Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

There are very few positive signs Nvidia would be a good caretaker of ARM IP, continue to push new innovated and open designs and mass adoption.

Nvidia's entire business and philosophy regarding their chip designs are kind of antithetical to what ARM was achieving.



All you have to see is NVIDIA's trashy effort on Linux and locking-down what-would-otherwise-be-open hardware from being usable by open-source drivers...


It's funny because NVIDIA used to work well on Linux, way back in the day. Then they just decided to not care, then they decided to be hostile. Now they're backtracking but only because they are forced to. I thought the amount of ML would have made them backtrack, but it seems that it is more AMD and Steam.


The quality of Nvidia's linux drivers hasn't changed much. They are about as good as they were 10 years ago.

But everything else has changed around them. The quality of other drivers has improved, the linux ecosystem and what it expects out of drivers has changed, and the nvidia windows drivers have gotten features that the nvidia linux drivers don't.


Wasn't there a situation where the Linux ecosystem decided to use an API that Nvidia officially opposed and never wanted to support? Resulting in Wayland/KDE/GNOME (or similar) not running quite a lot of features on Nvidia? Something about Nvidia only supporting EGLstream while everybody else wanted GBM?

It's very foggy in my mind, so I might be misremembering details.


You have the details precisely correct.

Nvidia are finally moving to support GBM (iirc, it's because someone from KDE asked an nvidia engineer to help get something working with wayland+eglstreams, and he found out it was impossible, as the community had been saying for a while. Slight citation needed on that as I can't find a reference).

Technically they have support in their latest 495 drivers, but I can speak from experience when I say it's not seamless for a user. (I have put a lot of effort into getting it to work and it recently broke again).


Found a reference[0]. I'm not gonna pretend that "thing I read by some random person on a forum" is an unimpeachable source, but it's where I heard it I think.

[0] https://www.gamingonlinux.com/2021/10/nvidia-beta-4952905-ro...


I cannot agree with this more! I'm so frustrated, since I actually managed to get nvidia + wayland (sway) working on my laptop, and an update of some kind broke my external monitors this week (I haven't had time to roll everything back individually to figure out what yet).

It's an incredibly frustrating experience, and I really wonder if nvidia realises the long-term impact they're having by alienating the tech crowd (i.e. the ones that give recommendations to friends, and decide what hardware their companies will buy).


It is important to remember that Hacker News and Phoronix and Linux kernel developers are within a communication bubble (like any forum).

It may be because, in practice, NVIDIA looks at the size of us all... and yawns because we're tiny and because most tech reviewers still recommend them and most gamers still buy them. And they have a point - AMD has better Linux support, but NVIDIA's got CUDA (which has basically killed OpenCL and AMD's ROCm is less than a proof of concept in quality), DLSS, Raytracing, much better video encoders, a bunch of stuff that people are willing to put up with subpar Linux for.


And ARM had constantly updated Linux kernels and drivers from Qualcomm before didn’t they?


ARM works very, very closely with the Linux kernel. The issues with updated code in the kernel isn't because of ARM CPU IP.


ARM hasn't been much better than Nvidia for GPU code.


Which is why I said CPU IP. That being said, it's a little hard to blame them there, GPUs are a bit of a patent minefield.


ARM has helped found Linaro (linaro.org) ~10 years ago to fix the problem with mainline Linux integration for all the participating SoC vendors.

You should expect the following manufacturers to have good mainline support:

https://www.linaro.org/membership/

I am sure your mileage will vary between vendors (I am pretty sure Qualcomm was NOT one of the founding members, but it's there now), but this at least signals an investment of money and resources.


ARM was never good for open hardware designs.

It wasn’t better before what you think Nvidia will do either. Look at the state of Linux kernel updates on ARM chips.


The openness of the hardware designs has little to do with Arm - it's the SoC designers adding closed GPU drivers for example. Nvidia taking over Arm won't solve that issue.


You seem to be making a lot of bad faith arguments in this comment section.

Why are you defending this deal so ferociously?


What is in bad faith? It’s assumed that ARM support is good now. I’d like to see more Tegra type chips, an open ARM device with updates Linux drivers from the manufacturer. I’d love to be pointed to a better one if you have it. I think what nvidia does with Linux GPUs is awful, but it is very different how they treat their ARM processors.

Please point me to good open ARM processor manufacturers that uploads good Linux updates, I’ll be happy to be wrong that the Jetson really isn’t the best.


All the SoCs available from https://www.96boards.org/ should have good mainline Linux support.

"Good manufacturers" today contribute directly to the upstream kernel, and frequently through Linaro kernel trees first.


Would you recommend any of these boards over a Jetson? I’m glad these exist but most of the specs look awful, and I’ve tried rockchip in pine hardware and it sucks.


I don't have much experience with the latest and greatest, and it surely depends on the types of workloads you plan to put on them.

Rockchip has a number of differently performing chips (RK3399Pro seems to perform well in comparison with Jetson: https://www.cnx-software.com/2019/05/15/toybrick-rk3399pro-b..., https://www.seeedstudio.com/blog/2019/12/05/rk3399pro-vs-ras...). Pine only recently introduced a RK3399-based board (non-Pro, basically lacking an NPU) version, so you might want to go with them for availability reasons.


Which is why the parent said that it would be good for RISC-V processors, not for ARM processors.


Yeah, just look at what they did with GSync




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: