Datasheet for local AI49 models98 GPUsData read 2026-09-25
No adsNo tracking
Check my GPU →

RTX 3090 24 GB for local AI

Which image and video models run on the RTX 3090 24 GB, which file to download for each, and how much VRAM they need.

NVIDIA24 GB GDDR6X · 384-bit · 936 GB/s · AmpereData 2026-09-25
VRAM24 GB
MemoryGDDR6X
Bus384-bit
Bandwidth936 GB/s
ArchitectureAmpere
Launched2020-09
Launch price$1,499
Runs well43 of 49

Specs: www.nvidia.com · launch: en.wikipedia.org

01What runs on it

ModelVerdictBest fileSizeNeeded
image models
FLUX.1 [dev]Runs wellQ8_012.7 GB15.0 GB
FLUX.1 [schnell]Runs wellQ8_012.7 GB15.0 GB
FLUX.1 Kontext [dev]Runs wellQ8_012.7 GB15.3 GB
FLUX.1 Krea [dev]Runs wellQ8_012.7 GB15.0 GB
FLUX.1 Fill [dev]Runs wellQ8_012.7 GB15.3 GB
FLUX.2 [dev]RunsQ4_K_M20.1 GB23.4 GB
FLUX.2 [klein] 9BRuns well16-bit18.2 GB20.5 GB
FLUX.2 [klein] 4BRuns well16-bit7.8 GB9.6 GB
Krea 2 (Turbo)Runs wellQ8_013.7 GB16.3 GB
Qwen-ImageRuns wellFP820.4 GB23.2 GB
Qwen-Image-Edit (2511)Runs wellFP820.5 GB23.5 GB
Qwen-Image 2.1Runs well16-bit14.2 GB16.5 GB
Z-Image TurboRuns well16-bit12.3 GB14.3 GB
Z-Image (base)Runs well16-bit12.3 GB14.3 GB
Ideogram 4Runs wellQ8_010.1 GB22.6 GB
Boogu-Image (Turbo)Runs well16-bit20.6 GB23.2 GB
ERNIE-Image (Turbo)Runs well16-bit16.1 GB18.4 GB
HiDream-O1-ImageRuns well16-bit16.4 GB19.2 GB
Mage-Flow (Microsoft)Runs well16-bit8.2 GB10.0 GB
Lumina Image 2.0Runs well16-bit5.2 GB7.0 GB
HiDream-I1 (Full)Runs wellQ8_018.7 GB21.5 GB
HiDream-I1 (Dev)Runs wellQ8_018.7 GB21.5 GB
Stable Diffusion 3.5 LargeRuns well16-bit16.5 GB18.8 GB
Stable Diffusion 3.5 MediumRuns well16-bit5.1 GB6.9 GB
Chroma1-HDRuns well16-bit17.8 GB20.1 GB
SDXL 1.0Runs well16-bit6.9 GB7.1 GB
Illustrious XL / Pony (SDXL anime)Runs well16-bit6.9 GB7.1 GB
Stable Diffusion 1.5Runs well16-bit2.1 GB3.3 GB
HunyuanImage 2.1Runs wellQ8_019.8 GB23.1 GB
video models
Wan 2.1 T2V 14BRuns wellQ8_015.9 GB20.2 GB
Wan 2.1 T2V 1.3BRuns well16-bit2.8 GB5.6 GB
Wan 2.1 I2V 14B 480PRuns wellQ8_018.1 GB22.4 GB
Wan 2.1 I2V 14B 720PRuns wellFP816.4 GB23.2 GB
Wan 2.1 VACE 14BRuns wellQ8_018.7 GB23.5 GB
Wan 2.2 T2V A14BRuns wellQ8_015.4 GB19.7 GB
Wan 2.2 I2V A14BRuns wellQ8_015.4 GB19.7 GB
Wan 2.2 TI2V 5BRuns well16-bit10.0 GB13.8 GB
Wan 2.2 Animate 14BRuns wellFP817.3 GB22.6 GB
Wan Animate 2 (14B)Runs wellQ8_018.1 GB23.4 GB
Wan 2.2 S2V 14BRuns wellFP816.4 GB21.2 GB
SCAIL-2 (character animation)Runs wellQ8_018.1 GB23.4 GB
HunyuanVideo (13B, original)Runs wellQ8_014.0 GB18.3 GB
LTX-Video 13B (0.9.8)Runs wellQ8_014.0 GB18.3 GB
HunyuanVideo 1.5Runs well16-bit16.7 GB21.0 GB
LTX-2 (19B)RunsQ6_K16.0 GB20.8 GB
LTX-2.3 (22B)RunsQ6_K17.8 GB22.6 GB
LTX-2.5 (22B)RunsQ6_K18.7 GB23.5 GB
MiniMax H3 (33B)TightQ3_K_M15.6 GB21.4 GB
MiniMax H3 PrunedRunsQ6_K16.7 GB22.5 GB

Calculated from real file sizes plus working memory. How the numbers work.

02Good to know

The RTX 3090 24 GB (Ampere) has no FP8 compute. FP8 files still load and save memory, but ComfyUI converts them back before the maths, so there is no speed gain — Q8_0 GGUF is the closer-to-original 8-bit choice. ComfyUI can compute INT8 files natively on NVIDIA, which is worth a try where a model offers one.

03Measured and reported results

LabelModelSetupResultPeak VRAMDateSource
reportedFLUX.1 devfp8 (ComfyUI template)
“Nvidia 3090: 26s”
ComfyUI 'GPU Benchmark Flux DEV fp8' thread: stock Flux dev fp8 workflow template, time of 2nd/3rd run (no loading); resolution/steps not stated in thread
26 s / image—2025-07-22github.com →
reportedHiDream-I1HiDream Full fp8 (T5 fp8, Llama3.1 fp8 scaled)
“ComfyUI Full example from their page at https://comfyanonymous.github.io/ComfyUI_examples/hidream/ runs at 115.09s on my RTX 3090, from cold start.”
HiDream-I1 Full via ComfyUI example workflow, Windows portable; cold start incl. load; resolution/steps not restated
115.09 s / image—2025-04-18huggingface.co →
reportedIllustrious / PonyIllustrious-XL-v2.0 · 1024x1024
“NVIDIA RTX 3090 | 4.00it/s | 0.25s/it | CUDA 12.9 | ComfyUI (Unknown) | Arch Linux”
Community speed list; ComfyUI, Euler/Normal, CFG 8, square 1:1, batch speed only (no per-image time); date = list last-updated date; SDXL 1024px section
0.25 s/it—2025-11-29huggingface.co →
reportedMiniMax H3transformer variant not stated; NVFP4 text encoder · 832x480
“render 23 min 17 s ... peak VRAM ~19.8 GB of 24 GB”
15.08 s clip at 24 fps with audio; needs --disable-pinned-memory on 32 GB RAM; 23 min 17 s = 1397 s; no date shown
1397 s / clip (362 frames)19.8 GB—github.com →
reportedSDXLsd_xl_base_1.0 · 1024x1024
“Prompt executed in 6.16 seconds”
ComfyUI 'GPU Benchmark' thread: default workflow, SDXL 1.0 base, 1024x1024, seed 1, second run
6.16 s / image—2024-03-06github.com →
reportedZ-Image TurboTongyi-MAI/Z-Image-Turbo · 1024x1024
“NVIDIA RTX 3090 | 0.50it/s | 2.01s/it | CUDA 12.6 | ComfyUI (5151cff) | Arch Linux”
Community speed list; ComfyUI, Euler/Normal, CFG 8, square 1:1, batch speed only (no per-image time); date = list last-updated date; Z-Image 1024px section; CFG 8 (above Turbo's us
2.01 s/it—2025-11-29huggingface.co →

Reported results are other people's numbers, copied as published, with a link. Settings, drivers and ComfyUI versions differ, so compare them with care. Send yours.