|
1 | 1 | --- |
2 | 2 | title: CachyOS Repository General Information |
3 | | -description: Benchmarks and information about the repository |
| 3 | +description: Benchmarks and information about our optimized repositories |
4 | 4 | --- |
5 | 5 |
|
6 | | -# Why does CachyOS utilize its own repositories? |
| 6 | +Our aim to deliver a performance-optimized distribution necessitates recompiling essential Arch Linux packages for generic x86-64, |
| 7 | +x86-64-v3, x86-64-v4, and Zen4 architectures. |
7 | 8 |
|
8 | | -## Performance and optimizations |
9 | | -We want to provide a performance-optimized distribution and this requires performance-optimized packages. Currently we are recompiling the Arch Linux repository core and extra with the generic x86-64-v3 and x86-64-v4 instruction set. |
| 9 | +- x86-64 (Most CPUs) - While lacking AVX2 support found in x86-64-v3, this repository offers optimized packages. LTO and PGO techniques have boosted performance in several packages. |
| 10 | +- x86-64-v3- (Intel Haswell and up, AMD Ryzen 1st gen and up) - 5%-20% performance uplift compared to x86-64. |
| 11 | +- x86-64-v4 (Intel Rocket Lake, Zen 4, Zen 5) - Delivers substantial performance gains through AVX512 support, depending on the workload. |
| 12 | +- Zen 4/5 - In addition to the x86-64-v4 instruction set, the following are added: |
10 | 13 |
|
11 | | -- x86-64-v4 - Enabling AVX512 can significantly improve application performance (Depending on the application/workload) |
12 | | -- x86-64-v3 - *5%-20% performance uplift* |
| 14 | +```text |
| 15 | +abm, adx, aes, avx512bf16, avx512bitalg, avx512ifma, avx512vbmi, avx512vbmi2, avx512vnni, |
| 16 | +avx512vpopctndq, clflushopt, clwb, clzero, fsgsbase, gfni, mwaitx, pclmul, pku, prfchw, |
| 17 | +rpdid, rdrnd, rdseed, sha, sse4a, vaes, vockmulqdq, wbnoinvd, savec, xsaveopt, xsaves |
| 18 | +``` |
13 | 19 |
|
14 | | -## Customized Packages |
| 20 | +Check out [this](https://en.wikipedia.org/wiki/X86-64#Microarchitecture_levels) page for more information about the instructions used in the different x86_64 micro-architectures |
15 | 21 |
|
16 | | -In the [CachyOS-PKGBUILDs](https://github.com/CachyOS/CachyOS-PKGBUILDS) repository we are maintaining a bunch of packages, which are constantly getting patches. PGO or BOLT Optimization to provide an additional uplift in performance. |
| 22 | +:::note |
| 23 | +Wikipedia page quotes that the Skylake architecture supports AVX512 which is wrong, only Skylake-X does. |
| 24 | +::: |
17 | 25 |
|
18 | | -# Tests and benchmarks |
19 | | -If you would like to know more about the performance uplift from our repositories. Please see the links below. |
| 26 | +### Customized Packages |
20 | 27 |
|
21 | | -- Phoronix has already benchmarked Cachy twice and it seems to be a win for CachyOS, which is shown mostly leading in the benchmark graphs and on the Geometric Mean of All Test Results. |
22 | | -These benchmarks are old by now, After these two reviews, CachyOS evolved and matured a lot more in terms of usability and performance. the following review was the first one Michael made of CachyOS on it's infancy. |
| 28 | +Our [CachyOS-PKGBUILDs](https://github.com/CachyOS/CachyOS-PKGBUILDS) repository contains packages that receive ongoing updates, patches and backported fixes. |
| 29 | +To boost performance, we selectively implement PGO, LTO, and BOLT optimizations depending on the need. |
| 30 | +We also maintain a couple of -git packages e.g mesa-git. |
| 31 | + |
| 32 | +### Tests and benchmarks |
| 33 | + |
| 34 | +If you would like to know more about the performance uplift from our repositories. Please check the links below. |
| 35 | + |
| 36 | +- Phoronix has already benchmarked CachyOS twice and it seems to be a win for CachyOS, which is shown mostly leading in the benchmark graphs and on the Geometric Mean of All Test Results. |
| 37 | +These benchmarks are old by now, After these two reviews, CachyOS has evolved and matured a lot more in terms of usability and performance. the following review was the first one Michael made of CachyOS on it's early days. |
23 | 38 | https://www.phoronix.com/review/cachyos-linux-perf |
24 | 39 |
|
25 | 40 | - Here, Phoronix has recently conducted another benchmark demonstrating the difference between x86-64-v4, x86-64-v3 and x86-64 (generic) Packages. Looking at the examples like PHP or GCC, where we customize our PKGBUILDs there is a noticeable performance improvement. |
26 | 41 | https://www.phoronix.com/review/cachyos-x86-64-v3-v4 |
27 | 42 |
|
28 | | - |
29 | | -- x86-64-v3 (- there is a moderate benefit of *-march=haswell* (x86_64-v3) - of around |
30 | | -10%-20% as compared to baseline for the tests performed by Arch Linux): |
31 | | -https://lists.archlinux.org/pipermail/arch-general/2021-March/048739.html |
| 43 | +- x86-64-v3 (there is a moderate benefit of *-march=haswell* (x86_64-v3) - of around |
| 44 | +10%-20% as compared to baseline for the tests performed by Arch Linux): |
| 45 | +https://lists.archlinux.org/pipermail/arch-general/2021-March/048739.html |
0 commit comments