[mobilio]

Announce was here: https://discourse.ubuntu.com/t/introducing-architecture-vari...

and key point: "Previous benchmarks we have run (where we rebuilt the entire archive for x86-64-v3 57) show that most packages show a slight (around 1%) performance improvement and some packages, mostly those that are somewhat numerical in nature, improve more than that."

[ninkendo]

> show that most packages show a slight (around 1%) performance improvement

This takes me back to arguing with Gentoo users 20 years ago who insisted that compiling everything from source for their machine made everything faster.

The consensus at the time was basically "theoretically, it's possible, but in practice, gcc isn't really doing much with the extra instructions anyway".

Then there's stuff like glibc which has custom assembly versions of things like memcpy/etc, and selects from them at startup. I'm not really sure if that was common 20 years ago but it is now.

It's cool that after 20 years we can finally start using the newer instructions in binary packages, but it definitely seems to not matter all that much, still.

[Amadiro]

It's also because around 20 years ago there was a "reset" when we switched from x86 to x86_64. When AMD introduced x86_64, it made a bunch of the previously optional extension (SSE up to a certain version etc) a mandatory part of x86_64. Gentoo systems could already be optimized before on x86 using those instructions, but now (2004ish) every system using x86_64 was automatically always taking full advantage of all of these instructions*.

Since then we've slowly started accumulating optional extensions again; newer SSE versions, AVX, encryption and virtualization extensions, probably some more newfangled AI stuff I'm not on top of. So very slowly it might have started again to make sense for an approach like Gentoo to exist**.

* usual caveats apply; if the compiler can figure out that using the instruction is useful etc.

** but the same caveats as back then apply. A lot of software can't really take advantage of these new instructions, because newer instructions have been getting increasingly more use-case-specific; and applications that can greatly benefit from them will already have alternative code-pathes to take advantage of them anyway. Also a lot of the stuff happening in hardware acceleration has moved to GPUs, which have a feature discovery process independent of CPU instruction set anyway.

[slavik81]

The llama.cpp package on Debian and Ubuntu is also rather clever in that it's built for x86-64-v1, x86-64-v2, x86-64-v3, and x86-64-v4. It benefits quite dramatically from using the newest instructions, but the library doesn't have dynamic instruction selection itself. Instead, ld.so decides which version of libggml.so to load depending on your hardware capabilities.

[ignoramous]

> llama.cpp package on Debian and Ubuntu is also rather clever … ld.so decides which version of libggml.so to load depending on your hardware capabilities

Why is this "clever"? This is pretty much how "fat" binaries are supposed to work, no? At least, such packaging is the norm for Android.

[mikepurvis]

> AVX, encryption and virtualization

I would guess that these are domain-specific enough that they can also mostly be enabled by the relevant libraries employing function multiversioning.

[izacus]

You would guess wrong.

[mikepurvis]

Isn’t the whole thrust of this thread that most normal algorithms see little to no speed up from things like avx, and therefore multiversioning those things that do makes more sense that compiling the whole OS for a newer set of cpu features?

[ploxiln]

FWIW the cool thing about gentoo was the "use-flags", to enable/disable compile-time features in various packages. Build some apps with GTK or with just the command-line version, with libao or pulse-audio, etc. Nowadays some distro packages have "optional dependencies" and variants like foobar-cli and foobar-gui, but not nearly as comprehensive as Gentoo of course. Learning about some minor custom CFLAGS was just part of the fun (and yeah some "funroll-loops" site was making fun of "gentoo ricers" way back then already).

I used Gentoo a lot, jeez, between 20 and 15 years ago, and the install guide guiding me through partitioning disks, formatting disks, unpacking tarballs, editing config files, and running grub-install etc, was so incredibly valuable to me that I have trouble expressing it.

[mpyne]

I still use Gentoo for that reason, and I wish some of those principles around handling of optional dependencies were more popular in other Linux distros and package ecosystems.

There's lots of software applications out there whose official Docker images or pip wheels or whatever bundle everything under the sun to account for all the optional integrations the application has, and it's difficult to figure out which packages can be easily removed if we're not using the feature and which ones are load-bearing.

[zerocrates]

I started with Debian on CDs, but used Gentoo for years after that. Eventually I admitted that just Ubuntu suited my needs and used up less time keeping it up to date. I do sometimes still pull in a package that brings a million dependencies for stuff I don't want and miss USE flags, though.

I'd agree that the manual Gentoo install process, and those tinkering years in general, gave me experience and familiarity that's come in handy plenty of times when dealing with other distros, troubleshooting, working on servers, and so on.

[michaelcampbell]

Someone has set up an archive of that site; I visit it once in a while for a few nostalgiac chuckles

https://www.shlomifish.org/humour/by-others/funroll-loops/Ge...

[viraptor]

Nixpkgs exposes a lot of options like that. You can override both options and dependencies and supply your own cflags if you really want.

[oivey]

This should build a lot more incentive for compiler devs to try and use the newer instructions. When everyone uses binaries compiled without support for optional instruction sets, why bother putting much effort into developing for them? It’ll be interesting to see if we start to see more of a delta moving forward.

[Seattle3503]

And application developers to optimize with them in mind?

[hajile]

According to this[0] study of the Ubuntu 16.04 package repos, 89% of all x86 code was instructions were just 12 instructions (mov, add, call, lea, je, test, jmp, nop, cmp, jne, xor, and -- in that order).

The extra issue here is that SIMD (the main optimization) simply sucks to use. Auto-vectorization has been mostly a pipe dream for decades now as the sufficiently-smart compiler simply hasn't materialized yet (and maybe for the same reason the EPIC/Itanium compiler failed -- deterministically deciding execution order at compile time isn't possible in the abstract and getting heuristics that aren't deceived by even tiny changes to the code is massively hard).

Doing SIMD means delving into x86 assembly and all it's nastiness/weirdness/complexity. It's no wonder that devs won't touch it unless absolutely necessary (which is why the speedups are coming from a small handful of super-optimized math libraries). ARM vector code is also rather Byzantine for a normal dev to learn and use.

We need a more simple assembly option that normal programmers can easily learn and use. Maybe it's way less efficient than the current options, but some slightly slower SIMD is still going to generally beat no SIMD at all.

[0] https://oscarlab.github.io/papers/instrpop-systor19.pdf

[badlibrarian]

Agner Fog's libraries make it pretty trivial for C++ programmers at least. https://www.agner.org/optimize/

[kccqzy]

The highway library is exactly the kind of a simpler option to use SIMD. Less efficient than hand written assembler but you can easily write good enough SIMD for multiple different architectures.

[JonChesterfield]

The sufficiently smart vectoriser has been here for decades. Cuda is one. Uses all the vector units just fine, may struggle to use the scalar units.

[suprjami]

I somehow have the memory that there was an extremely narrow time window where the speedup was tangible and quantifiable for Gentoo, as they were the first distro to ship some very early gcc optimisation. However it's open source software so every other distro soon caught up and became just as fast as Gentoo.

[harha]

Would it make a difference if you compile the whole system vs. just the programs you want optimized?

As in, are there any common libraries or parts of the system that typically slow things down, or was this more targeting a time when hardware was more limited so improving all would have made things feel faster in general.

[juujian]

Are there any use cases where that 1% is worth any hassle whatsoever?

[wongarsu]

If every computer built in the last decade gets 1% faster and all we have to pay for that is a bit of one-off engineering effort and a doubling of the storage requirement of the ubuntu mirrors that seems like a huge win

If you aren't convinced by your ubuntu being 1% faster, consider how many servers, VMs and containers run ubuntu. Millions of servers using a fraction of a percent less energy multiplies out to a lot of energy

[vladms]

Don't have a clear opinion, but you have to factor in all the issues that can be due to different versions of software. Think of unexposed bugs in the whole stack (that can include compiler bugs but also software bugs related to numerical computation or just uninitialized memory). There are enough heisenbugs without worrying that half the servers run on a slightly different software.

It's not for nothing that some time ago "write once, run everywhere" was a selling proposition (not that it was actually working in all cases, but definitely working better than alternatives).

[sumtechguy]

That comes out to about 1.5 hours faster per week for many tasks. If you are running full tilt. But that seems like an ok easy win.

[alkonaut]

If I recompile a program to fully utilize my cpu better (use AVX or whatever) then if my program takes 1 second to execute instead of 2, it likely did not use half the _energy_.

[darkwater]

Obviously not. But scale it out to a fleet of 1000 servers running your program continuously, you can now shut down 10 for the same exact workload.

[zymhan]

Sure, but we're talking about compiled packages being distributed by a package manager.

[alkonaut]

Yes but my point is: if I download the AVX version instead of the SSE version of a package and that makes my 1000 servers 10% _quicker_ that is not the same as being 10% more _efficient_.

Because typically these modern things are a way of making the CPU do things faster by eating more power. There may be savings from having fewer servers etc, but savings in _speed_ are not the same as savings in _power_ (and some times even work the opposite way)

[duskdozer]

how much energy would we save if every website request weren't loaded down with 20MB of ads and analytics :(

[Aissen]

You need 100 servers. Now you need to only buy 99. Multiply that by a million, and the economies of scale really matter.

[iso1631]

1% is less than the difference between negotiating with a hangover or not.

[gpm]

What a strange comparison.

If you're negotiating deals worth billions of dollars, or even just millions, I'd strongly suggest not doing so with a hangover.

[Pet_Ant]

> If you're negotiating deals worth billions of dollars, or even just millions, I'd strongly suggest not doing so with a hangover.

...have you met salespeople? Buying lap dances is a legitimate business expense for them. You'd be surprised how much personal rapport matters and facts don't.

In all fairness, I only know about 8 and 9 figure deals, maybe at 10 and 11 salespeople grow ethics...

[bregma]

I strongly suspect ethics are inversely proportional to the size of the deal.

[glenstein]

That's more an indictment of sales culture than a critique of computational efficiency.

[squeaky-clean]

Well sure, because you want the person trying buy something from you for a million dollars to have a hangover.

[tclancy]

Sounds like someone never read Sun Tzu.

(Not really, I just know somewhere out there is a LinkedInLunatic who has a Business Philosophy based on being hungover.)

[gpm]

Appear drunk when you are sober, and sober when you are drunk

- Sun Zoo

[wat10000]

It's rarely going to be worth it for an individual user, but it's very useful if you can get it to a lot of users at once. See https://www.folklore.org/Saving_Lives.html

"Well, let's say you can shave 10 seconds off of the boot time. Multiply that by five million users and thats 50 million seconds, every single day. Over a year, that's probably dozens of lifetimes. So if you make it boot ten seconds faster, you've saved a dozen lives. That's really worth it, don't you think?"

I put a lot of effort into chasing wins of that magnitude. Over a huge userbase, something like that has a big positive ROI. These days it also affects important things like heat and battery life.

The other part of this is that the wins add up. Maybe I manage to find 1% every couple of years. Some of my coworkers do too. Now you're starting to make a major difference.

[PeterStuer]

A lott of improvements are very incremental. In agregate, they often compound and are vey significant.

If you would only accept 10x improvements, I would argue progress would be very small.

[ilaksh]

They did say some packages were more. I bet some are 5%, maybe 10 or 15. Maybe more.

Well one example could be llama.cpp . It's critical for them to use every single extension the CPU has move more bits at a time. When I installed it I had to compile it.

This might make it more practical to start offering OS packages for things like llama.cpp

I guess people that don't have newer hardware aren't trying to install those packages. But maybe the idea is that packages should not break on certain hardware.

Blender might be another one like that which really needs the extensions for many things. But maybe you so want to allow it to be used on some oldish hardware anyway because it still has uses that are valid on those machines.

[locknitpicker]

> Are there any use cases where that 1% is worth any hassle whatsoever?

I don't think this is a valid argument to make. If you were doing the optimization work then you could argue tradeoffs. You are not, Canonical is.

Your decision is which image you want to use, and Canonical is giving you a choice. Do you care about which architecture variant you use? If you do, you can now pick the one that works best for you. Do you want to win an easy 1% performance gain? Now you have that choice.

[adgjlsfhk1]

it's very no uniform. 99% see no change, but 1% see 1.5-2x better performance

[2b3a51]

I'm wondering if 'somewhat numerical in nature' relates to lpack/blas and similar libraries that are actually dependencies of a wide range of desktop applications?

[adgjlsfhk1]

blas and lapack generally do manual multi-versioning by detecting CPU features at runtime. This is more useful 1 level up the stack in things like compression/decompression, ode solvers, image manipulation and so on that are still working with big arrays of data, but don't have a small number of kernels (or as much dev time), so they typically rely on compilers for auto-vectorization

[Insanity]

I read it as, across the board a 1% performance improvement. Not that only 1% of packages get a significant improvement.

[IAmBroom]

In a complicated system, a 1% overall benefit might well be because of a 10% improvement in just 10% of the system (or more in a smaller contributor).

[darkwater]

The announcement is pretty clear on this:

   > Previous benchmarks (...) show that most packages show a slight (around 1%) performance improvement and some packages, mostly those that are somewhat numerical in nature, improve more than that.

[godelski]

  > where that 1% is worth any hassle

You'll need context to answer your question, but yes there are cases.

Let's say you have a process that takes 100hrs to run and costs $1k/hr. You save an hour and $1k every time you run the process. You're going to save quite a bit. You don't just save the time to run the process, you save literal time and everything that that costs (customers, engineering time, support time, etc).

Let's say you have a process that takes 100ns and similarly costs $1k/hr. You now run in 99ns. Running the process 36 million times is going to be insignificant. In this setting even a 50% optimization probably isn't worthwhile (unless you're a high frequency trader or something)

This is where the saying "premature optimization is the root of all evil" comes from! The "premature" part is often disregarded and the rest of the context goes with it. Here's more context to Knuth's quote[0].

  There is no doubt that the holy grail of efficiency leads to abuse. Programmers waste enormous amounts of time thinking about, or worrying about, the speed of noncritical parts of their programs, and these attempts at efficiency actually have a strong negative impact when debugging and maintenance are considered. We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil.

  Yet we should not pass up our opportunities in that critical 3%. A good programmer will not be lulled into complacency by such reasoning, he will be wise to look carefully at the critical code; but only after that code has been identified.

Knuth said: "Get a fucking profiler and make sure that you're optimizing the right thing". He did NOT say "don't optimize".

So yes, there are plenty of times where that optimization will be worthwhile. The percentages don't mean anything without the context. Your job as a programmer is to determine that context. And not just in the scope of your program, but in the scope of the environment you expect a user to be running on. (i.e. their computer probably isn't entirely dedicated to your program)

[0] https://dl.acm.org/doi/10.1145/356635.356640 (alt) https://sci-hub.se/10.1145/356635.356640

[dehrmann]

Anything at scale. 1% across FAANG is huge.

[dabinat]

If those kinds of optimizations are on the table, why would they not already be compiling and optimizing from source?

[darkwater]

I'm not an hyperscaler, I run a thousand machines. If by just changing the base image I use to build - in an already automated process - those machines, well, the optimization is basically for free. Well, unless it triggers some new bug that was not there before.

[Havoc]

Arguable same across consumers too. It’s just harder to measure than central datacenters

[notatoad]

nah, performance benefits are mostly wasted on consumers, because consumer hardware is very infrequently CPU-constrained. in a datacentre, a 1% improvement could actually mean you provision 99 CPUs instead of 100. but on your home computer, a 1% CPU improvement means that your network request completes 0.0001% faster, or your file access happens 0.000001% faster, and then your CPU goes back to being idle.

an unobservable benefit is not a benefit.

[bandrami]

Isn't Facebook still using PHP?

[dehrmann]

They forked PHP into Hack. They've diverged pretty far by this point (especially with data structures), but it maintains some of PHPs quirks and request-oriented runtime. It's jitted by HHVM. Both Hack and HHVM are open-source, but I'm not aware of any major users outside Meta.

[speed_spread]

Compiled PHP. I'm pretty sure they ran the numbers.

[bandrami]

Hell, I still get Perl CGI contracts so I'm in no place to talk

[gwbas1c]

> some packages, mostly those that are somewhat numerical in nature, improve more than that

Perhaps if you're doing CPU-bound math you might see an improvement?

[rossjudson]

Any hyperscaler will take that 1% in a heartbeat.

[colechristensen]

Very few people are in the situation where this would matter.

Standard advice: You are not Google.

I'm surprised and disappointed 1% is the best they could come up with, with numbers that small I would expect experimental noise to be much larger than the improvement. If you tell me you've managed a 1% improvement you have to do a lot to convince me you haven't actually made things 5% worse.

[noir_lord]

No but a lot of people are buying a lot of compute from Google, Amazon and Microsoft.

At scale marginal differences do matter and compound.

[dang]

Thanks - we've merged the comments from https://news.ycombinator.com/item?id=45772579 into this thread, which had that original source.

[horizion2025]

How many additions have there even been outside of AVX-x? And even AVX-2 is from 2011. If we ignore AVX-x the last I can recall are the few instructions added in the manipulation sets BMI/ABM, but they are Haswell/Piledriver/Jaguar era (2012-2013). While some specific cases could benefit, doesn't seem like a goldmine of performance improvements.

Further, maybe it has not been a focus for compiler vendors to generate good code for these higher-level archs if few are using the feature. So Ubuntu's move could improve that.

[jwrallie]

Is it worth it losing the ability to just put your hdd on your older laptop and booting it in an emergency?

[pizlonator]

That 1% number is interesting but risks missing the point.

I bet you there is some use case of some app or library where this is like a 2x improvement.

[alternatex]

Aggregated metrics are always useless as they tend to show interesting and sometimes exciting data that in actuality contains zero insight. I'm always weary of people making decisions based on aggregate metrics.

Would be nice to know the per app metrics.