Is this used GPU a good deal for local AI?
Used cards are the value sweet spot for inference, because the market prices gaming performance and local AI needs memory. That gap is the opportunity — and the risk.
The short answer
Computed from this tool’s default settings — card on offer and the rest as most people start. Change them below for your own case.
4 catalogued cards deliver a usable gigabyte more cheaply than $31.7/GB. Negotiate, or buy one of those instead.
Used GPU Value Checker
Price per usable gigabyte against every alternative at the same capability.
4 catalogued cards deliver a usable gigabyte more cheaply than $31.7/GB. Negotiate, or buy one of those instead.
| Device | Usable | Price | Per GB | |
|---|---|---|---|---|
| This offer | RTX 3090 | 22.1 GB | $700 | $31.7/GB |
| Alternative | Raspberry Pi 5 | 9.60 GB | $120 | $12.5/GB |
| Alternative | Arc B580 | 11.0 GB | $249 | $22.6/GB |
| Alternative | RTX 5060 Ti 16GB | 14.7 GB | $429 | $29.1/GB |
| Alternative | RTX 4060 Ti 16GB | 14.7 GB | $449 | $30.5/GB |
| Alternative | RX 9070 XT | 14.7 GB | $599 | $40.7/GB |
| Alternative | RX 7900 XTX | 22.1 GB | $999 | $45.2/GB |
What each setting changes
Every input moves the result for a reason. This is what each one does and where to find the value for your own machine.
| Setting | Default | What it changes |
|---|---|---|
| Card on offer | RTX 3090 · 24 GB | The machine the model runs on. Usable memory decides what fits and memory bandwidth decides how fast it runs, so this moves every figure below. |
| Asking price | 700 USD | Your own figure in USD, starting from 700. Change it to match what you actually run. |
| Model size | 8B | 6 options, from 3B to 405B. |
Real answers across model size
The same calculation run at a range of settings, with everything else left at its default. These are computed by the tool itself, not written by hand.
| Model size | Verdict | Price per GB | Usable memory | Runs a 3B |
|---|---|---|---|---|
| 3B | Overpriced | $31.7 | 22.1 GB | Yes |
| 8B | Overpriced | $31.7 | 22.1 GB | Yes |
| 14B | Overpriced | $31.7 | 22.1 GB | Yes |
| 32B | Good deal | $31.7 | 22.1 GB | Yes |
| 70B | Good deal | $31.7 | 22.1 GB | No |
5 catalogued cards deliver a usable gigabyte more cheaply than $31.7/GB. Negotiate, or buy one of those instead.
How this is calculated
No lookup tables and no invented constants. Here is the arithmetic, so you can check it against your own numbers.
A used card is worth what it does per dollar, not what it cost new. The comparison here is against every catalogued card that runs the same model, ranked on price per usable gigabyte.
Two things this cannot check: remaining life, and whether the card was mined on. Memory is what degrades under sustained load, and memory is exactly what you are buying it for — so test it under load before paying.
These are well-founded engineering estimates, not benchmark results. Your quantisation, runtime and context length all move the real number, and usable memory is an assumption rather than a specification. See the full methodology for every assumption behind these figures.
Who this is for
The three situations that bring people to this calculation.
Evaluating a listing
Check the asking price against every alternative.
Negotiating
Know what the card is worth for this workload.
Selling
Price a card fairly for AI buyers.
How to use this calculator
Four steps, no account, nothing leaves your browser.
Set card on offer
Start at the top of the panel. Every figure recalculates as you change it — there is no submit button, because watching the number move is the point.
Adjust the rest to match your setup
2 further settings: asking price, model size. Defaults are the common case, so change only what differs for you.
Read the headline, then the table
The large figure answers the question. The table underneath shows how the answer changes across nearby settings, which is usually where the decision actually gets made.
Check it against the method
The arithmetic is written out above. If a number looks wrong for your hardware, the assumptions are the first place to look — usable memory and quantisation are the two that vary most.
Is this used GPU a good deal for local AI: common questions
The questions people ask about this, answered without hedging.
Is buying a used GPU for AI safe?
Reasonably, with testing. Memory is what degrades under sustained load and memory is exactly what you are buying, so run a memory stress test and a long inference session before paying.
Are mined-on cards a problem?
Less than reputation suggests for the core, but mining runs memory hot for years. Test under sustained load rather than trusting a short benchmark.
Which used cards are worth buying for local AI?
Previous-generation cards with large frame buffers. They are priced on gaming performance while delivering capacity that newer mid-range cards do not have.
What should I check before paying?
Fan and thermal behaviour under load, memory errors under a stress test, the original packaging or purchase history if claimed, and that the card holds its clocks rather than throttling after a few minutes.
Related local AI tools
All 50 run on the same arithmetic, so answers across them agree.
More on hardware
Will it fit
Context
Speed
Quantisation
Runtime & errors
Cost
Searches this page answers
Different ways of asking the same question, all resolved above.
Now find the models that fit
Sizing is only half the problem. Model Radar takes your hardware and shows which models actually run on it, ranked by what they are good at — the same arithmetic as this page, applied to every model worth running.