Originally posted by bridgman
View Post
Announcement
Collapse
No announcement yet.
Core i9 7900X vs. Threadripper 1950X On Ubuntu 17.10, Antergos, Clear Linux
Collapse
X
-
Originally posted by boxie View PostAlso pretty sure that AMD is punching above its weight with regards to marketshare/revenue to OSS contribution ratio
- Likes 6
Comment
-
tiwake
It seems that Michael is not willing to do benchmarks on Gentoo.
About -march=native, that is not the whole story when it comes to performance. Gentoo's USE flags allow you to leave out all the cruft you don't need and which would otherwise clog your system.
franglais125
Likely the threshold won't change until there is considerable perfomance gain. In the meantime, the distro's focus appears to be on GCC function multi-versioning to achieve some of the benefits of more recent CPU instructions without giving up compatibility with older CPUs.
Comment
-
Originally posted by chuckula View PostIt would be nice if AMD at any point in the history of the company had made a contribution to open source that extended past getting its own hardware to run.
- Likes 16
Comment
-
chithanh In that case he doesn't really need to waste time on Gentoo, he can compile his custom kernel and other system critical things with those flags on any distribution, it would still take more time, but far less than doing Gentoo. That comes from the person who never even tried Gentoo, so my understanding of that OS is that everything needs to be compiled, correct me of I'm wrong.
Comment
-
Originally posted by arjan_intel View Post
Clear Linux runs on Atom as well, baseline is "Westmere" (2010 CPU, pre-AVX)
Maybe it's the whole OS, but I'm wondering if one could not recompile certain key packages to get about the same performance gain without having to go all source like gentoo.
- Likes 2
Comment
-
Originally posted by leipero View Posthe can compile his custom kernel and other system critical things with those flags on any distribution, it would still take more time, but far less than doing Gentoo.
While performance can be among the reasons why people run Gentoo, it is typically not the main reason. It is rather having more choice than a binary distro can give you, and having compiler optimized packages for your CPU is a free benefit from that.
Originally posted by geearf View PostDo you guys have an idea of which packages matter the most in terms of compilation optimizations?
Maybe it's the whole OS, but I'm wondering if one could not recompile certain key packages to get about the same performance gain without having to go all source like gentoo.
To answer the question which packages matter most, you need to analyze and profile the things which are performance critical to you. For example, if you do lots of media encoding, then ffmpeg would be a good candidate. Or imagemagick/graphicsmagick if you process a lot of pictures. Here is a report from a G'MIC user who saw huge performance gain from recompiling this particular application with optimized compiler flags.
Or if your code spends lots of time in libc (calling math functions) or in kernel space then look at optimizing those.
- Likes 1
Comment
-
Originally posted by chithanh View PostThis is probably not what you want to hear, but: It squarely depends on your use case.
Originally posted by chithanh View PostTo answer the question which packages matter most, you need to analyze and profile the things which are performance critical to you. For example, if you do lots of media encoding, then ffmpeg would be a good candidate. Or imagemagick/graphicsmagick if you process a lot of pictures. Here is a report from a G'MIC user who saw huge performance gain from recompiling this particular application with optimized compiler flags.
Or if your code spends lots of time in libc (calling math functions) or in kernel space then look at optimizing those.
I tried once with o3 but got many issues with the reencoded stuff and never tried again.
But yeah globally you make a good point.
I was more thinking of whatever app you use + whatever core stuff needs it. Like I don't think the kernel would change a lot based on grayski's benchmarks, but maybe glibc as you mentioned or something else...
Comment
Comment