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)
NVIDIA GeForce RTX 3090comfortable
Consumer crossover 24GB
$269
best price
NVIDIA RTX A4000tight
Workstation 16GB
$466
best price
NVIDIA RTX 4000 Adatight
Workstation 20GB
$600
best price
NVIDIA GeForce RTX 4090comfortable
Consumer crossover 24GB
$699
best price
NVIDIA RTX A4500tight
Workstation 20GB
$888
best price
NVIDIA GeForce RTX 5090comfortable
Consumer crossover 32GB
$969
best price
NVIDIA RTX 5000 Adacomfortable
Workstation 32GB
$1,399
best price
NVIDIA RTX A5000comfortable
Workstation 24GB
$1,400
best price
NVIDIA RTX 4500 Adacomfortable
Workstation 24GB
$1,925
best price
NVIDIA RTX A6000comfortable
Workstation 48GB
$3,300
best price
NVIDIA RTX 6000 Adacomfortable
Workstation 48GB
$3,700
best price
NVIDIA L40comfortable
Datacenter 48GB
$5,950
best price
NVIDIA A100comfortable
Datacenter 80GB
$6,296
best price
NVIDIA L40Scomfortable
Datacenter 48GB
$6,500
best price
NVIDIA H100comfortable
Datacenter 80GB
$6,750
best price
VRAM figures are practical guidance for typical settings/precision, not vendor benchmarks. Real usage varies with model, quantization, batch size, and context length.