graphic card recommendations appreciated (old hardware)

[linuux]

There are 2 1030 versions. One with DDR5 DRAM and the other with slower, DDR4 DRAM. Be careful which you buy. There's a big performance difference. I have a fanless model with DDR5. Again, I'd get an AMD GPU if Linux support were important. There's a famous photo showing what Linus thinks about nvidia that is easily found. nvidia support is much better on Ubuntu these days, if you run an LTS and use the ubuntu-drivers autoinstall command. I know LTS support was a goal, but cannot say anything about non-LTS Ubuntu releases and nvidia.

Are you having any trouble using your card on Ubuntu? I thought Ubuntu was criticized a little while ago for providing proprietary drivers (already) installed - because the Linux community supports FOSS so having them 'pre-installed' went against the ethics/philosophy etc. Maybe I didn't explain that accurately but you probably know what I'm talking about.

In Ubuntu, AFAIK, you only have to 'click' the proprietary driver in Additional Drivers to 'enable' its use. I'm not sure why people have so much trouble with it when installing. I realize that there are potential problems whenever you upgrade your Linux version i.e. upgrading the kernel. That's because the kernel modules/headers need to match up with the Nvidia modules/info (I don't know the exact terms/terminology but it's mostly a problem with version upgrades). When there's a mismatch, the drivers don't work with X(org) or whatever which results in a black screen. I have seen this when I used to install nvidia drivers 'the hard way' or manually.

I think, as long as you understand the dynamics and concepts of what is happening and what is going wrong (and have an online connection with another computer or phone - so you can use internet to look up 'fixes' and recovery methods), you can deal with the nvidia proprietary driver.

In saying all that, though, the AMD(GPU) driver sounds a lot simpler and since it's ingrained properly in the kernel, much easier to deal with. However, I think it also has a problem with kernel upgrades - if AMD doesn't have an updated driver ready or if there are changes, there's potential for it to cause problems too. I read that bleeding edge distros have a lot of changes and thus, it is possible that using the free AMD driver can lead to some of these 'out-of-sync' issues or there is a delay in AMD driver updates.

It's a pick your poison situation but the main reason for me to look at GT 1030 (DDR5) instead of the RX 550 - is because I am finding more used (cheaper) nvidia cards else I would pick an RX 550 (I still might but they are at least $60 more here).

[linuux]

Just for teh lolz I did a short search: Using a site for comparing market prices I found the cheapets 1030 for about 95€ while the cheapest 550 stated at around 120€.
I also tried to search on ebay - but as I don't like that platform for several reasons I not invested any more time after still in the 10€-15€ range full of old pci/agp cards, lots and lots of fans, brackets, anti-sagg-stands and other crap. Yes, I came accross ongoing auctions, one for a gtx670 (currently at around 12€, 3 days left) and a 1060 (about the same, but 6 days or so left) - and I'm sure someone with experience on the site is able to find the few good deals to save quite a lot ... but hey: I'm sending over my 640 for free - I don't need it anyway and it seems a perfect fit - even if it's just to get the card a couple more months before it finally ends up as e-waste anyway.

Oh yea, I had to learn THAT the hard way: Although off-topic, I explain why anyway: As I started to stream quite some time ago now I figured why not built a cheap 2nd system to offload the streaming to it. I already had a FM2 APU in mind - and designed the system around such. Little did I know back than that the elgato hd60 I chose (was the best option back then) is a bit of false advertising. Yes, it does have a h.264 asic on the card - BUT: it's only ever used with direct input-to-stream when using elgatos own software. And even if using that style it's only a 1-to-1 direct copy of the incoming hdmi stream (1080p / stereo audio) packed into rtmp with at most 8mbit/s - 10mbit/s of quality. No overlays, no alerts, no nothing - just directly streaming the hdmi input directly to the streaming service. If one want's anything more fancy or use any other streaming software than the elgato one the card only behaves as a very expensive paperweight offering an hdmi input - pointing a cheap webcam at your screen would get you far cheaper. One requires quite some hardware to use it - both a rather powerful cpu as well as a good gpu as encoder-accelerator.
I though any cheap modern gpu would do and so I ordered a 1030 - only to find out: it doesn'T offer any encoding acceleration (it ended up as the "system" gpu in kvm host). So I ended up ordering a 1650 - just as encoding accelerator for my stream system.
Fun fact: as my r9 290x died from overheating as I didn't noticed a failed fan in time at least I can use it to game and default back to the direct-stream with elgatos software ... I don't have a facecam anyway.

Yeah, that's the only negative for the 1030 as far as I can tell - the feature set doesn't include NVENC. The next card up that does is the 1050 ti. I would look for one of those but they are expensive and I can only find them used - $100 more than the 1030. But, it's more powerful than the RX 550 and I need a lp card anyway since I foolishly bought a sff computer. I was desperate though and someone was selling a sff for $100.

The other reason I need a discrete gpu is that the Intel igpu doesn't support 4k resolution - the sff only has display port so I need to use an adapter. Thus, Linux OS only supports 1080p res on my 4K tv and in Windows, I am stuck at 30 hz. A discrete gpu will solve all these issues. They aren't major but kinda annoying. I should have bought a tower but that would have been $80+ more.

I don't know what advantages you get with NVENC over AMD encoding but many people tend to choose the nvidia card, right? The RX 550 supports video encoding so it has more features over the GT 1030. There's only one LP choice for that card though. I can find normal, dual slot RX 550s (used and new) for various prices including as cheap as the used GT 1030s but they won't fit because they are normal height (although some are single slot and short in length). Note to everyone: don't choose sff computers unless you have measured it and know that the cards you want will fit!

[TheFu]

Whenever I've played with GPU encoding, the results haven't been great and always seemed fuzzy. I don't live-stream, so a no-brand $65 HDMI 1080p capture device off amz running handbrake on a Ryzen creates very nice, small, videos. For example:
5890333202 (5.5G) for the original "real-time" capture and
1173110950 (1.1G) for the handbrake encoded version. Both are h.264/AAC encoded, though I add a vorbis audio track to the 2nd version. The audio encoding is just stereo since the source stopped using DD 5.1 audio for DD+, which isn't supported by any of the cheapo options. To get **any** audio, I have to use stereo audio settings from the source. I have a few players that do not support h.265, so I don't use that vcodec.

I've not had issues with Intel iGPUs on 4K displays. Perhaps the wrong type of physical connection is being used? DisplayPort and VGA appear to work fine.
https://ark.intel.com/content/www/us/en/ark.html is the Intel Specs search page. My Intel G3258 from 2014 supports 2560x1600@60Hz over display port and 1920x1200@60Hz over VGA, but only 1080p over HDMI. The connection type matters.

My Core i5-8250 does 4096x2304@60Hz over displayport, but only 4096x2304@24Hz over hdmi.

Perhaps we are saying the same thing?

For playback, I have raspberry pi media players. I don't have a full computer next to the projector. We strive to keep the noise down and computers or Xbox systems are just too noisy. A raspberry pi is silent.

Glad I've had the 1030 GT 2.5 yrs now as it was still 2x the cost of what I really wanted for a GPU at $70. Mine is in a KVM host. I got a 10xx-series to extend the support period 1 more generation, in theory.

BTW, not everyone uses the GUI for package management. I don't have an "additional drivers" option on my systems, but the ubuntu-drivers autoinstall accomplishes the same thing and is much easier to post in these forums as an answer. That command produces output to help folks struggling if anything bad happens, unlike using a GUI.

[cryptearth]

BTW, not everyone uses the GUI for package management. I don't have an "additional drivers" option on my systems, but the ubuntu-drivers autoinstall accomplishes the same thing and is much easier to post in these forums as an answer. That command produces output to help folks struggling if anything bad happens, unlike using a GUI.

According to the ubuntu wiki there'Re several different versions of the nvidia driver all supporting different ranges of models. Does the auto-install also take this into account and only installs the highest useable driver that still supports the installed gpu?
For some reason I experienced a strange driver issue where the 1030 required a lower version but my 1650 required a higher one. For some odd reason both are incompatible with oneanother: The driver supporting the 1030 doesn't support the 1650 and vise-versa.
Unfortunately I'm not that fluent in linux to have install both drivers just with different names and then set the specific version for each device - but this shows that a multi-gpu system is not just limited by the back then "it has to be the very same card for them work together" but also of specific driver versions. Even one generation difference can break a system.

[TheFu]

I know that sudo ubuntu-drivers autoinstall works on my KVM host with a 1030 GT. Due to some other oddities, I run that as part of my weekly patch script. I don't allow automatic patching - or any patch nagging. On 18.04, nvidia-driver-460 is the correct driver for that card.

On another KVM host, nvidia-driver-390 is the correct driver package for NVIDIA GF108 [GeForce GT 430], but I haven't actually looked at the screen on that system in a few months. All management and use is through ssh. It is connected to a kvm-switch, so looking isn't THAT hard, but it doesn't run a GUI by default.

That would seem to say that sudo ubuntu-drivers autoinstall does the right thing some of the time, perhaps most of the time, but I really don't know. Whatever people do, don't go grab the driver from nvidia's website. Always, always, stick with drivers provided through Canonical's package management. It is safe to get information about the correct driver version from the nvidia website, just don't download drivers from there.

[cryptearth]

Thanks for that advice. Will stick to it even I'm used to goin to vendors sites and download drivers from there on windows.

[linuux]

Whenever I've played with GPU encoding, the results haven't been great and always seemed fuzzy. I don't live-stream, so a no-brand $65 HDMI 1080p capture device off amz running handbrake on a Ryzen creates very nice, small, videos. For example:
5890333202 (5.5G) for the original "real-time" capture and
1173110950 (1.1G) for the handbrake encoded version. Both are h.264/AAC encoded, though I add a vorbis audio track to the 2nd version. The audio encoding is just stereo since the source stopped using DD 5.1 audio for DD+, which isn't supported by any of the cheapo options. To get **any** audio, I have to use stereo audio settings from the source. I have a few players that do not support h.265, so I don't use that vcodec.

I've not had issues with Intel iGPUs on 4K displays. Perhaps the wrong type of physical connection is being used? DisplayPort and VGA appear to work fine.
https://ark.intel.com/content/www/us/en/ark.html is the Intel Specs search page. My Intel G3258 from 2014 supports 2560x1600@60Hz over display port and 1920x1200@60Hz over VGA, but only 1080p over HDMI. The connection type matters.

My Core i5-8250 does 4096x2304@60Hz over displayport, but only 4096x2304@24Hz over hdmi.

Perhaps we are saying the same thing?

For playback, I have raspberry pi media players. I don't have a full computer next to the projector. We strive to keep the noise down and computers or Xbox systems are just too noisy. A raspberry pi is silent.

Glad I've had the 1030 GT 2.5 yrs now as it was still 2x the cost of what I really wanted for a GPU at $70. Mine is in a KVM host. I got a 10xx-series to extend the support period 1 more generation, in theory.

BTW, not everyone uses the GUI for package management. I don't have an "additional drivers" option on my systems, but the ubuntu-drivers autoinstall accomplishes the same thing and is much easier to post in these forums as an answer. That command produces output to help folks struggling if anything bad happens, unlike using a GUI.

I don't know about that method. 'Additional Drivers' is probably the most common method. Others:

https://phoenixnap.com/kb/install-nvidia-drivers-ubuntu

I have a 4k tv with hdmi ports so an hdmi to hdmi connection is the preferred one. The sff computer only has display port and vga video ports. The dp to hdmi adapter can only get 1080p resolution in Linux. I suppose there are tweaks and configurations (CLI, probably) but I don't want to do that. I want to connect the computer and tv and leave it (PnP). A discrete video card will allow hdmi-to-hdmi. But, since I don't have one, I'm stuck with intel igpu and the limitations.

[cryptearth]

But, it's more powerful than the RX 550 and I need a lp card anyway since I foolishly bought a sff computer. I was desperate though and someone was selling a sff for $100.


I don't know what advantages you get with NVENC over AMD encoding but many people tend to choose the nvidia card, right? The RX 550 supports video encoding so it has more features over the GT 1030. There's only one LP choice for that card though. I can find normal, dual slot RX 550s (used and new) for various prices including as cheap as the used GT 1030s but they won't fit because they are normal height (although some are single slot and short in length). Note to everyone: don't choose sff computers unless you have measured it and know that the cards you want will fit!

Well - although I have a few cards around from both AMD and nVidia they're all too different to compare them fair to figure out if there're any diference in the hardware encoder blocks at all. So I can'T say anything about that point ...
But about the other ... limited to a not-full-atx-formfactor - I made the same mistake.
It's another reason why I original ordered the 1030: As I original used an asus a68hm-plus with the pci-e x16 slot and the x1 slot right next to each other but the x16 slot closer to the cpu I also required a single-slot card. Bonus: As it's all in a low-profile case (antec minuet 350) I was limited to a low-profile card. So the challenge was: Finding a decend enough single-slot low-profile card but with hardware encoder - the only card I was able to find which meet these specifics was some AMD Radeon Pro WX ... one of the early models - can't remember which exactly - the 2100 or 3100 or something like this. And this might had worked out - but as I had to order a card I'm not even sure would actually work the way I intended from outside of the EU (can't remember if it was from usa or china (doesn't matter - it was manufactured in china anyway)) the price tag of over 200€ made my finally opt-out of this idea.
Long story short: I found another mini-ATX board which had a slot in between - a biostar a68mde. This finally gave me the required space for using a dual-slot card - although still limited to low-profile. So I ended up with a 1650. It all worked out as nice day-trip to berlin (I live about 2h from berlin - so no big deal for a day-trip over the about 150km) into one of biggest pc-hardware stores in my local area (berlin really was the best option for me anyway as I combined it with a visit to my back then landlord) and as I had given my ASUS board and some random CPU (can'T even remember what it was) it turned out as a pretty much 1-to-1 exchange (I got about 5€+ out of the deal I spent for lunch).

If I had built it in a standard midi-tower I could had gotten away with some single-slot standard-height cards ... but going the small case route had some challenges for me.
Today, after the swap of the motherboard and added 1650 as encoder accelerator I'm happy with it ... plus: due to the 1650 it's quite a power system able to handle most games I play (as I use the 1650 right now after my big r9 290x died) - so when it gets back after a new main gpu it's available again for some 2-player LAN action.

[linuux]

Well - although I have a few cards around from both AMD and nVidia they're all too different to compare them fair to figure out if there're any diference in the hardware encoder blocks at all. So I can'T say anything about that point ...
But about the other ... limited to a not-full-atx-formfactor - I made the same mistake.
It's another reason why I original ordered the 1030: As I original used an asus a68hm-plus with the pci-e x16 slot and the x1 slot right next to each other but the x16 slot closer to the cpu I also required a single-slot card. Bonus: As it's all in a low-profile case (antec minuet 350) I was limited to a low-profile card. So the challenge was: Finding a decend enough single-slot low-profile card but with hardware encoder - the only card I was able to find which meet these specifics was some AMD Radeon Pro WX ... one of the early models - can't remember which exactly - the 2100 or 3100 or something like this. And this might had worked out - but as I had to order a card I'm not even sure would actually work the way I intended from outside of the EU (can't remember if it was from usa or china (doesn't matter - it was manufactured in china anyway)) the price tag of over 200€ made my finally opt-out of this idea.
Long story short: I found another mini-ATX board which had a slot in between - a biostar a68mde. This finally gave me the required space for using a dual-slot card - although still limited to low-profile. So I ended up with a 1650. It all worked out as nice day-trip to berlin (I live about 2h from berlin - so no big deal for a day-trip over the about 150km) into one of biggest pc-hardware stores in my local area (berlin really was the best option for me anyway as I combined it with a visit to my back then landlord) and as I had given my ASUS board and some random CPU (can'T even remember what it was) it turned out as a pretty much 1-to-1 exchange (I got about 5€+ out of the deal I spent for lunch).

If I had built it in a standard midi-tower I could had gotten away with some single-slot standard-height cards ... but going the small case route had some challenges for me.
Today, after the swap of the motherboard and added 1650 as encoder accelerator I'm happy with it ... plus: due to the 1650 it's quite a power system able to handle most games I play (as I use the 1650 right now after my big r9 290x died) - so when it gets back after a new main gpu it's available again for some 2-player LAN action.

At least, you were able to use a dual slot card.

I noticed all the gtx 1650 lp cards are dual slot. There's quite a few. That would be a nice option for me to have but the sff I have only has room for single slot, low profile so it's a double whammy of limitations. The gtx 1650 lp cards are really expensive still so it's a bit of a moot point but if it ever goes back to a reasonable price, I still can't use one as an option with the computer I have.

I guess Nvidia version 470 is out now? How is it with your card?

[ianwhite7]

ATI Radeon HD 2600 XT