Can the RTX 4090 24 GB run MiniMax H3 (33B)?
Not at full quality. The 16-bit and 8-bit files are too big for the RTX 4090 24 GB; the largest that fits is Q3_K_M (15.6 GB).
Take Q3_K_M (15.6 GB). In use it comes to about 21.4 GB, with 2.6 GB left over on 24 GB. Quality: noticeable loss of detail. For better quality, Q4_K_M (19.9 GB) also runs, with about 1.7 GB spilling into system RAM - a little slower, still practical.
Also try the INT8 file (34.0 GB) on this GPU. It does not fit entirely, but a current ComfyUI streams the overflow from RAM and native 8-bit files often beat a small GGUF that does fit. Dynamic VRAM →
01What to download
| Part | File | Folder | Size | |
|---|---|---|---|---|
| Model | MiniMax-H3-FL2VA-Q3_K_M.gguf Q3_K_M · Abiray/MiniMax-H3-GGUF | models/unet | 15.6 GB | Download → |
| Text encoder | qwen3vl_32b_minimax_h3_int8_convrot.safetensors Qwen3-VL 32B INT8 (convrot) · Comfy-Org/MiniMax-H3 | models/text_encoders | 27.1 GB | Download → |
| VAE | minimax_h3_video_vae_fp16.safetensors MiniMax H3 video VAE FP16 · Comfy-Org/MiniMax-H3 | models/vae | 5.2 GB | Download → |
| Also needed | minimax_h3_audio_vae_fp32.safetensors MiniMax H3 audio VAE FP32 · Comfy-Org/MiniMax-H3 | models/vae | 0.6 GB | Download → |
| Total download · keep about the same free on disk | 48.5 GB | |||
Alternatives: Qwen3-VL 32B NVFP4 (AWQ) (15.7 GB); Qwen3-VL 32B BF16 (51.5 GB). FP4 files are listed only as alternatives unless the GPU is an RTX 50 (Blackwell). Sizes from Hugging Face, 2026-10-01.
System RAM: 64 GB or more. These files take about 48.5 GB of RAM while loaded, 54.5 GB with Windows and ComfyUI. What happens with less. calculated
02Every MiniMax H3 file on 24 GB
| File | Size | Needed | On this card | Quality | Download |
|---|---|---|---|---|---|
| 16-bit SAFETENSORS · Comfy-Org | 66.3 GB | 72.1 GB | 48.1 GB too big | the original weights | Hugging Face → |
| INT8 SAFETENSORS · Comfy-Org | 34.0 GB | 39.8 GB | 15.8 GB too big | practically identical to the original | Hugging Face → |
| Q8_0 GGUF · Abiray | 36.0 GB | 41.8 GB | 17.8 GB too big | practically identical to the original | Hugging Face → |
| Q6_K GGUF · Abiray | 28.2 GB | 34.0 GB | 10.0 GB too big | very close to the original | Hugging Face → |
| Q5_K_M GGUF · Abiray | 23.9 GB | 29.7 GB | 5.7 GB too big | close; small differences in fine detail | Hugging Face → |
| Q4_K_M GGUF · Abiray | 19.9 GB | 25.7 GB | spills 1.7 GB | good; some loss in fine detail and text | Hugging Face → |
| Q3_K_M ← GGUF · Abiray | 15.6 GB | 21.4 GB | fits · 2.6 GB spare | noticeable loss of detail | Hugging Face → |
“Needed” adds 5 GB of working memory for MiniMax H3 and a 0.8 GB reserve. Sizes 2026-10-01.
03The text encoder
Take the smallest Qwen3-VL 32B file (15.7 GB): it is the only one that fits in 24 GB on its own. Sizes: 16-bit 51.5 GB, INT8 27.1 GB, NVFP4 15.7 GB. Why the encoder does not count against VRAM →
04Measured and reported results
Nobody has sent measured numbers for this pair yet. If you run MiniMax H3 on a RTX 4090 24 GB, send your time per image and peak VRAM and it will appear here, credited.
05With more VRAM
With 48 GB you could run MiniMax H3 at 8-bit or better (INT8, 34.0 GB) with no offloading.
What changes from the RTX 4090 24 GB to the RTX 5090 32 GB →
06About the RTX 4090 24 GB
The RTX 4090 24 GB is an Ada Lovelace GPU with hardware FP8, so ComfyUI can compute Comfy-Org's FP8 files natively: small and fast. (Plain FP8 files use FP8 maths with the --fast fp8_matrix_mult option.)
24 GB GDDR6X · 384-bit · 1008 GB/s · Ada Lovelace. Everything that runs on it →
07Questions
How much VRAM does MiniMax H3 need?
Around 21.4 GB with the smallest file (Q3_K_M) and 39.8 GB with an 8-bit file (INT8). The full 16-bit file needs about 72.1 GB.
Which MiniMax H3 file should I download for the RTX 4090 24 GB?
Q3_K_M (15.6 GB) from Abiray/MiniMax-H3-GGUF. Load it with Unet Loader (GGUF) from the ComfyUI-GGUF node pack.
How much do I need to download for MiniMax H3 on the RTX 4090 24 GB?
48.5 GB: MiniMax-H3-FL2VA-Q3_K_M.gguf, qwen3vl_32b_minimax_h3_int8_convrot.safetensors, minimax_h3_video_vae_fp16.safetensors, minimax_h3_audio_vae_fp32.safetensors.