Hacker News Reader

AMD is not competing enough with NVIDIA, so they are not a solution.

What I mean is that whenever NVIDIA removed features from their "consumer" GPUs in order to reduce production costs and increase profits, AMD immediately followed them, instead of attempting to offer GPUs that have something that NVIDIA does not have.

Intel at least tries to be a real competitor, e.g. by offering much, much better FP64 performance or by offering more memory.

If Intel's discrete GPUs disappear, there will be no competition in consumer GPUs, as AMD tries to compete only in "datacenter" GPUs. I have ancient AMD GPUs that I cannot upgrade to newer AMD GPUs, because the newer GPUs are worse, not better (for computational applications; I do not care about games), while Intel offers acceptable substitutes, due to excellent performance per $.

Moreover, NVIDIA also had excellent Linux driver support for more than 2 decades, not only for games, but also for professional graphics applications (i.e. much better OpenGL support than AMD) and for GPU computing applications (i.e. CUDA). AMD gets bonus points for open-source drivers and much more complete documentation, but the quality of their drivers has been typically significantly worse.

NVIDIA always had good support even for FreeBSD, where I had to buy discrete NVIDIA GPU cards for computers with AMD APUs that were not supported for any other OS except Windows and Linux.

AMD "consumer" GPUs are a great choice for those who are interested only in games, but not for those interested in any other GPU applications. AMD "datacenter" GPUs are good, but they are far too expensive to be worthwhile for small businesses or for individuals.

I've found the amdgpu Linux driver to be fairly buggy running dual monitors with my Radeon VII, and found things like the fTPM to be highly buggy on Threadripper 2k/x399 to the point that I had to add a dTPM. They never got things truly working properly with those more-niche products before they just.. kind of... stopped working on them. And of course ROCm is widely regarded to be a mess.

On the other hand, my Steam Deck has been exceedingly stable.

So I guess I would say: Buy AMD but understand that they don't have the resources to truly support all of their hardware on any platform, so they have to prioritize.

I seem to recall the Vega era as 'when I wouldn't buy a GPU because AMDs were just unstable' (and of course never closed source Nvidia).

Took me almost 5 min to drill through enough Wikipedia pages to find the Radeon VII string.

https://en.wikipedia.org/wiki/List_of_AMD_graphics_processin... https://en.wikipedia.org/wiki/Radeon_RX_Vega_series

Contrast that with the earlier R9 285 that I used for nearly 10 years until I was finally able to get a 9070XT that I'm very happy with. They were still refining support for that aged GCN 1.2 driver even today, even if things are a lower priority to backport.

Overall the ONLY things I'm unhappy about this GPU generation.

* Too damned expensive * Not enough VRAM (and no ECC off of workstation cards?) * Too hard for average consumers to just buy direct and cut out the scalpers

The only way I could get my hands on a card was to buy through a friend that lives within range of a Microcenter. The only true saints of computer hardware in the whole USA.

> What we need? Reliability. And Linux support

Both of which NVidia does a lot better in practice! I'm all for open-source in-tree drivers, but in practice, 15 years on, AMD is still buggy on Linux, whereas NVidia works well (not just on Linux but on FreeBSD too).

> I don’t judge whether implicit sync or explicit are better.

Maybe you should.

> Both of which NVidia does a lot better in practice!

Correction - if they care. And they don't care to do it on Linux, so you get them dragging feet for decades for something like Wayland support, PRIME, you name it.

Basically, the result is that in practice they offer abysmally bad support, otherwise they'd have upstream kernel drivers and no userspace blobs. Linux users should never buy Nvidia.

> And they don't care to do it on Linux

I don't understand what you're saying here. I've used NVidia on Linux and FreeBSD a lot. They work great.

If your argument is they don't implement some particular feature that matters to you, fair enough. But that's not an argument that they don't offer stability or Linux support. They do.

Taking very long to implement stuff is a perfect argument of bad support for the platform. Timely support isn't any less important than support in general.

Are you a product manager? Or do you just not see the irony on your comment?

Long term support means my thing that has been working great continues to work great. New feature implementation has nothing to do with that and is arguably directly against long term support.

And Nvidia seems justified in this since effectively no distro dropper X11 until Nvidia had support.

If you think taking decades is an acceptable rate while others do it in a timely manner it's your own problem. For any normal user it's completely unacceptable and is the opposite of great (add to it, that even after decades of dragging their feet they only offer half cooked support and still can't even sort out upstreaming their mess). Garbage support is what it is.

AMD is notorious for not having ROCM support on in production currently sold GPUs, and horrendous bugs that actually make using the devices unusable.

I use AMD gpus on linux, I generally regret not just buying an Nvidia GPU purely because of AMDs lacklustre support for compute use cases in general.

Intel is still too new in the dGPU market to trust and on top of that there is so much uncertainty about whether that entire product line will disappear.

So at this point the CUDA moat makes is a non issue, on top of that what works works and keeps working, whereas with AMD I constantly wonder whether something will randomly not work after an update.

A timeline of decades for “features” your biggest consumers don’t care about is a reasonable tradeoff, even more so if actually pushing those features would reduce stability.

That's exactly the point. Nvidia might care about industrial use cases, while they don't care about desktop Linux usage and their support is garbage in result.

I've been using Nvidia gpus exclusively on debian linux for the past 20 years, using the binary Nvidia drivers. Rock solid stability and excellent performance. I don't care for Wayland as I plan to stay on Xorg + Openbox for as long as I can.

Wayland support hasn't been an issue since GLX was depreciated for EGLStream. I think the Nvidia backend has been "functional" for ~3 years and nearly flawless for the past year or so.

Both Mutter and KWin have really good Nvidia Wayland sessions nowadays.

It got better, but my point is how long it took to get better. That's the indicator of how much they care about Linux use cases in general. Which is way below acceptable level - it's simply not their priority (which is also exacerbated by their hostile approach to upstreaming).

I.e. if anything new will need something implemented tomorrow, Nvidia will make their users wait another decade again. Which I consider an unacceptable level of support and something that flies in the face of those who claim that Nvidia supports Linux well.

Buying AMD (for graphics) has been the only ethical choive for a long time. We must support the underdogs. Since regulation has flown the coop, we must take respondibility ourselves to fight monopolies. The short term costs may be a bit higher but the long term payoff is the only option for our self-interest!

/ steps down from soap box /

> Conclusion: Buy AMD. Excellent Linux support with in-tree drivers.

Funnily, AMD's in-tree drivers are kind of a pain in the ass. For up to a year after a new GPU is released, you have to deal with using mesa and kernel packages from outside your distro.. While if you buy a brand new nVidia card, you just install the latest release of the proprietary drivers and it'll work.

Linux's driver model really is not kind to new hardware releases.

Of course, I still buy AMD because Nvidia's drivers really aren't very good. But that first half a year was not pleasant last time I got a relatively recently released (as in, released half a year earlier) AMD card.

Use a better distro that includes drivers for new hardware.

A lot of people want to use Ubuntu or Ubuntu-based distros.

I have since switched from Ubuntu to Fedora, maybe Fedora ships mesa and kernel updates within a week or two from release, I don't know. But being unable to use the preferred distro is a serious downside for many people.

> Excellent Linux support with in-tree drivers. For 15 years!

Linux support has been excellent on AMD for less than 15 years though. It got really good around 10 years ago, not before.

ATI/AMD open source linux support has been blowing hot and cold for over 25 years now.

They were one of the first to actually support open source drivers, with the r128 and original radeon (r100) drivers. Then went radio silence for the next few years, though the community used that as a baseline to support the next few generations (r100 to r500).

Then they reemerged with actually providing documentation for their Radeon HD series (r600 and r700), and some development resources but limited - and often at odds with the community-run equivalents at the time (lots of parallel development with things like the "radeonhd" driver and disagreements on how much they should rely on their "atombios" card firmware).

That "moderate" level of involvement continued for years, releasing documentation and some initial code for the GCN cards, but it felt like beyond the initial code drops most of the continuing work was more community-run.

Then only relatively recently (the last ~10 years) have they started putting actual engineering effort into things again, with AMDGPU and the majority of mesa changes now being paid for by AMD (or Valve, which is "AMD by proxy" really as you can guarantee every $ they spend on an engineer is $ less they pay to AMD).

So hopefully that's a trend you can actually rely on now, but I've been watching too long to think that can't change on a dime.

It is possible that at some point, maybe 15 years ago, AMD provided sufficient documentation to write drivers, but even 10 years ago, a lot of documentation was missing (without even mentioning that fact), which made trying to contribute rather frustrating. Not too bad, because as you said, they had a (smallish) number of employees working on the open drivers by then.

Agreed! This is great news for AMD and users.

Those who want to run Linux seriously will buy AMD. Intel will be slowly phased out, and this will reduce maintenance and increase the quality of anything that previously had to support both Intel and AMD.

However, if Microsoft or Apple scoop up AMD, all hell will break loose. I don’t think either would have interest in Linux support.

> Agreed! This is great news for AMD and users.

Less competition is NOT good news for AMD users. Their CPUs are already at lot less competitively priced now that they beat Intel for market share.

Oh boy that strikes a nerve with the "Video memory could be gone after Suspend/Resume". Countless hours lost trying to fix a combination of drivers and systemd hooks for my laptop to be able to suspend/hibernate and wake up back again without issues... Which makes it even more complicated when using Wayland.

I have been looking at high-end laptops with dedicated AMD Graphics chip, but can't find many... So I will probably go with AMD+NVidia with MUX switch, let's see how it goes... Unless someone else has other suggestions?

> Basically, video memory could be gone after Suspend/Resume, VT-Switch and so on.

This actually makes sense: for example, a new task has swapped out previous task's data, or host and guest are sharing the GPU and pushing each others data away. I don't understand why this is not a part of GPU-related standards.

As for solution, discarding all the GPU data after resume won't help? Or keeping the data in the system RAM.

Last I tried to file a bug for a crash in an AMD Windows driver I had to through an anonymous developer I found on Discord, and despite weeks of efforts writing and sharing test case they choose to ignore the bug report I the end. The developer even asked not to be named as he might face repercussions for trying to help out.

Excellent Linux support. Except for ROCm which is a big mess.

I once had an mini pc with Nvidia. I got it for Cuda dev. One day the support for it was dropped for it so I was unable to update my system without it messing things up. So regardless of Cuda I decided Nvidia is not for me.

However, doing research when buying a new pc, I've found that AMD kind of sucks too. ROCm isn't even supported on many of the systems i was looking into. Also, I've heard their Linux graphics drivers are poor.

So basically I just rock a potato with Intel integrated graphics now. GPUs cost too much to deal with that nonsense.

I would really disagree with AMD's Linux graphics drivers are poor. Only ROCm is.

In your case maybe, but not according to some of the comments here in this very thread and also in some forums and YouTube videos back when I'd last checked.

FWIW, my experience gaming/web browsing/coding on a 3070 with modern drivers has been fine. Mutter and KWin both have very good Wayland sessions if you're running the new (>550-series) drivers.