Solved An available graphics card that's known to work, for < $300?

Yes it does!

Intel i7-10510U (8) @ 2.300GHz
CometLake-U GT2 [UHD Graphics]

Interesting, mine is a bit different... I forget what exactly, it doesn't show up anymore now that I have an nvidia in there. But I'm pretty certain it had a different letter suffix after CometLake, and it's UHD 530 Graphics. So maybe that slightly different version means it doesn't get recognized? I will get that info when I swap in the new 730 card and file a PR.
 
Well, and my intention here was specifically to preempt inevitable GT 710/730 recommendations. It's EOL. It's effectively a brick by this point. Buying one is stupid even if it's cheap. Suggesting that card is irresponsible. (By the way, if somebody wonders why I'm not recommending GT 1030 in addition to or instead of GTX 1630, the newly open-sourced kernel modules do not support anything earlier than Turing.)

so I ordered a ZOTAC GeForce GT 730.
What the hell?
 
Well, and my intention here was specifically to preempt inevitable GT 710/730 recommendations. It's EOL. It's effectively a brick by this point. Buying one is stupid even if it's cheap. Suggesting that card is irresponsible.
I did not recommend that card - I just wrote what I was using and *explicitly* stated WHY (PCIe x1). I also mentioned that it is already only supported by the legacy driver.
 
What the hell?

I don’t know anything about graphics cards, if you can’t tell.

I saw a fanless card, which I want to minimize noise, and saw that it had native drivers. So, it sounded great to me.

You mention EOL - so I guess the risk is that it won’t run on 14+? Although there’s a chance that my iGPU works by then, so the 730 really only needs to get me to that point.

Can you recommend a currently supported passively-cooled card? I’m only using this for emacs and web browser (although with hundreds of tabs).
 
There’s the Palit GTX 1650 KalmX which looks interesting, but $450ish and doesn’t appear to be readily available.

Asus sells a fanless 1030, that nvidia shows as being on the current driver. So I guess that’s the one?

edit: okay yeah, did some comparisons, and the 1030 performs way better than the 730. Plus it’s in the current drivers.
 
There’s the Palit GTX 1650 KalmX which looks interesting, but $450ish and doesn’t appear to be readily available.

Asus sells a fanless 1030, that nvidia shows as being on the current driver. So I guess that’s the one?

edit: okay yeah, did some comparisons, and the 1030 performs way better than the 730. Plus it’s in the current drivers.
IMHO, you should make a strategic decision - AMD or Nvidia. Personally, I was a long-term Nvidia user until the support to my otherwise good card stopped. Nvidia drivers are closed-source with just a lightweight wrapper. In the other hand, AMD drivers are open source.
Partially because of worldview reasons I moved to AMD cards and not regretted. I have RX550 and RX570 cards and work perfectly. Just a bit old.
I can also see that local retailers here are selling brand new Asus RX50 cards for € 129,00 (including VAT). (ASUS AMD Radeon RX 550 4 GB 128 Bit PCIE 3.0 16x GDDR5 Memory 6000 MHz GPU 1183 MHz Dual Slot Fansink 1xDVI 1xHDMI) I think this a pretty good choice...
 
There’s the Palit GTX 1650 KalmX which looks interesting, but $450ish and doesn’t appear to be readily available.

Asus sells a fanless 1030, that nvidia shows as being on the current driver. So I guess that’s the one?

edit: okay yeah, did some comparisons, and the 1030 performs way better than the 730. Plus it’s in the current drivers.
Is that really a concern for you? Outside of games modern GPU fans are not particularly loud unless they are defective. (For example, the fans on my card spin at 1020 RPM when there is no load.) Some comparatively expensive cards don't even spin their fans at all in that scenario.
 
Is that really a concern for you? Outside of games modern GPU fans are not particularly loud unless they are defective. (For example, the fans on my card spin at 1020 RPM when there is no load.) Some comparatively expensive cards don't even spin their fans at all in that scenario.
Okay, it usually goes without saying, but let me state this explicitly: the fans on the modern card should be 8-10 cm in diameter and PWM-controlled. Don't buy cards having a single small fan with only two wires going to it.
 
I did get a CPU with integrated graphics - Intel 10900k. It turns out that, despite my best research effort, it is not yet supported by i915kms.

I really don’t care about a discrete graphics card at all. But, I’m on my second motherboard, have (re-)assembled this machine half a dozen times, have a third graphics card on the way (nvidia this time, fingers crossed!).

The time I’ve spent researching, and assembling, and testing, has been way more costly than the components. So at this point, I just want the fastest path to booting a desktop environment. Which means adding a working graphics card.

Knowing what I know now, I would have gone for a beefier AMD processor. But, pulling out this mobo and CPU and starting all over again is an absolute last resort.

It sounds somehow like a config matter. Next to the i915kms there must be other options. Sure you checked the Handbook on x11 config?
 
IMHO, you should make a strategic decision - AMD or Nvidia. Personally, I was a long-term Nvidia user until the support to my otherwise good card stopped. Nvidia drivers are closed-source with just a lightweight wrapper. In the other hand, AMD drivers are open source.

I did, and my decision is Nvidia. Here’s how I arrived at it:

- I wanted a good processor with an iGPU. I knew that the latest generations don’t have graphics support. I read through the source code of the various drm-kmod ports, linux kernel announcements, lots of forum and list posts. I determined that 10th gen graphics are supported - but alas, the 10900k appears to be an outlier.

- I began my search for discrete graphics. Having an open source preference, I immediately dismissed Nvidia and focused on AMD. That was a confusing mess of correlating model names with NAVI versions with colorful-fish code names. Again I read the ports’ sources and read lots of forum posts. I saw that dimgrey-cavefish is in the amd-firmware-kmod port, and thought I was good to go. It didn’t work, and a developer informed me that while the firmware is included in the port, it is not yet supported. I guess it’s included in preparation for a forthcoming release? I don’t know.

- Frustrated, I posted this thread. I figured that since my 30+ hours of researching had yielded nothing, I might as well ask what other people use. Then I can just pick one, knowing that it actually works. Here I was informed that Nvidia has a page where you can input the model, and it will tell you what drivers you can use. One of those versions is included in the ports tree, so just find a model that runs with that driver, and you’re good to go.

In short, I spent 30+ hours researching Intel iGPU and AMD discrete GPU, and didn’t get a working card. I spent 5 minutes looking at Nvidia’s driver page, installed the port, took about 10 minutes to read through their instructions, and got a desktop.

I’m all for open source, all other things being equal. But they are not equal here. AMD produces open source drivers for Linux, not FreeBSD. It is through (what appears to be immense) effort on the part of volunteers, that some of those drivers work on FreeBSD. It is unclear, to me at least, which models are actually supported.

Nvidia develops native drivers, clearly communicates the models that run them, and provides instructions on how to use them. They support FreeBSD directly. As someone enthusiast about FreeBSD, and disappointed by the low adoption rate in the development world relative to Linux and MacOS, I am very happy to support a company that directly supports my OS of choice.

Related: I think the single biggest barrier to entry for developers using FreeBSD is how difficult it is to procure hardware that will work, especially laptops. The conventional wisdom is “a ThinkPad from a couple generations back” - but even then you still have to research the graphics card and network card and hope it all works. This is why so many developers use Macs - you can go to the Apple store, buy the top spec MBP, plug it in and get to work.

The other developers on my team are getting more interested in FreeBSD because of the things they see me doing with it. But the only reliable way I know of to get a machine is to boot a cloud VM and SSH to it. I can’t (yet) give my boss a link to a laptop and say “buy one of these for everyone on the team and we’re good to go.”

Is that really a concern for you?

Yes, I turned the machine on and the noise difference was immediately noticeable. My GPU needs are minimal, so if I can get them met with 0dB vs 25+ dB then that’s a big win. 25 dB is about a whisper, but I wouldn’t want someone standing next to me whispering all day long either.
 
Yes, I turned the machine on and the noise difference was immediately noticeable. My GPU needs are minimal, so if I can get them met with 0dB vs 25+ dB then that’s a big win. 25 dB is about a whisper, but I wouldn’t want someone standing next to me whispering all day long either.
Lucky you. I won't ever be in that situation unless I kill my neighbors, their neighbors, the neighbors of their neighbors and especially the local muni kindergarten personnel (which isn't a bad idea at all, but it's kind of frowned upon) — 25 dB is way lower than the normal city noise floor.
 
Last edited:
patmaddox
So the thing about nvidia is that the hardware is pretty cutting edge, but their proprietary drivers tend to tick off the free software crowd, so nvdia is a swear word in some circles. I don't much care if the driver is closed as long as it works as expected. My only complaint right now is the whole lack of support for CUDA/GPU programming with nvidia on freeBSD. Well, that and their non-windows, non-OEM support channels really do suck.
 
patmaddox
So the thing about nvidia is that the hardware is pretty cutting edge, but their proprietary drivers tend to tick off the free software crowd, so nvdia is a swear word in some circles. I don't much care if the driver is closed as long as it works as expected. My only complaint right now is the whole lack of support for CUDA/GPU programming with nvidia on freeBSD. Well, that and their non-windows, non-OEM support channels really do suck.

Well, try using the existing, working drivers on -current? And it can be difficult to find out which kernel versions, exactly, are supported by any given NVidia binary package. Big problem for Linux, too.

As for support, it went to hell after the nvnews.com forum was given up in favor of NVidia's own channels.
 
Back
Top