GGPUVet
Workload matcher

What GPU do I actually need?

Stop guessing from VRAM charts. Pick what you're running and get a concrete recommendation with the cheapest card that can actually do it — priced against today's used market.

$
FLUX.1 image inference
Min VRAM
12GB
Comfortable
24GB

FLUX.1 [dev] fp8 runs in ~12GB; bf16 + LoRAs and large batches want 24GB+.

Recommended pick

Cheapest card with enough VRAM for flux.1 image inference. 24GB · comfortable headroom.

All capable cards (15)

VRAM figures are practical guidance for typical settings/precision, not vendor benchmarks. Real usage varies with model, quantization, batch size, and context length.