Hacker News Reader

Comment by pavlov | Hacker News Reader

During the late PowerPC era (2000-2005), Apple was at risk of losing this industry entirely. Every new professional content creation application was being built primarily for the Windows NT platform which had already successfully killed off SGI workstations and their proprietary Unix software stack. Intel and AMD CPUs were consistently delivering both better performance and pricing, despite the brief glimmer of hope from IBM's PowerPC G5 CPU in 2003.

What kept Apple alive in this market was 1) Pro Tools, 2) their own suite of applications like Final Cut Pro and Logic, 3) the 2006 Mac Pro that finally delivered everything users had hoped for — latest Intel CPUs, Windows compatibility, enough fast PCI slots for everyone. It was really the best of both worlds and became the x86 media desktop to beat. (In typical Apple fashion, the 2013 Mac Pro swung too far the other way towards Apple-specific integration and was an abysmal flop with massive heat problems and non-upgradeable GPUs stuck in the past.)

View on HN

Ironically Pro Tools is still not M1 native.

One cool thing in MacOS that Windows can't do is aggregates of audio devices/interfaces.

Honestly this area is so frustrating technologically.

MacOS has easily the best Audio subsystem across MacOS & Windows.

In Linux we have JACK, which is as good as MacOS (if not better honestly).

But then Linux suffers from having very little commercial support and worse than that, a lot of problems with hardware video encoding.

Windows doesn't have any problems encoding video, but it has a terrible filesystem (which is forced on you) that causes problems for the insanely large files that you must work with on video production and in addition: the audio subsystem is the worst of all three platforms.

Linux could be a contender, if someone threw $100M USD at the problem, but getting adoption for a linux solution would be hard because the Apple stuff works "fine" and movie productions are more time constrained than cost constrained.

> In Linux we have JACK, which is as good as MacOS (if not better honestly).

And now we have Pipewire which is not only better, but can also pretend it's Jack for compatibility.

Pipewire still had its issues (mixing setups can easy cause weird sound issues) but it's much better than anything that came before it.

I'd go as far as to say Pipewire made connecting to my Bluetooth headphones more reliable on Linux than on Windows, though neither audio stack is particularly good.

> In Linux we have JACK, which is as good as MacOS (if not better honestly).

screams in terror

The ideas are great, the realization is jackshit. Sooo much finicky crap that partly also depends of whether your sound output is special enough to work well with it. And trying to make it play nice with pulseaudio is another nightmare.

Pulseaudio itself also suffers massively from UI problems, pretty much all non-basic use cases is either mess with configs or "just run a bunch of commands on runtime or every time your USB interface reconnects". Jack at least gets that part right

Trying to get some audio interface to be connected properly was... an experience, it just showed in system as one stereo input (it was 2 mono inputs) and 4.0 output (it had 4 separate outs) so there was a good deal of pulseaudio fuckery to run to just split it properly

> Linux could be a contender, if someone threw $100M USD at the problem, but getting adoption for a linux solution would be hard because the Apple stuff works "fine" and movie productions are more time constrained than cost constrained.

I wonder if someone figuring out how to run the usual tools via Wine/Proton and somehow integrate nicely with pipewire would've been enough...

Pulseaudio is so extremely over complicated that it took years of distro maintainers trying to wrangle it into a working default configuration for many people, and it was adopted because ALSA didn’t (but was going to) support multiple inputs to a single output.

It is a prime example of short-sighted “just rewrite it then!” Style Behaviours in FOSS and continues to be brought up in conversations about subsystem replacements (such as systemd).

Pipewire is a good replacement, but pulseaudio is to JACK what windows movie maker is to Final Cut.

It doesn’t deserve to be brought up when talking about professional solutions.

It really is amalgamation of "whatever works fine as default on Lennart Poettering's private laptop". Hell, I still remember where "sec, I need to restart pulseaudio" was common thing I've read from my colleagues any time they needed to run some voice chat app...

ALSA before that wasn't great either, the fact you didn't get any software mixing by default and needed to fuck around with dmix gave way to endless problems with various software, and the fact dmix was still working subtly different than "real" device also reared its ugly head in many "pro audio" apps.

> ALSA before that wasn't great either, the fact you didn't get any software mixing by default

It was being developed.

People jumped on Pulseaudio and made it the default before it was finished because the ALSA contributers wanted to do it in a way that was either going to work for embedded or be optional in a reasonable way, IE; they were trying to do it right, took too long, and it got replaced with something much worse that did the job.

Which, happens unfortunately frequently.

Similar story for how ALSA came to exist, really. Linux folks looked at OSS and said "this blows, let's replace it" and made ALSA.

The FreeBSD folks looked at OSS and said "this blows, let's fix it". And did. And they're still on that while Linux is headed into its third major audio shake-up in the same time span (OSS -> ALSA -> PulseAudio -> PipeWire).

Saying this as FreeBSD user: FreeBSD didn't switch this many times because no one is working on a new thing. The community is already small, the community of people using it on desktop is even smaller. Developers that can work on any of this - almost doesn't exist at all.

Corporate money and contribution go to: storage, network, scheduler. Anything desktop related doesn't get any love. I bet that PR for EoL drm-kmod from linux still not merged.

I could never understand why it was adopted. It was written to solve problems that nobody faced. I still remember Fedora Core 2 (??) being useless with audio due to them adopting PulseAudio. I try to minimise my contact with any of Poettering's software as much as possible since then.

I still remember being shocked at the bloat. It sat there eating something like 100x the memory of my entire sound stack before that (granted, easy to hit a large multiple when the baseline is very low), with pretty frequent CPU spikes, all while no sound was playing. WT actual F. And the CPU use was straight-up vulgar when it was actually doing anything.

It was then that I decided whoever was responsible for it had no software-architectural taste whatsoever. I didn't yet know his name. His work since then hasn't changed my mind.

I thought to myself, I think I'll stream some Codewars challenges on Twitch. I have a Shure mic, a Sony ZV-E10, and some speakers for listening to music. Surely Linux can handle all this and pipe the audio into OBS...

Yes... it is possible. No, it's not something I ever want to do ever again.

I don't understand why you bring pulsaudio to the table when answering about JACK. This is unrelated.

windows still has the audio mixer to beat

Which is why every professional audio tool uses ASIO and every audiophile player is going to great lengths to avoid the audio mixer and uses things like kernel streaming.

Does MacOS even have OS level independent volume control for each app yet?

Sort of? The kernel supports this and has forever, but the OS doesn’t ship with an app that exposes the capability. It’s super annoying. There are a number of third party apps, some free, some paid.

Not that I've found!

Why should the OS solve app problems?

Yeah.

Why do we need `open()` anyway?

(this is sarcasm, the OS exists only to solve app problems).

This was what everyone said was the reason for wrecking Linux with PulseAudio, all those years ago.

I've still never cared and don't know why I should. What am I missing? I've used systems with that, but never bothered to try using it because... why would I? Every media player has a volume control, YouTube has a volume control, every game has a volume control (usually multiple, for different types of sound), and 99% of the time I just want everything at about the same level anyway. The last thing I want to do is have another place a given program could have its volume set or muted or whatever, so then I have another place to look when it's not doing what I want. One per-program (built in to the programs) and one global is quite enough.

For me it wasn't so much the per-app/stream mixer but per-stream dynamic routing that was the important feature of Pulse (though I've moved on to PW now).

I'm not too sure how this is handled on other platforms, but pre-PA the in-app volume control would often modify the global mixer, rather than implement an in-app mixer/attenuator, which is definitely not what you usually want. In most Linux apps these days, those controls manipulate and are synchronized with the PA mixer, so there's still only one actual mixer, they are the same control. I assume that other OSes also do per-stream mixing in a similar manner, and just choose not to expose the mixer for whatever reason.

As much as people complain about PulseAudio it's always worked well for me, and from what I've seen of Windows and macOS alternatives, it's more featureful and usable out of the box.

You can’t even mute Zoom on Mac OSX. Frequently I want to leave video up but mute anyone that is talking.

I use SoundSource for this, but I did pay $39 for it.

https://rogueamoeba.com/soundsource/

> Ironically Pro Tools is still not M1 native.

that does not surprise me at all, unfortunately.

> One cool thing in MacOS that Windows can't do is aggregate of audio devices/interfaces.

it does come at the cost of latency & jitter: fundamental issues that ultimately stem from having two separate audio clocks. and it's not really a specific limitation of windows, which tend to use ASIO for low-latency audio. there's nothing stopping a an aggregate ASIO driver from being written, i just can't imagine it'd be that useful, indeed, i've only ever used the macOS aggregate device a handful of times, mostly only to try it out.

I used it in Linux, but that's because it detected 2-in/4-out audio interface as just three stereo pairs and I needed to split those inputs off to mono.

It was a lot of CLI fuckery, not a pleasant experience... and of course pulseaudio can't just remember it, need to be re-applied after every USB reconnect.

Can be done with UDEV but not exactly something random user would know how to do.

The one time I've actually used an aggregate device, I synced the clock of one device to the other. It worked fine as far as I remember (it's been a couple of years).

Mine didn't in practice. I think one was a FireWire audio interface and the other USB. Never could get it to work properly.

"One cool thing in MacOS that Windows can't do is aggregates of audio devices/interfaces."

Some of us didn't need it, software at the time allowed us to record and mix from and playback to different audio hardware devices simultaneously already. I remember doing that in Windows ME with some audio editing software.

> Ironically Pro Tools is still not M1 native.

Public beta this month, though. https://duc.avid.com/showthread.php?t=422724

Public beta coming soon. https://www.pro-tools-expert.com/production-expert-1/apple-s...

How did NT beat SGI? Cheaper hardware?

Cheaper hardware, faster CPUs, open OS platform.

In the end SGI tried to enter the Windows NT market with its own system that had a unique GPU with unified memory and also used SGI's own firmware instead of a PC BIOS:

https://en.wikipedia.org/wiki/SGI_Visual_Workstation

Despite the GPU, it wasn't competitive with cheaper, more standard x86 hardware.

I think "NT", here, should more accurately read "the Wintel platform". NT bringing a "modern GUI" to DOS computers (as well as some RISC systems) and intel retaking some of the high end previously lost to Sun, Silicon Graphics, DEC, etc. So yes: price, primarily.

> NT bringing a "modern GUI" to DOS computers

NT wasn't a GUI on DOS. NT was different, not based on DOS in any way. Dave Cutler brought his pet project NT (which was technically Digital IP) with him when he left Digital. NT was new and entirely unrelated to any other project at Microsoft.

Traditional Unix workstation vendors made a bet for Itanium and against their own architectures, so in the late 90s / early 2000s their investment into their archs became minimal. But Itanium flopped and the Dotcom bubble burst. The latter impacted Sun a lot. At the same time x86 kept improving massively, mostly pushed by AMD, which also put AMD64/x86-64 in the market by the mid 2000s. Also in the same time frame nVidia and ATI created monolithic GPUs which started to perform better than the proprietary GPUs by SGI and HP. The last Unix workstations from HP actually had ATI or maybe nVidia COTS GPUs.

All these factors together largely eliminated Unix workstations from most of their native markets (3D visualization, CAD, CAE, media) so that category of system just ceased to exist in the mid 2000s.

I thought Itanium was just an HP thing? HP killed their own PA-RISC and DEC’s Alpha (which HP acquired via Compaq) in favour of Itanic, which left POWER (still holding on at IBM), SPARC (pretty much died with Sun), and MIPS (dunno, tbh).

SGI bet what little remained of the farm on Itanium and even sold a handful of Itanium-based supercomputers.

It was an odd strategy from a company that obviously had no future and nothing left to offer. At the time, I had built up so much animus against SGI for their outrageous pricing and unbelievably annoying FlexLM software licensing that it was amusing to me. Their god damned MIPS Pro compiler spent more cycles checking its license file than actually compiling a small C file. And the FlexLM system was so badly written that it did nothing to stop piracy and any user could trick it into overwriting any file on the system. They claimed to patch it, but never actually fixed it. At least they open-sourced XFS. That was nice of them.

https://www.zdnet.com/article/nasa-gets-sgi-2048-core-itaniu...

So to reword what you said, nobody was writing games for SGI workstations.

The video game industry is so huge that whatever hardware benefits from it, it will eventually crush the professional competition.

If AI is possible today, it's because billions have been invested in making our computers beasts at multiplying matrices.

For the record, I played a lot of Creeper CTF mod quakeworld on my Octane MXE. I don’t even want to remember what that machine cost me. Unfortunately, the impact graphics chipset the MXE used in a 4x SLI-like configuration had an issue with sub texture support that kept frame rates in the 20 FPS range.

A single voodoo2 in a pentium2 absolutely trounced the Octane, so my gaming eventually moved to pc and stayed there.

First cheaper hardware, and later simply better performance. I worked at a small animation studio around 1999-2003 and the sudden price/performance improvement that Windows on Intel + Nvidia all of a sudden offered was unlike any I've seen since. A Geforce 2 Ultra and dual Pentium IIIs stomped all over our SGI workstation at most workloads, at less than quarter of the price.

Plus since virtually all the animators used Windows at home and where very comfortable with it, the need for support and hand-holding was much less than with Irix.

> later simply better performance

You probably know this but a reminder: it was a double whammy of scale (network effect): SGI was selling to roughly three markets (video production, high end CAD, and midrange number crunching). NT was selling to anybody, and despite its multi-architecture support, Intel was selling to the same people. This threw off cash both of them could invest in improving their offering.

There is a good argument, though I don't know how good, for Apple to do an SGI with the Mac Pro. Its main value would be a "halo" effect: make the argument that Apple is so technically great that the people with the hardest problems turn to them, so you can too. Also Apple helps bring you the stuff you like, movies and music and such. That was an important argument back when Apple was lagging. If this argument still makes sense it might even worth doing so at cost or even a small loss paid for by the marketing department. But the green eyeshade side of Apple's planning hates that kind of excuse for good reason.

The Mac Pro has some minor benefits in being a practical way to push the design envelope in ways that will later trickle down, though as this article points out it requires investment in areas not otherwise critical, or even important to the company.

It's always strange to hear Windows being actually good in the 90s and early 2000s, but so much of that was the fact that the consumer oriented, DOS based OSes, were so poor.

The switch to the NT platform in Windows XP was incredible. Of course, a lot of that was lost in teh beginning with the "Fisher Price" UI, and the massive rise of malware, but by SP2 most of these issues were resolved... Windows XP SP2 was one of the most stable and long lasting consumer OS of probably all time.

I wonder if Microsoft could have switched to NT for its consumer OSes earlier, and if so, how much better Windows based computing for most people would have been. Combine that with the Longhorn disaster, and MS lost so many years of precious development time.

You could always turn off the Fisher Price UI and switch to Windows Classic style, up to Windows 7 actually.

Windows NT 4.0 was tight in its time. (Despite the name, NT 4 was really version 2.0 of the NT operating system.)

It had the new kernel that could compete with (and even outperform) commercial Unix for workstation use, and also the new GUI from Windows 95 which meant it was both familiar and had an enormous range of apps easily available.

Not only was Windows NT 4 enormously better than DOS, but it was also objectively better as a total package than Mac OS 8, the Sun Solaris CDE desktop, and everything else. Open source Linux desktops were still very early.

Windows 2000 (a.k.a. NT 5) was still good and a much more practical desktop OS than the first Mac OS X which had absolutely horrendous performance. Somehow Microsoft managed to squander this lead by focusing on consumer Windows XP and their own endless .NET API turf wars that ended up destroying the credibility of the Windows platform.

> Not only was Windows NT 4 enormously better than DOS, but it was also objectively better as a total package than Mac OS 8, the Sun Solaris CDE desktop, and everything else.

The issues were that Windows NT needed much more RAM than consumer Windows at a time when RAM was still extremely expensive, and only a subset of the hardware that would work under consumer versions of Windows had drivers for Windows NT.

However, NT (and especially Windows 2000) was night and day better than Windows 9x.

The sad thing is that we remember that era and being used to fast user interfaces and applications even on the ancient hardware. Today's software is so sluggish and slow and all web based even when native toolkits are perfect. Makes me weep.

At the time a single seat of PowerAnimator, the predecessor to Maya, cost $50k-$60k. $30k was for the SGI machine to run it and the other half was the software license.

Microsoft bought Softimage, one of the top competitors to Maya at the time. They charged iirc $4k or $8k and a PC to run it would run $4k-$5k

Softimage is long gone but it's what Valve used for the Source Engine and Half Life 2

The same old story: the rise of «worse is better». UNIX workstations hardware was always miles and miles better[0], although it could no longer compete with the commodity PC hardware due to being exorbitantly priced. Coupled with highly aggressive and anticompetitive practices of Microsoft, the demise was inevitable and quick.

[0] With the exception of the lower end UNIX workstations which were still better, yet a tad bit more expensive compared to the commodity PC hardware.

> UNIX workstations hardware was always miles and miles better[0], although it could no longer compete with the commodity PC hardware due to being exorbitantly priced

This was true early on, as in, early 90s-ish. But in the second half of the decade the balance shifted significantly and by the late 90s, Unix workstations could no longer compete with the commodity PC hardware due to being slower. I was doing sysadmin grunt work for a studio at the time and beige boxes ran circles around later generation Octane and Onyx machines.

They were also a nightmare to maintain by then. The market for peripherals and the like wasn't exactly open and there were all sorts of weird servicing deals. With beige boxes we didn't even need a formal procurement process for peripherals like keyboards and the like, we let people buy their own shit and reimburse them within some limit.

> […] The market for peripherals and the like wasn't exactly open and there were all sorts of weird servicing deals. With beige boxes we didn't even need a formal procurement process for peripherals like keyboards and the like

Yes, this is true enough as the situation was incomparably better with commodity PC's, with upgrade components and peripherals.

But to be fair, any branded PC was exactly the same, e.g. memory sticks they required were not compatible with the affordable or reasonably priced commodity ones (sometimes even across different models from the same PC vendor), expansion cards (even if made by a third party) also had to come from the same vendor etc, and everything that was branded came with an exorbitantly priced component price list. Compaq was notorious (and nearly universally hated) for requiring special SIMM's that could only work in Compaq PC's and only in a specific model. Procurement loathed those. IBM, HP and Dell were just the same. Then, there was RAMBUS trying to corner and lock everyone into their über-expensive memory chips and stuff (which, luckily, faltered and the company went bust). Again, I am talking about the desktop PC's not the servers.

The situation only changed after the market got flooded with cheap Taiwanese made PC's that could be easily and cheaply upgraded, so the big brands feeling the market force and pressure had no choice, eventually budged and stopped adding proprietory control pins to their memory sticks and stuff.

“Worse is better” is just a framing of “economies of scale beat niche products” from a niche enthusiast’s viewpoint.

Cheaper hardware, cheaper software, and PC vendors didn't try to force you into ludicrously expensive support contracts. All in all just way more customer friendly than the workstation companies.

And of course the huge number of people buying x86 hardware meant PC hardware manufacturers could afford big R&D budgets that allowed them to surpass the workstation CPUs and GPUs in performance.

Reply on news.ycombinator.com