Low-resource open source developers pretty much screwed now

[OpenFreeNet]

Sapling follow a "stack approach". You have polite commits, and "working" commits. You can fold and split working commits. Every operation is saved also in a private "operation log", so you have both a clean repo history, and low level details of every dirty operation you have done. You can revert anything. You can move working commits like in a stack. When you are happy, you push finished commits to a Git repo, and during review they can improve them. In practice the operation on the history of commits are more natural and composable.

In a Git repo you can have many branches with various experiments/features. In a Sapling repo usually you have a linear history of the already pushed/public commits on the bottom, and a tree of working commits on the top of the log. But this "chaos" of commits is fully visible, as a tree. You can continuosly move commits in this tree. When something become production-ready, you push the finished commits in the bottom part. In the top you have always working commits.

Every refinement you do with the commits is visible in the private repo with sl journal. So when you manage commits, you don't loose anything, if you do the wrong command.

 

[drhowarddrfine]

something of good there must be in the language

No language determines the value or usefulness of any app

 

[drhowarddrfine]

What's wrong with git?

Everyone started using it only because Linus invented it over a weekend to get rid of the version control system he was using then. No other reason. Just like my previous comment about Rust. There was better version control then and still now

 

[CShell]

With LLMs that changes. You either need a monthly subscription to an AI service (and the $20/month option doesn't reach very far), or you need hardware to run an LLM locally. You'd spend about $3500, which gets you either:

I disagree.

You don't need AI to be competitive as a computer programmer. In fact you had better know how to survive WITHOUT AI or your going to have a very rough time as a programmer. Programmers regularly get programming solutions from AI that do not work ... as in the AI provided solutions do not function at all, or within spec, or perform extremely badly. For a programmer "to know" if an AI solution is going to work or not... they literally have to know how to program without AI.

As for everything going up in price $$$$$ -- that was the entire point of AI to begin with.

You would have to be pretty naive to think an "AI PC" was going to be cheap . It's an "AI PC" -- it is "the Porsche of PCs", even though it is really not. No one can even explain in plain English "WHAT an AI PC actually is" - even today.

Color Montors? Dialup Modems? Internet? Neural Networks? Cloud? Crypto? Next "Valley buzz word"? - same thing. AI was always an excuse by (everyone) to raise prices.

Computer programmers were writing "AI" in the 1970s at the same time they invented the EMACS editor - what do you think the LISP and PROLOG programming languages were used for back then? AI isn't something "everyone suddenly discovered" in the year 202x ....

Heck we were running (Link Wikipedia: ELIZA) using the BASIC programming language during the 1980s (The original ELIZA was designed between 1964 - 1967) and we thought we had achieved the "REAL DEAL" AI back then ! Maybe ... we had?

 

[kpedersen]

Naive yes, because I didn't checked. It seems so natural to me that a rather modern ARM CPU can beat the worst processor Intel made, 25 years ago, also in single-core tasks.

An ARM CPU for an embedded SoC originally intended for set-top boxes?
This isn't a laptop-class Snapdragon or Apple's desktop offerings we are talking about here.
So no, your instincts may not quite be on point here. At face validity, "new" may feel better but this is certainly not always the case.

The first benchmark I found. In single-core, it is 1104-991 (the Cortex 2.4 GHz and Cortex 1.8 GHz) vs 237 (P4 3.83 GHz). Benchmarks can lie, but it is 5 times better, and you have 4 cores, instead of 1, and a better instruction set.

Yeah, many real-world tasks cannot be parallelized. You mentioned you do not have access to a P4 era machine and unfortunately its one of those things you just need to check out for yourself.

These are benchmarks of the VideoCore VII.. To be honest, they are not very good. You cannot play many games. But it is a full featured GPU. It decodes all video formats in hardware. It supports all modern API like Vulkan. You can use it on the desktop without problems.

For all practical use, it beats the old ATI card, especially if paired with an ancient P4.

Not a chance. Try running the following on the VideoCore chip (ignoring the expensive translation you would need to do on the incompatible ARM CPU):

• Half-Life 2
• BioShock (2007)
• Call of Duty: Modern Warfare (2007)
• Need for Speed: Most Wanted (2005) & Carbon

These run very well on the ageing radeon. The VideoCore chip may support Vulkan, but has zero power to actually make use of it. I feel you may be falling for the same marketing "tactics" that saw the GeForce 2 MX (PCI) get sales (a very under-performant GPU that happened to support the latest DirectX at the time)

FreeBSD 15 does not support i386, so it is even worse. At least I can hope that the Linux drivers will be ported to FreeBSD.

It is more likely that FreeBSD picks up i386 again (to keep parity with OpenBSD and NetBSD) than the Linux drivers (and proprietary blobs) ever get released for FreeBSD.

In any case this is the last time I will answer you. I replied to non-sense, and it is a waste of time. Probably you are thinking the same of mine answers.

You are speaking unfortunately with no hands on experience. You don't have the machine to test. You can only fall back on the words "modern" and "old" which is simply childish and consumer-centric. I also suspect that if you had more development experience, you would realize the toolchains can run on a potato and you don't need to keep shelling out for the latest kit.

 

[kpedersen]

Everyone started using it only because Linus invented it over a weekend to get rid of the version control system he was using then. No other reason. Just like my previous comment about Rust. There was better version control then and still now

I do tend to agree. Git is absolute overkill for many scale projects and teams. Yes the branching is still a little easier than SVN 1.6 but frankly, that could be integrated. It doesn't need to be a distributed VCS for that.

I think Epic and their new lore VCS is quite an interesting idea. I am glad to see work is still happening in this area.

 

[kpedersen]

Nothing. The FreeBSD 14.x or 15 DVD doesn't boot, the kernel hangs immediately.
This machine was my daily driver until 2015, it doesn't have issues. It is a rock solid system based on Intel, cpu, mainboard, chipset and SSD drive.

Interesting. Since you do have better machines, its probably not worth exploring but I did get similar with OpenBSD. I had to disable ACPI, APM and radeon during the install. Only after I installed the firmware did the radeon not cause a panic and I could also re-enable ACPI, APM.

As you know, there is more to a computer than the CPU. It could be a whole heap of things. Its strange because older machines tend to be better supported but its always a sliding window.

Oh I do not disagree with that. Agreed fully that open source development can be done on an older system, even a legacy platform.
What I add to that, is that 30 pounds can also buy a much more powerful, newer, 64 bit CPU platform, compared to P4.

Oh right. Yeah, of course you can optimise budget. if you play the cards right and go fishing at the right time of the (tax) year, then you can grab a decent-ish X220 Thinkpad for around that cost. This is also avoiding the concerns of extra costs (i.e SSD, PSU, GPU, etc).

To clarify, my original response was really a rebuttal to the statement that you need to shell out $1000 to get involved in open-source development. My reason for using the P4 as an example was precisely because it was cheap and not considered powerful today. Gatekeeping "open-source dev" behind needing expensive, powerful machines is absurd.

 

[OpenFreeNet]

Yeah, many real-world tasks cannot be parallelized. You mentioned you do not have access to a P4 era machine and unfortunately its one of those things you just need to check out for yourself.

C code compilation is easily parallelizable. Every core is 5x faster than the fastest P4. You have 4 of them, so 20x the speed. There is no way a P4 can be better than a Raspberry Pi 5 for coding OSS projects. I haven't this hardware at hand, and I cannot benchmark, but I doubt I will be surprised.

Not a chance. Try running the following on the VideoCore chip:

• Half-Life 2
• BioShock (2007)
• Call of Duty: Modern Warfare (2007)
• Need for Speed: Most Wanted (2005) & Carbon

You said P4 + its GPU is better for programming. Video games are not important. The Raspberry Pi GPU can run web browsers, play videos, manage modern GUI using OpenGL ES and Vulkan. The P4 combo no, or with serious limitations. For coding is better a computer that can display GitHub home page, respect one that can play Half-File 2.

You are speaking unfortunately with no hands on experience. You don't have the machine to test. You can only fall back on the words "modern" and "old" which is simply childish and consumer-centric. I also suspect that if you had more development experience, you would realize the toolchains can run on a potato and you don't need to keep shelling out for the latest kit.

I installed P4 in the past for sons of my friends. Youtube videos where unplayable after a while, because Youtube changed codec, and the video where not anymore hardware accelerated, and the computer become nearly unuseable for their tasks.

I regularly bought (before the prices increases) i3/i5/i7 dismissed from companies, for my friends and parents, and with Linux they are still happily navigating in internet. Intel integrated GPU are very well supported and snappy. I spent 50-90 EUR for each of them, and their speed is top-notch for their usage scenario. So I value old hardware. We lived in a dream era, where hardware was improving more than the real needs, and prices where down-falling.

I still use one of the first Intel 64bit CPU, with also an IDE connector and an old IDE HDD plus various SATA1 HDD, as offline ZFS backup servers. Sometime I power on, and it backup the data.

The P4, except for retro-computing, is the worst architecture sold by Intel. I'm happy you love your P4 computer, but you are confusing love with logic, when debate its utility.

 

[kpedersen]

You said P4 + its GPU is better for programming. Video games are not important. The Raspberry Pi GPU can run web browsers, play videos, manage modern GUI using OpenGL ES and Vulkan.

So the VideoCore XII doesn't support OpenGL 3.3+.

Games aside, I don't believe it can even run Blender. You need to fall back to a software renderer on the weak bean 4 core ARM CPU.

Also, OpenCL doesn't look possible. Unlike a Pentium 4 machine, you can't easily add an NVIDIA GPU for CUDA. The Raspberry Pi 5 may be newer, but it is a toy. As such I don't see why you think open-source development is possible on a toy vs an ageing workstation.

The P4, except for retro-computing, is the worst architecture sold by Intel. I'm happy you love your P4 computer, but you are confusing love with logic, when debate its utility.

I've got loads of various machines. I am stating that a software engineer can easily develop software on all of them. Not sure why you feel you would struggle.

 

[OpenFreeNet]

So the VideoCore XII doesn't support OpenGL 3.3+.

Games aside, I don't believe it can even run Blender. You need to fall back to a software renderer on the weak bean 4 core ARM CPU.

You can probably run Blender on the Raspberry because last version of Blender supports Vulkan for the UI. You need to use the CPU for the effects.

For sure you cannot use the P4, because it is far away from the minimum requirements of CPU instruction sets and RAM.

Smartphone chipsets tend to support OpenGL ES and Vulkan. The full OpenGL is usually not supported, IIRC. But probably the future will be Vulkan, because it uses the GPU shader logic in a rather native way.

Also, OpenCL doesn't look possible. Unlike a Pentium 4 machine, you can't easily add an NVIDIA GPU for CUDA.

CUDA on a P4!? Are there NVIDIA drivers running on i386? This discussion cannot be serious.

The Raspberry Pi 5 may be newer, but it is a toy. As such I don't see why you think open-source development is possible on a toy vs an ageing workstation.

Raspberry Pi 5 is a toy. True 100%. In fact for every serious OSS development use a i3, i5 and so on. Install discrete GPU. Use CUDA. But speaking about P4 for development is a border-line trolling activity.

 

[drhowarddrfine]

Are there NVIDIA drivers running on i386?

Yes cause the 9600GT I had in my current machine was pulled from a Gateway P4. But I pulled it from my machine cause I found out that the Intel processor I have could do better web graphics processing than the Nvidia card could.

 

[kpedersen]

You can probably run Blender on the Raspberry because last version of Blender supports Vulkan for the UI. You need to use the CPU for the effects.

The Raspberry Pi can run the UI... What about the rest of it? The actual 3D viewport
Try it out. Look around some forums, you will see it can't run a recent Blender. You would know this if you tried. The error you would get is:

blender
Error! Unsupported graphics card or driver.
A graphics card and driver with support for OpenGL 3.3 or higher is required.
The program will now close.

So the best you have is LLVMpipe (software rasterisation entirely on the CPU)

For sure you cannot use the P4, because it is far away from the minimum requirements of CPU instruction sets and RAM.

I can run Blender on it no problem. FreeBSD and OpenBSD aren't even listed on the minimum requirements. As a technical user, you are generally safe to ignore them.

Check out the Radeon HD 4670. You will see OpenGL 3.3 support. This is the key requirement that Raspberry Pi's GPU lacks.

CUDA on a P4!? Are there NVIDIA drivers running on i386? This discussion cannot be serious.

You run CUDA on a GPU, not the P4 processor so that does not affect things. Yes there are i386 drivers, why wouldn't there be? How do you think 32-bit games work?

The Radeon HD 4670 has ATI Stream (the predecessor to modern AMD ROCm/HIP) and OpenCL 1.1 support.
GeForce cards from that same era (i.e GeForce 9600 GSO) of course have support for CUDA and obviously include 32-bit drivers.

But speaking about P4 for development is a border-line trolling activity.

Yes I can see how someone with Dunning-Kruger would arrive at this conclusion. But I feel it has been a useful exercise for you to realize that you don't need "new" to get involved in open-source development.

To clarify and conclude, whilst I have spoken quite frankly, I don't blame you, the industry wants to keep people on the hardware upgrade treadmill to extract money from us, so I can see how misinformation of "new is necessary" has propagated.

And finally, some trivia for those who have been mislead that 32-bit is always worse than 64-bit. Modern GPUs focus mostly on 32-bit floating point. Not 64. Have a read of this.

FP32 is the most widely used for its good precision, and reduced size. The larger the floating-point number, the longer it takes to run those highly specific values through calculations. A number of workloads don’t benefit from the higher precision, while some workloads are only feasible using lower precision formats

 

[Beastie7]

The Raspberry Pi 5 may be newer, but it is a toy.

The RPi 5 has 6x the floating point performance of that P4 at a TDP that’s 4x lower. It supports newer ports, codecs, and graphics libraries, arch extensions. It’s anemic (for heavy 3D stuff at least), but I wouldn’t call it a toy either. It wasn’t designed for that anyway.

I don’t think comparing a dGPU to a SoC is a fair argument either. The RPi 5 is great hardware for its size and power consumption.

New and cheap do exist.

 

[OpenFreeNet]

I can run Blender on it no problem. FreeBSD and OpenBSD aren't listed on the minimum requirements. As a technical user, you are generally safe to ignore them.

Are you using a very old version of Blender? Are you compiling it yourself?

A web search tell me that Blender does not run on i386. But it can be hallucination. FreeBSD freshports tell me that Blender cannot be built for FreeBSD 14 on i386. But sometime FreeBSD packages are broken. NetBSD too, cannot support Blender on i386.

But yes (I checked better): the GPU of Raspberry Pi 5 is its weakest part. It has a lot more powerful CPU in comparison. So, no GPU scientific software on it. The GPU is good only for the UI.

In theory the Raspberry Pi 5 should be supported from recent versions of Blender, because the UI now it is Vulkan based. But the CPU must be used for the rendering. So far from optimal. But I don't want to buy a Raspberry for winning a discussion on a forum

You run CUDA on a GPU. Not the P4 processor. Yes there are i386 drivers, why wouldn't there be? How do you think 32-bit games work?

The Radeon HD 4670 has ATI Stream (the predecessor to modern AMD ROCm/HIP) and OpenCL 1.1 support.

Retro gaming with legacy drivers is a thing and it makes sense for full native experience.

OpenCL and CUDA OSS development on an archaic platform, using legacy drivers with tricks is madness, considering that with a very limited cost, you can use full supported AMD64 drivers and a vastly superior CPU, like an i3 or some free Core Duo CPU. It is an usage scenario that has no common sense.

Yes I can see how someone with Dunning-Kruger would arrive at this conclusion. But I feel it has been a useful exercise for you to realize that you don't need "new" to get involved in open-source development. To clarify, whilst I have spoken quite frankly, I don't blame you, the industry wants to keep people on the hardware upgrade treadmill to extract money from us, so I can see how misinformation of "new is necessary" has propagated.

I'm a regular user of old hardware. But the line I use is an amd64 CPU able to run a modern web-browser. At best of my knowledge a P4 is not that type of system.

 

[hedwards]

I do tend to agree. Git is absolute overkill for many scale projects and teams. Yes the branching is still a little easier than SVN 1.6 but frankly, that could be integrated. It doesn't need to be a distributed VCS for that.

I think Epic and their new lore VCS is quite an interesting idea. I am glad to see work is still happening in this area.

It is, but given how easy it is to self host just in a directory or via gitea and one of the other clone webservers, I'm not sure there's much of a reason not to use git at this point. There is a bit of a learning curve, but there are some excellent tutorials covering how to use it and it didn't take that much time for me to get it up and running. I think the bigger issues for me personally were working out how best to actually utilize it rather than just how to operate it.

That being said, depending upon what sort of files you're dealing with, it might not be a great choice. It's definitely geared more towards source code than configuration files.

 

[kpedersen]

It is, but given how easy it is to self host just in a directory or via gitea and one of the other clone webservers, I'm not sure there's much of a reason not to use git at this point.

Popularity and wide range of tools is indeed a large reason people choose git I suspect. That said, a small light standalone server like svnserve does tend to be lacking in the Git ecosystem (gitdaemon is close but has no authentication) and actually my main reason for sticking with svn for many projects.

Gits options don't seem ideal. For example SSH as transport is too powerful that it is a potential hazard, large web systems like gitea, etc have a very wide attack surface compared to i.e svnserve.

Granted, not many people actually *serve* git repos so they probably don't care about this stuff.

 

[blackbird9]

Doesn't anyone use sccs nowadays? It's still available, from the open group. Fine for small projects. Even rcs can work for small projects.

Source Code Control System - Wikipedia

en.wikipedia.org

sccs

Maybe sccs isn't powerful enough for modern work though.

There's always Subversion, from apache https://en.wikipedia.org/wiki/Subversion_(software)

Git is a whole ecosystem, the tools themselves, and the web repositories, all that infrastructure, and everyone knows it.

 

[OpenFreeNet]

In theory the Raspberry Pi 5 should be supported from recent versions of Blender, because the UI now it is Vulkan based. But the CPU must be used for the rendering. So far from optimal. But I don't want to buy a Raspberry for winning a discussion on a forum

This user is using Blender on the Raspberry, but yeah, it can be used only for basic modelling, because nearly all the work is delegated to the CPU instead of the GPU.

The user says that the Raspberry GPU does not even support OpenCL in a good/full way. So for any serious scientific work, or advanced 3D, the GPU is unusable. It is good only for UI and video.

In any case, also Debian does not support Blender on i386. So, between nothing and slow, slow is the winner.

I can agree on many things said from kpedersen , if instead of a P4 workstation, there is some old i3.

 

[rbranco]

All I can say is:

- Don't take anything for granted. Things are a lot more unpredictable now.
- Always strive to have an open mind. This is hard to do because our natural tendency is to have an opinion before getting all the facts, which is also not easy because there are many unknown unknowns.

I regret not buying some powerful machines when they were cheap to do distributed inference and experiment with all the crazy stuff Salvatore Sanfilippo is doing:

antirez (@antirez) on X

Reproducible bugs are candies. I like programming too much for not liking automatic programming.

x.com

I'm waiting for the "bubble" to pop so I can buy lots of hardware before they increase in price again.

LLMs came here to stay in one way or the other. There's a lot of stuff I can't bother to do manually now like test code which has always been a PITA.

And don't tell me that the Internet is being inundated with slop because of AI... We had lots of slop before AI in the form of cookie banners, brain-rot videos, pornography, podcasts from people giving unsolicited life advice, etc. This cheap entertainment is going to be generated by LLMs now. You can't fix mankind to stop consuming it anyway.

I think the future is local open-source models. And I agree with the main premise that those who can run those have a natural advantage over those who cannot. But this can be fixed.

 

[blackbird9]

LLM's are definitely here to stay. Once that box was opened, they're never going to shut it again.

 

[kpedersen]

In any case, also Debian does not support Blender on i386. So, between nothing and slow, slow is the winner.

Without digging on with this discussion too much, I would like to present a third option:

Running Blender ~2.8 at full speed on even an ageing Pentium 4 workstation.

It would give the better experience and allow you to get relatively large work done. Chances are we don't even need the newer features (Embree ray tracing?). Again, I know the concept of "older is sometimes better" does feel uncomfortable if you are not used to it but kludging along with a modern and but slow solution is not "better" in my opinion.

 

[CShell]

LLM's are definitely here to stay. Once that box was opened, they're never going to shut it again.

I agree - they are.

But it will be interesting to see what happens to AI when the next "shiny object" takes over in popularity.

 

[OpenFreeNet]

But it will be interesting to see what happens to AI when the next "shiny object" takes over in popularity.

There were the Personal Computer revolution, the Internet revolution, the Smartphone revolution, the Cryptocoin buzz, and now AI. Of all these revolutions, the most doubtful are the cryptocoins, because the majority of buzz was driven purely by financial speculations. But there is also some real usage of block-chains.

If we separate the marketing/financial aspects of these technologies, from the innovative aspects, the innovations were reals. And the same will be for AI.

Many Personal Computer companies that seemed giants, are now failed, but in the meantime they generated a lot of money and they were useful. So the volume of good money they moved was some order of magnitude more than the debts they left in their final phase (or I hope). The problem is that in current world, dominated by greedy finance, we risk that failures in AI companies will generate a disaster.

In the long term, we need also to find a new economic order, because consumerism and finance are destroying the planet. AI can be the first technology removing more jobs from economy, respect the new jobs that will be created. We need to find a better way for distributing work, but also the richness generated from the technological improvements.

 

[blackbird9]

In the long term, we need also to find a new economic order, because consumerism and finance are destroying the planet. AI can be the first technology removing more jobs from economy, respect the new jobs that will be created. We need to find a better way for distributing work, but also the richness generated from the technological improvements.

Let's hope it works out ok for the humans.

 

[blackbird9]

"An artificial-intelligence bust (and thus bubble), inflation and fiscal stress are the three the most alarming threats to global prosperity at present, the Bank for International Settlements warned."
The full BIS report is here:-

Annual Economic Report 2026

This year's Annual Economic Report examines how the global economy is faring as progress meets rising perils – including a new fiscal-financial stability nexus and shifting inflation dynamics.

www.bis.org

"The assessment highlighted AI-led risks prominently in a report that arrived on the eve of the ECB’s three-day annual symposium in Sintra, where a host of global policymakers will also scrutinize such stability dangers closely."

Of course, they are questioning the current AI and semi stocks valuations; who wouldn't, really. Look at these two sets of charts. Capex expected to reach 800 bn dollars this year; it's getting close to a trillion.

“Disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust, with potential knock-on effects on financial conditions,” the BIS said, before observing that “a major equity-market correction could have larger macroeconomic consequences today than in the past.”
"On AI specifically, officials highlighted vulnerabilities linked to funding, including complex arrangements such so-called “circular financing” deals that can mix equity and debt with supplier-client contracts"
- in other words, companies buying each other's stuff in a circular pattern, all ultimately dependent on the current astronomical stocks valuations.

In essence, they are saying we may well be in DotCom 2.0. All we can do is 'watch this space', I guess. Admittedly, not a staggeringly surprising conclusion, however, for the normally rather staid and reserved BIS to come out and say so publically, is noteworthy. Let's hope we don't get "dotcom crash 2.0" at some point in the next couple of years.

A couple of other headlines I noticed today.

GM replaces 1000 factory workers with robots.

GM Fired Over 1,000 Factory Zero Workers And Installed 50 Robots

General Motors is facing renewed scrutiny over automation at its flagship EV assembly plant after adding dozens of robots to the production line months after cutting more than 1,000 jobs. The changes at Factory Zero ... Read more

finance.yahoo.com

BAT fires 9000 workers, 1/5 of its workforce, to be replaced by AI.

https://www.thisismoney.co.uk/money/markets/article-15937839/British-American-Tobacco-cut-9-000-jobs-turns-AI.html?ico=mol_desktop_home-newtab&molReferrerUrl=https%3A%2F%2Fwww.dailymail.com%2Fhome%2Findex.html

And it seems 'tokenmaxxing' is suddenly a thing of the past in the Valley, who knew? For once, actually quite a well written article, from the torygraph.
“No one is safe anymore. Morale is down the toilet." Was anyone ever "safe"?

https://www.telegraph.co.uk/business/2026/06/29/silicon-valleys-ai-jobs-purge-is-a-warning-for-the-world/

It's not just open-source developers, who are feeling the pinch.