Ask HN: What to Do with an Old GPU?

[jupiterelastica]

Hello all,

for many months my custom bulit gaming PC from 2014 has been sitting around my appartment, not finding any use. I am working mostly on my laptop from work nowadays and I stopped playing games for the most part. The GPU is a Nvidia GTX 780.

Recently, I started using machine learning tools on my work laptop and realized I have a reasonably strong GPU sitting around - or so I thought. Looking up the requirements for pytorch I found that the GTX780's CUDA version has long since lost support and now I wonder what to do with the GPU, or even the whole system, as gaming is not important to me anymore and for an ML box it's too old apparently.

I know I could sell it, but I'm wondering if anybody might have ideas on what such old hardware could still be used for?

Cheers, - V

Edit: spelling

[idonotknowwhy]

I sold a 750ti for $70 recently. Just ebay it.

Also, I think that's the most powerful gpu with analogue video output. Some Vga CRT geeks would want it, to avoid the additional lag using an hdmi/usbc/dp to vga adapter.

[tluyben2]

Donate it. Many many people on this planet will be happy with it. Not everyone is on the latest AAA games, especially in poorer communities and on a discrete gpu from 2014 you can have enough fun. Think a lot of people even in the poorer part of your city would bite off your hand for it.

[jupiterelastica → tluyben2]

Good point, I haven't considered this. Thanks!

[guard0g]

Use it for a multi-monitor setup? I've got 8 monitors connected to my desktop thru two old GPUs.

[angrais → guard0g]

What do you use 8 monitors for?

[krnlpnc → angrais]

To look at 8 HN submissions at the same time

[technobabbler → krnlpnc]

7 max, and a dashboard on the 8th to tell you what you're looking at on the other 7

[guard0g → angrais]

Monitors: 1 - slack, teams, discord 2 - email, zoom 3 - markets monitor 4,5,6 - streaming news, tv 7, 8 - coding/work

I have the monitors mounted, stacked 2x2 on each side and tilted forming a semi-circle.

[fancyfredbot]

I'd suggest installing an old version of CUDA and pytorch using it for machine learning anyway? It's still possible to learn the principles with an older GPU like this, although you'll get less than half the performance of a bottom of the range Ampere.

[epirogov → fancyfredbot]

Every time I tried this way, because of no budget, I got obsolete apis and libs include old glibc, old linux versions to return to GPU age, hours of compiling from the sources to run on new hardware and kernel. As a reward you will get a lot of internal knowledge but rare chanse to see CUDA and pytorch works as expected.

[jupiterelastica → fancyfredbot]

I could, yes. Yet it makes more sense to run what I need on my work laptop or if I ever really need it buy a could GPU on demand, since I'm not fiddling with the lower layers of ML tools, but learning the basics.

[Wowfunhappy]

This GPU is quite close to the most powerful nVidia GPU you can use in macOS 10.14 and above. It's beaten only by the GTX 780 6GB (since you didn't specify, I assume yours is a 3GB model), the GTX 780 Ti, and the original Titan. (And I'm not actually sure about the Titan.)

For the same reason, it also happens to be close to the most powerful GPU you can use with OS X 10.9, period. That's my favorite OS, so that makes your card special. But I'm probably the only person in the world who cares about that, and I already have a GTX 780 6GB and a 780 Ti. :)

[khoobid_shoma]

What happens to the next GPUs you buy?

I have wasted a lot of money buying nVidia GPUs. They used to be quickly outdated for different reasons (e.g. unsupported by the next versions of CUDA). I decided to stop buying new ones until I really really need a modern GPU. Now, that's around 7 years, I am happily away from nVidia GPUs. If needed, I would get a GPU instance on cloud. The same issue is with Apple iPads (my iPad One still looks new but software upgrade is not supported anymore).

[jupiterelastica → khoobid_shoma]

It is the first and only GPU I ever bought. Since it is the most expensive part in a PC, I have also kept to the "do I really really need it?"-mantra and have never upgraded it.

With this PC I will not be doing anything anymore, I believe. I just don't have a use for it. So I guess I will follow the advice of most comments here and just sell it after all :)

[smoldesu]

Sell it? There will never be a better time to get rid of garbage GPUs at unreasonable prices. Dump it on Ebay, pop the popcorn, and watch the desperate bidders roll in!

[gtirloni → smoldesu]

I really doubt it. People that want to play recent games will be looking for 900-series at least and mining on a 780 is nonexistent.

[6yyyyyy → gtirloni]

https://www.ebay.com/sch/i.html?_from=R40&_trksid=p2334524.m...

Looks like you can fetch $120 for a working one.

[speedgoose]

To be honest, you can put it in the electronic items bin at your local recycling centre.

[karmakaze]

Donate it to someone who wants to play either non-AAA games or retro ones. StarCraft II plays well on basically any discrete GPU. Age of Empires may work well too (I haven't tried but graphics don't look too demanding.)

[goldname]

Maybe mine crypto in the winter if there's some software compatible with it.

[PaulHoule]

So far a GPGPU is concerned that is not a ‘powerful’ GPU at all.

[Procedural]

Donate it to a kid in Yemen.

[kimown]

Hackintosh GTX780