Regarding NVIDIA

The Water Cooler
1

This topic is somewhat off-topic, so I decided to place it in the 'Water Cooler" section. It may not be as friendly as the category’s description suggests it should be, however, as you probably know, real office water coolers are not free from various gossip.

NVIDIA is a large company. They’re effectively multinational, in fact. They distribute their own proprietary GPU hardware, which is tied to a specific AI/ML stack. This stack is their primary selling point, with their GPUs being the gateway.

I don’t have a problem with this in itself, and I don’t believe that most people do. However, it does become a problem when the company attempts to shove this stack down people’s throats via hardware enablement initiatives. Enablement seems to be a recently coined word, since even this forum’s autocorrect doesn’t recognize it. Now, I’m not saying NVIDIA is doing that, however, the way that enablement has been communicated across the FOSS community makes it seem so.

This may rightfully make various FOSS communities, including some within Fedora, very suspicious that something sinister may be happening. They will also implicate NVIDIA, no matter if NVIDIA is involved or not, since it involves NVIDIA’s drivers.

There is a coordinated push for hardware enablement, since there is a CentOS SIG centered around it, and Fedora leadership has brought it up as an initiative worth implementing, whether directly, or indirectly via Remixes:

https://docs.centos.org/centos-accelerated-infrastructure-sig/

Enablement language has also been used by at least one journalist, to describe IBM’s virtualization initiatives:

Since this initiative seems to be NVIDIA-specific, this enablement initiative will not bode well with FOSS communities. I haven’t brought about any conclusive evidence, however, I believe it is important to bring attention to this matter, and make way for others to comment their objections/concerns.

2

I hear you, and I basically agree. But Fedora is a community, and communities can be varied and have different aspects while still being a community. More that one person can get what they want at the same time.
I’ve only seen you post about anti-AI and pro open topics, with which I am actually on side with - but I’m also on side with making everyone in the community have a voice, and even on side with Red Hat and Jef having a voice and a say. So I would find it easier to have a longer discussion with you if I saw you work on other little topics in the the community rather than continually push one agenda.
You could start helping out in Ask Fedora, our help desk. You could package some software, or anything you want really. Doing the technical work to keep Fedora open and free is what will carry us into the next generation.

1 Like
3

With the influx of new users switching away from Windows there is a need to educate many of them in effective ways to collect details and report issues to appropriate upstream or distro forums.

There is also a need to shape the evolution of developing technologies. In particular, AI is not going away. It has proven useful at finding bugs, but without careful management, can easily overwhelm the people doing the actual work of triaging and fixing bugs while at the same time providing bad actors with unprecedented quantities of new exploits. Well intentioned users need guidance in reporting bugs and encouragement to learn what they need to understand the information an AI provides and work towards a suitable fix.

4

I could not agree more. Dealing one on one with new users and helping them learn about the values of Open Source is super important work.

1 Like
5

I get that, maybe you don’t trust me yet. However, some say that the people on the other side also have their agenda, which may compromise the future of Fedora. In the case of NVIDIA, their stack is being pushed heavily.

There is something to be said about FESCo having an enablement initiative, in addition to the other AI initiatives proposed to them with the Fedora AI Developer Desktop initiative. Who exactly was involved in this process, and why? Should this have been proposed at all, and discussed to the degree it has been? It could turn out that the reason is benevolent, but I’m not sure as of yet.

Dependency chains are also a huge problem, that no software engineer has opted to solve as of yet. I can solve this problem if given the time, but it should have been solved yesterday. Developing software from near scratch is hard, but it always pays off in the end, since now, the software would no longer require a chain of dependencies, it would only depend on the libraries already packaged in Fedora by default (GLIBC, maybe add Mesa3D and libX11/libwayland if the application is a GUI).

Clearly, the initiatives/proposals over the years that have addressed this problem, never solved the root problem. I don’t know if that’s due to a lack of developers, or a lack of understanding in regards to the root problem, but whatever it may be, this isn’t good.

Python isn’t a bad language in my opinion, however, the amount of libraries/packages that are associated with Python seem to be enormous, and almost unmanageable. If I just search python on https://packages.fedoraproject.org/, there will be ~3K results. I really don’t understand how you guys do it. If I were a Fedora leader, I would simply tell everyone to not bother.

I don’t think this number of python packages is necessary. It would lessen the load on packagers, if the root problem is solved. Maybe some of the packagers will be able to take on new roles, and in turn help out the community more than they already have been?

I am certainly pushing an agenda or two, I’ll admit. But I believe they’re important ones to pay heed to.

6

Maybe surprisingly, the leader does not always get their way. Those Python packages are maintained because someone uses them.
We give in to the leaders for little things, and keep doing our own thing, that’s Linux.

I do welcome your agenda. But may I suggest that short, focussed posts can be far more efficacious than arguments or repeating points. I want the same things as you, but I think our working methods are different. Also the suggestion of making little and tangible improvements to Fedora will, in a rather short time, ensure your trust and integration into the community.

The problem with Nvidia is that they have sold so much damn hardware. And now people jump over to Fedora and Linux (because we are better) but they still have their hardware. And the hardware they sold, for Windoze, is really good kit. Fast. Cheap. It works. It’s just a shame they are a black box.

7

Speaking of black boxes, look at these:

The difference speaks for itself. Take a look at those internal clock speeds Intel gives away all willy nilly :P. Intel also describes the internal paths far more than NVIDIA does.

Maybe the difference isn’t as pronounced in the newer ones:

– but Intel still has a history of being more open with their hardware. I think they should be commended for this, since I hope they continue doing this well into the future. MBAs and CEOs may not value it, but many lower level software engineers certainly do. It allows them to better understand the hardware they’re working with.

Intel’s newer whitepaper is far more terse, and straight to the point, when compared with NVIDIA’s newer whitepaper

The strong point of Intel’s new whitepaper, is that it details the number of operations per clock cycle each EU can execute. With this statistic, when combined with the ~560GB memory bandwidth figure given, it gives an engineer a solid ground to estimate roughly what the performance should be for any compute workload he may give the GPU. Of course, real performance will be ~80% of the maximum estimate, but that’s just how these chips are.

If I have one complaint, it is that Intel doesn’t have bandwith/per-clock figures for their L1/SLM memory.

NVIDIA’s paper just has a lot of fluff in my opinion. I believe it is better to let the chips’ specifications speak for themselves, rather than describe in dozens of pages what they can do for you. Interpreting these statistics for the general population is the Journalist’s job, in my opinion.

I’ll keep posting, no matter if people agree with me or not. My primary methods involve shifting the focus onto what the root problems actually are, rather than discussing the symptoms that they bring about. I hope you understand. :slight_smile:

8

Both AMD and Intel found that it’s profitable to have fully open drivers.

Chip vendors that only had closed linux drivers fail to get design wins.
These wins are usually $1M+ deals in the commercial world.