call it.

I paid close to 300$ for a GPU which architecture exists more than 2 years now (so plenty of time for the company to develope functional perfoming drivers for it-which is something they should have done prior releasing it anyway and not after the fact-) and which company still does not supply basic service such as freesync,audio over hdmi,relive etc and on which the drivers (open or closed) suck balls out of the box and need great research and tinkering to just make them suck less (in terms of gaming because when you pay 300$ for a GPU unless you are a miner or a scientist you probably do so for having better experience in gaming and that is what you get not with recent amd products on linux at the time we speak. period.)


Edit: + exactly because you get a great performance hit when switching on linux they should try to make acceptable linux drivers because there is no room to spare for halfassing your job (I am talking towards AMD any unpaid generous devs who just participate to contribute you have my gratitude )

[papajo]

so you hate amd because you have some ridiculous pictures in your head? apparently amd does, but community didn't write driver which is consistently better than nvidia blob yet. it is only sometimes better.

Ignoring all those childish and stupid insults. show me in which game it is better.

maybe because most of opensource developers working on amd driver are employed by amd(apparently contrary to pictures in your head) and amd has limited resource because it has smaller market share and it currently does two drivers?

That's BS so if you go to buy a shoe and pay 80$ for it and it sucks you are going to be like... oh its ok this shoe I bought has already worn off and is not as good as the 80$ nike shoe and thats is fine with me because the shoe I bought is made from a smaller company which can not compete?

Do you even listen to yourself?

If they cant make competitive drivers then they should sell their products for less. I paid close to 300$ for a card and I expect 300$ card performance.

[bridgman]

Ok could you please point me to a specific download link to download a specific kernel you consider to be ideal? because there are like 30 different branches for 4.15, Thanks.

Sorry, just saw this. Probably best bet would be the Ubuntu 4.15 rc2 kernel build at:

http://kernel.ubuntu.com/~kernel-ppa...ine/v4.15-rc2/

You will need to specify amdgpu.dc=1 in the grub command line.

EDIT - from the discussion in the 4.15rc2 thread these may be debug builds - not sure offhand what the perf impact would be. Probably worth giving it a try anyways though.

Prior to 4.15 the best option would have been building from the most recent "highly active" amd-staging-<version> tree, currently amd-staging-drm-next. I say "highly active" to distinguish between a just-created branch (with one or two developers working at rebasing the previous staging branch onto a new kernel tree) from the branch where most of the developers are working.

Another option would have been installing the latest AMDGPU-PRO kernel, since that has had DC and HDMI audio since Polaris launch, although those builds have targeted enterprise/LTS distros and so won't always work with latest Ubuntu non-LTS.

New subject - I am looking through the "cpu bottleneck" posts now but not really understanding the point you are making.

If I understand correctly you are saying that since no single core is at 100% you are not CPU bottlenecked, but from what I have seen the scheduler tends to spread load across cores so once your "sum of all cores" utilization gets anywhere near 100% (ie anything over ~10% on each of your 8 cores) you are starting to approach the point where CPU single-thread performance can impact FPS depending on how single-threaded the game engine is at that point. Unfortunately Linux ports tend to be more single-threaded than the original Windows games, although this seems to be starting to improve.

From what I could see (some pics didn't show all cores) your CPU utilization was at or above that point in all cases.

Michael posted another article which included both NVidia and AMD results for Linux vs Windows, which helps to distinguish the performance differences resulting from "native vs port" from the differences resulting from "Linux vs Windows" - worth a read if you have not already seen it:

https://www.phoronix.com/scan.php?pa...10ubuntu&num=1

[VikingGe]

And to put an end to any doubt here is The Witcher 2 running on high settings at 20 FPS

That is certainly not normal, getting 50+ most of the time in that area (screenshot). That's like ~85% of native Windows performance.

Settings used: screenshot

[papajo]

That is certainly not normal, getting 50+ most of the time in that area (screenshot). That's like ~85% of native Windows performance.

Settings used: screenshot

In that spot I get about 40 FPS tops could you go in that exact spot again please only this time having also the draw-calls parameter enabled in your gallium-hud?

EDIT: lol I didnt check on your settings I use even lower ones!!!! https://i.snag.gy/eaEAVn.jpg ... So I have a slightly better CPU slightly better GPU but FPS are lower by a lot... it has to be software related or I have some sort of hardware malfunction...

I use MESA 17.4, LLVM 6 ubuntu 17.10 64bit with kernel 4.15.0-041500rc2-lowlatency/DRM-NEXT 4.15.0-996-lowlatency (same avg fps with either) and when booting both in wayland or in xorg I see no difference (actually wayland is more jittery in some games but avg stay the same more or less)...

So what else different do you have? Maybe you dont run gnome? I run gnome 3.26.2 but this shouldnt matter that much really..

Ill check my powersupply.. maybe that could be the issue wtf though... (still please add on the gallium-hud line the draw-calls parameter its important to me to see if you have more or less draw calls than me on that spot. thanks )

[papajo]

Sorry, just saw this. Probably best bet would be the Ubuntu 4.15 rc2 kernel build at:

http://kernel.ubuntu.com/~kernel-ppa...ine/v4.15-rc2/

You will need to specify amdgpu.dc=1 in the grub command line.

EDIT - from the discussion in the 4.15rc2 thread these may be debug builds - not sure offhand what the perf impact would be. Probably worth giving it a try anyways though.

Prior to 4.15 the best option would have been building from the most recent "highly active" amd-staging-<version> tree, currently amd-staging-drm-next. I say "highly active" to distinguish between a just-created branch (with one or two developers working at rebasing the previous staging branch onto a new kernel tree) from the branch where most of the developers are working.

Another option would have been installing the latest AMDGPU-PRO kernel, since that has had DC and HDMI audio since Polaris launch, although those builds have targeted enterprise/LTS distros and so won't always work with latest Ubuntu non-LTS.

New subject - I am looking through the "cpu bottleneck" posts now but not really understanding the point you are making.

If I understand correctly you are saying that since no single core is at 100% you are not CPU bottlenecked, but from what I have seen the scheduler tends to spread load across cores so once your "sum of all cores" utilization gets anywhere near 100% (ie anything over ~10% on each of your 8 cores) you are starting to approach the point where CPU single-thread performance can impact FPS depending on how single-threaded the game engine is at that point. Unfortunately Linux ports tend to be more single-threaded than the original Windows games, although this seems to be starting to improve.

From what I could see (some pics didn't show all cores) your CPU utilization was at or above that point in all cases.

Michael posted another article which included both NVidia and AMD results for Linux vs Windows, which helps to distinguish the performance differences resulting from "native vs port" from the differences resulting from "Linux vs Windows" - worth a read if you have not already seen it:

https://www.phoronix.com/scan.php?pa...10ubuntu&num=1

Too late thanks though I already tried my luck here http://kernel.ubuntu.com/~kernel-ppa/mainline/ and downloaded first the 4.15 rc1 then the 4.15 RC2 came out then I tried the latest NEXT-DRM (all low latency).

The low latency version is a must with MESA 17.3 ALL the hiccups and jitter was gone and everything went buttery smooth ^_^ (although no significant FPS boost)

Then I tried the unstable padoka and got all the goodies starting with 17.4dev but (although not as bad as before) it brought back some small jitter and again 0 to no significant FPS boost (audio over HDMI is not my concern it just bothered me for not being there, mainly I get bothered for not being able to use freesync and VCE)

I would have posted earlier (I planned on posting actual video footage of my benchmarks etc) hoping that I would be able to get AMD VCE working but unfortunately it still doesnt work... and OBS even on light settings creates significant hiccups and drops the average FPS quite frequently.

As for your CPU question no if you noticed I set gallium hud so that it shows not only each separate CPU core but also the total CPU usage in a percentage and it was all the time bellow 60.

and even at 100% my CPU isnt bottlenecking the card for example in tomb raider benchmark I get one core at 100% but also my GPU usage is 100% almost all the time

As for Michaels article I really dont get any similar results... not even close (e.g ultra settings HD dota ~70 FPS avg CSGO highest settings HD resolution ~70 FPS avg) only the superposition bench seems to agree (I get 15.8 he gets 16.02)

And also for some reason the GTX 1060 results seem to have regressed compared to the older benchmark he had done last time with the same card on 17.3 I think which I also find strange.

[papajo]

Latest OBS studio does not cause significant hiccups and drops to the average fps with the Ryzen 5 1600 cpu so you have a bit weaker cpu. In TR 2013 fullhd normal settings benchmark with RX560 OBS studio drops average fps from 116 to 105 so that does not affect to gaming. Also your Linux system is not optimised.

Obviously it does not if you dont know about something then dont talk about it. Ryzen 5 has 12 threads and about 46% better IPC than FX who has 8 threads. So obviously it will do better in combinatory tasks such as gaming and recording screenplay they even beat the recent intel CPUs in that area that doesnt mean anything though OBS when using software encoder is still a heavy task its only that the CPU you mentioned can facilitate it.

EDIT as for my linux system it has the latest low latency kernel the latest drivers and is on the latest version of my distro there is no optimization to make other than engaging all the cores to performance with the governor which I also did.

[puleglot]

a) That CPU bottleneck exists when the CPU can not manage to send draw calls to the GPU in the needed pace or in other words when CPU is at 100% load which in my case neither of my cores reaches that load.

Witcher 2 doesn't set CPU affinity for its thread(s). And the number of (busy) threads is lower than number of your CPU cores.

An other way to prove the bottleneck is caused by the CPU is that when lowering the settings the FPS rate stays the same which doesnt happen either.

Lowering the settings can decrease the number of draw calls and improve FPS event in CPU-limited games.