GPUs for image generation
As of — SDXL at 16-bit: from $0.07 per GPU-hour on the NVIDIA GeForce RTX 3060 (Edgevana), 62 cards qualify; FLUX.1 [dev] at 16-bit: from $0.16 per GPU-hour on the NVIDIA V100 32GB (Geodd), 29 cards qualify. Rates are each card’s cheapest live on-demand rate card across the providers we track, read hourly, every one linked to the page it was read from.
An image model also keeps its weights in GPU memory at 2 bytes per parameter in 16-bit, and its text encoders and the decoder come on top. The parameter counts below are the models’ own.
SDXL at 16-bit
2.6 billion UNet parameters × 2 bytes = 5.2 GB, plus two text encoders and the decoder: a card with 12 GB or more.
| Card, memory | Cheapest on-demand | Provider, read | Median rate card | $ per GB-hour | Deploy |
|---|---|---|---|---|---|
| NVIDIA GeForce RTX 3060 · 12 GB | $0.07 |  Edgevanacaptured | — 1 provider | $0.006 | Deploy ↗ |
| NVIDIA TITAN Xp · 12 GB | $0.08 | Edgevana captured | — 1 provider | $0.006 | Deploy ↗ |
| NVIDIA Tesla P100 12GB · 12 GB | $0.11 | Edgevana captured | — 1 provider | $0.009 | Deploy ↗ |
| NVIDIA GeForce RTX 2060 12GB · 12 GB | $0.11 | Edgevana captured | — 1 provider | $0.009 | Deploy ↗ |
| NVIDIA RTX A4000 · 16 GB | $0.12 from |  GPU.aicaptured | $0.17 9 providers | $0.007 | Deploy ↗ |
| NVIDIA GeForce RTX 4070 SUPER · 12 GB | $0.14 | Edgevana captured | — 1 provider | $0.012 | Deploy ↗ |
| NVIDIA GeForce RTX 5060 Ti · 16 GB | $0.14 | Edgevana captured | — 1 provider | $0.009 | Deploy ↗ |
| NVIDIA GeForce RTX 4070 · 12 GB | $0.15 | Edgevana captured | — 1 provider | $0.013 | Deploy ↗ |
| NVIDIA GeForce RTX 4060 Ti · 16 GB | $0.15 | Edgevana captured | — 1 provider | $0.010 | Deploy ↗ |
| NVIDIA T4 · 16 GB | $0.15 | Edgevana captured | $0.53 13 providers | $0.010 | Deploy ↗ |
52 more cards qualify, dearer: NVIDIA V100 32GB $0.16, NVIDIA GeForce RTX 3090 $0.17, NVIDIA A2 $0.20, NVIDIA RTX 4000 Ada $0.20, NVIDIA V100 $0.21, NVIDIA GeForce RTX 5070 $0.23, NVIDIA GeForce RTX 5070 Ti $0.23, NVIDIA RTX 2000 Ada $0.24, NVIDIA GeForce RTX 4080 $0.24, NVIDIA GeForce RTX 3080 Ti $0.24, NVIDIA TITAN RTX $0.24, NVIDIA Quadro RTX 6000 $0.27, NVIDIA RTX A5000 $0.27, NVIDIA L4 $0.27, NVIDIA GeForce RTX 5080 $0.30, NVIDIA A30 $0.32, NVIDIA A10 $0.32, NVIDIA RTX A6000 $0.35, NVIDIA GeForce RTX 4080 SUPER $0.37, NVIDIA RTX 4090 $0.39, NVIDIA RTX PRO 4500 Blackwell $0.39, NVIDIA RTX 4000 SFF Ada $0.44, NVIDIA RTX 5000 Ada Generation $0.45, NVIDIA GeForce RTX 3090 Ti $0.46, NVIDIA L40 $0.53, NVIDIA RTX 5090 $0.54, NVIDIA RTX 4500 Ada Generation $0.54, NVIDIA A40 $0.59, NVIDIA GeForce RTX 4070 Ti SUPER $0.62, NVIDIA RTX 6000 Ada $0.65, NVIDIA RTX PRO 6000 Blackwell $0.65, NVIDIA RTX PRO 5000 Blackwell $0.71, NVIDIA L40S $0.72, NVIDIA A100 40GB $0.76, NVIDIA A100 80GB $0.89, NVIDIA A10G $1.00, AMD Instinct MI300X $1.71, NVIDIA H100 80GB (form factor not stated) $1.71, NVIDIA H100 SXM $1.79, NVIDIA RTX PRO 6000 Blackwell Max-Q $1.80, NVIDIA H100 PCIe $1.83, NVIDIA H200 $2.09, AMD Instinct MI325X $2.25, NVIDIA GH200 $2.29, NVIDIA H100 NVL $2.92, AMD Instinct MI355X $2.95, NVIDIA H200 NVL $3.29, NVIDIA B200 $3.69, AMD Instinct MI350X $4.00, NVIDIA B300 $4.89, NVIDIA GB200 NVL72 $6.94, NVIDIA GB300 NVL72 $10.74.
FLUX.1 [dev] at 16-bit
12 billion parameters × 2 bytes = 24 GB for the transformer alone, with its text encoders on top: a card with more than 24 GB holds it without offloading.
| Card, memory | Cheapest on-demand | Provider, read | Median rate card | $ per GB-hour | Deploy |
|---|---|---|---|---|---|
| NVIDIA V100 32GB · 32 GB | $0.16 |  Geoddcaptured | $0.76 8 providers | $0.005 | Deploy ↗ |
| NVIDIA RTX A6000 · 48 GB | $0.35 |  Thunder Computecaptured | $0.55 17 providers | $0.007 | Deploy ↗ |
| NVIDIA RTX PRO 4500 Blackwell · 32 GB | $0.39 | Edgevana captured | $0.82 6 providers | $0.012 | Deploy ↗ |
| NVIDIA RTX 5000 Ada Generation · 32 GB | $0.45 | Edgevana captured | $0.64 2 providers | $0.014 | Deploy ↗ |
| NVIDIA L40 · 48 GB | $0.53 | Edgevana captured | $0.90 10 providers | $0.011 | Deploy ↗ |
| NVIDIA RTX 5090 · 32 GB | $0.54 | Edgevana captured | $0.72 15 providers | $0.017 | Deploy ↗ |
| NVIDIA A40 · 48 GB | $0.59 from | GPU.ai captured | $0.64 6 providers | $0.012 | Deploy ↗ |
| NVIDIA RTX 6000 Ada · 48 GB | $0.65 from | GPU.ai captured | $1.06 11 providers | $0.014 | Deploy ↗ |
| NVIDIA RTX PRO 6000 Blackwell · 96 GB | $0.65 from |  Hostingercaptured | $2.20 35 providers | $0.007 | Deploy ↗ |
| NVIDIA RTX PRO 5000 Blackwell · 48 GB | $0.71 | Edgevana captured | $0.83 2 providers | $0.015 | Deploy ↗ |
19 more cards qualify, dearer: NVIDIA L40S $0.72, NVIDIA A100 40GB $0.76, NVIDIA A100 80GB $0.89, AMD Instinct MI300X $1.71, NVIDIA H100 80GB (form factor not stated) $1.71, NVIDIA H100 SXM $1.79, NVIDIA RTX PRO 6000 Blackwell Max-Q $1.80, NVIDIA H100 PCIe $1.83, NVIDIA H200 $2.09, AMD Instinct MI325X $2.25, NVIDIA GH200 $2.29, NVIDIA H100 NVL $2.92, AMD Instinct MI355X $2.95, NVIDIA H200 NVL $3.29, NVIDIA B200 $3.69, AMD Instinct MI350X $4.00, NVIDIA B300 $4.89, NVIDIA GB200 NVL72 $6.94, NVIDIA GB300 NVL72 $10.74.
How these figures are read
A card qualifies by its memory alone; speed, interconnect and software support are not ranked here. The rate is the card’s cheapest live on-demand rate card - each provider once, at its cheapest on-demand rate, community capacity and marketplace floors out, read within 48 hours - and the median is taken over the same rate cards, as on every card page. Memory figures are floors from the arithmetic above, not measurements: context length, batch size and the serving stack change them. Methodology.
Sources
- Podell et al., SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis (2023) - a 2.6B-parameter UNet
- Black Forest Labs, FLUX.1 [dev] model card - a 12 billion parameter rectified flow transformer
Other workloads: GPUs for LLM inference · GPUs for fine-tuning · GPUs for training · Cheapest GPU memory per hour.
Data updated: