OpenSUSE Developers Continue Discussing x86_64 Microarchitecture Feature Levels

[Jedibeeftrix]

personally i'd vote for v3, but that is because by the time ALP arrives my entire PC estate will be Zen3 or newer.

[Developer12]

So much slicing and dicing, trying to find the least evil, when better tooling would be a simple and straightforward fix. We could sit here all day arguing over where to draw the line in the sand, or just patch the package manager to read the CPU info and check some package tags.

Then all the packages are compiled for v1 if you like, and anything performance sensitive ranging from the kernel to browsers to databases can be additionally compiled with v4 (or whatever combination) and selected where supported. Minimum compilation, no lost users, potential support for future package customizations.

[pWe00Iri3e7Z9lHOX2Qx]

Jasper & Elkhart Lake still don't support AVX. Therefore, requiring v3 shuts out a lot of chromebook, embedded, and even NUC/mini-desktop users.

Maybe not a deal-breaker for SuSE Enterprise, but I hope OpenSUSE definitely doesn't start requiring v3!

The feature level debates are just one of a million reasons that distros need some instrumentation to make better product decisions. What percentage of current Leap users are on systems that don't support AVX? Is it 2% and you could just tell them to go use i686 Tumbleweed if they want to stay in the family? Is it 20%? Hell, what's the total population of Leap users? I get the whole "but muh freedoms!!!" thing and why there will never be useful telemetry, but it's really hard to build and refine a great product when you have no damn clue how it is consumed, tweaked, etc.

[coder]

I think there needs to be a v2.5 -- v2 with avx or v3 without avx2. The compromise level.

Requiring AVX, but not AVX2, would only get you Sandybridge + Ivy Bridge. However, it would still be a deal-breaker for the bottom-tier Atom-derived CPUs.

I was originally annoyed to see the cut for v3 make at AVX2, but I've come to accept it as a sensible point to draw the line. Especially if you plot them in time:

v1 - 2005
v2 - 2009
v3 - 2013

There's a nice 4-year spacing between the first 3. Even v4 uses features introduced in 2017, but only to Intel's server & HEDT platforms. Rocket Lake brought them to the mainstream desktop, but then Alder Lake snatched them away.

[coder]

I definitely agree with SSE 4.2, I think it's a bit early perhaps for AVX requirements and maybe make the cut off for AVX for the release after.

Thoughts please folks ?

Jasper & Elkhart Lake still don't support AVX. Therefore, requiring v3 shuts out a lot of chromebook, embedded, and even NUC/mini-desktop users.

Maybe not a deal-breaker for SuSE Enterprise, but I hope OpenSUSE definitely doesn't start requiring v3!

[SilverBird775]

Raising the x86_64 micro-architecture feature level is generally a way to go, yet with aware. SSE3 age CPU is the most safe common ground today. Such a decisions should be made strongly on global hardware surveys data.

Here is an example:

https://store.steampowered.com/hwsurvey/

SSE2 - 100.00%
SSE3 - 100.00%
LAHF / SAHF - 100.00%
CMPXCHG16B - 99.99%

99.99% to 100% coverage is high enough to make use the feature. IMHO.

[pWe00Iri3e7Z9lHOX2Qx]

It seems kind of strange not to use the same feature level as SUSE when they just went through all that effort to use the same binary packages as SUSE in 15.3.

¯\_(ツ)_/¯

[skeevy420]

I think the main obstacle for distro maintainers to default to x86_64_v3 is that Pentium and Celeron CPUs don't support AVX2 until recently.

That's basically it. Every time this discussion comes up it always boils down to "so what about those Intels?" and we begrudgingly all come to the consensus that v2 is the best in regards to an optimized compromise.

[skeevy420]

I think there needs to be a v2.5 -- v2 with avx or v3 without avx2. The compromise level.

[Setif]

I think the main obstacle for distro maintainers to default to x86_64_v3 is that Pentium and Celeron CPUs don't support AVX2 until recently.