RADV vs. NVIDIA Vulkan/OpenGL Performance For Serious Sam 2017

[geearf]

Also remember that Fury (as opposed to Fury X) is <$300 using those prices...

Going back to the earlier topic, it's never been clear to me why anyone would expect graphics performance to be determined by shader core throughput but not by any of the other subsystems that contribute to graphics rendering.

Because it's easier to be wrong that way

[smitty3268]

Also remember that Fury (as opposed to Fury X) is <$300 using those prices...

Going back to the earlier topic, it's never been clear to me why anyone would expect graphics performance to be determined by shader core throughput but not by any of the other subsystems that contribute to graphics rendering.

Because they have no idea how the hardware actually works and it's an easy number to latch onto.

It's like people who used to focus on the GHz as the sole number that mattered for CPU performance, while the rest of us just rolled our eyes at them.

[dungeon]

Going back to the earlier topic, it's never been clear to me why anyone would expect graphics performance to be determined by shader core throughput but not by any of the other subsystems that contribute to graphics rendering.

GCC's -O3 is known to improve and break performance, but some people even think that is the rule of thumb and only better perf come with that But no optimization is safe, -O2 consist of collection of optimizations which are on average considered safe but even that might break something

Just reading bugs on fdo bugzilla, man enabled -Ofast for mesa probaly because word fast looks like thing would be fast, resonable people would name that -Obroken to clearfy things

https://bugs.freedesktop.org/show_bug.cgi?id=98856

Very fast but more and more quite possibly broken... -O0 the King, King is dead, viva la King

[indepe]

GCC's -O3 is known to improve and break performance, but some people even think that is the rule of thumb and only better perf come with that But no optimization is safe, -O2 consist of collection of optimizations which are on average considered safe but even that might break something

Just reading bugs on fdo bugzilla, man enabled -Ofast for mesa probaly because word fast looks like thing would be fast, resonable people would name that -Obroken to clearfy things

https://bugs.freedesktop.org/show_bug.cgi?id=98856

Very fast but more and more quite possibly broken... -O0 the King, King is dead, viva la King

It looks like this is a 10 year old myth. Nowadays the downside of -O3 is mostly that it takes longer to compile, and it may produce larger binaries, depending on the app.

[dungeon]

It looks like this is a 10 year old myth. Nowadays the downside of -O3 is mostly that it takes longer to compile, and it may produce larger binaries, depending on the app.

I agree, nowdays is the same as 10 years ago -O3 isn't recommended on average as safe that is always -O2 ... more optimization might break things and even instead of being faster to be slower.

[duby229]

I agree, nowdays is the same as 10 years ago -O3 isn't recommended on average as safe that is always -O2 ... more optimization might break things and even instead of being faster to be slower.

I use a gentoo system here, so it's pretty easy to play around with cflags. But in my experience it really isn't a good idea at all. If you build the whole system with -O3 it results in random strange behavior that is hard to describe. Rebuild with -O2 and random strangeness goes away.

[dungeon]

Blah always use -O2 is average safe mantra, anything added to that or going to -O3 or up is considered sort of overclocking which might work very good in some special scenario, but to use it generally... no way, that good might became very bad instead

To me it is exactly like overclocking hardware, where there is no really safe overclocking. While some people think there is safe OC instead Same way there is no safe generally recommended flags beyond -O2 also Of course no one stops you and wish you luck pushing things further, but that is a push into unpredictable area and user should aknowladge and take all possible consenvences Same like with OC, you lose warranty if you do that. Kernel might panic, apps suddenly closed up, random missing things, elements, slower instead of faster, etc... that no one can predict nor debug, same thing mght happen with too much compiler optimizations

It is just that consumers or Desktop users, might not care about possible breakages like that into areas they might no care so much so firing things up here and there is considered fine

[indepe]

I agree, nowdays is the same as 10 years ago -O3 isn't recommended on average as safe that is always -O2 ... more optimization might break things and even instead of being faster to be slower.

Recommended where?

"More optimization might breaks things"... that doesn't sound very meaningful. Any software "might break".

[-O3 is not like -Ofast.]

[indepe]

I use a gentoo system here, so it's pretty easy to play around with cflags. But in my experience it really isn't a good idea at all. If you build the whole system with -O3 it results in random strange behavior that is hard to describe. Rebuild with -O2 and random strangeness goes away.

"The whole system" doesn't build with clang either, however that doesn't mean you can't use clang.

Besides I wonder what all that has to do with the original topic.

[duby229]

"The whole system" doesn't build with clang either, however that doesn't mean you can't use clang.

Besides I wonder what all that has to do with the original topic.

I was following along with the conversation. I can quote the relevant posts for you or you can read the thread I suppose. I provided one example where in my experience -O3 isn't a good idea.