On: The GPU Sadness Index: Tracking eBay Pricing

[segmondy]

I want to experiment with deep learning by training on the GPU, but the price even for used cards just makes me put it off. :-O. If it doesn't come down soon tho, I might cave and buy one.

[dgellow → segmondy]

Have you considered cloud services? For example you have GCloud compute engine: https://cloud.google.com/compute/gpus-pricing

[indiv0 → dgellow]

The GCE free tier is nice but ran out for me after just about a month of continuous use of one (cheap) GPU. Maybe some people have deeper pockets than I do but I can’t afford to pay $300 CAD/month on a single cloud GPU for hobbyist purposes.

[Something1234 → indiv0]

Kaggle seems to offer a lot of free GPU services if you want to have your notebooks be public.

[IdiocyInAction → dgellow]

vast.ai is also an option, which is cheaper (though also not as nice as GCP)

[cbozeman → segmondy]

Well how much VRAM do you feel you need for your projects? Because there's a lot of 16 GB cards out there that are still perfectly serviceable for that.

Vega Frontier Editions (16 GB HBM2) are going for around $600-700 on eBay and Craigslist, also Facebook Marketplace in some areas.

Radeon VIIs (16 GB HBM2) are going for around $650-1000 on the same platforms.

You can easily put two of those into a machine and have a pretty impressive deep learning rig.

[segmondy → cbozeman]

Thanks for the suggestion, will check those out!

[cbozeman → segmondy]

My pleasure and happy hunting.

Here's the original article that reminded me of this. Its four years old, but I still feel its a worthwhile read: https://medium.com/intuitionmachine/building-a-50-teraflops-...