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

Can the RX 9070 XT 16 GB run MiniMax H3 (33B)?

Video model · 33.1B16 GB GDDR6 · 256-bit · 640 GB/s · RDNA 4Data 2026-09-25
Offload
Not entirely in VRAM, but it can still run.

The smallest sensible file, Q4_K_M (19.9 GB), needs about 25.7 GB — 9.7 GB more than RX 9070 XT 16 GB has. ComfyUI can still run it by streaming part of the model from system RAM. How much slower that is depends on your ComfyUI version, the file format and the PCIe link — see the note below.

Best fileQ4_K_M
File size19.9 GB
VRAM needed~25.7 GB
System RAM64 GB+
Memory map · RX 9070 XT 16 GB25.7 GB needed · 9.7 GB over 16 GB
013 GB26 GB
Weights Q4_K_M · 19.9 GBWorking memory · 5.0 GBReserve · 0.8 GBSpills to system RAM · 9.7 GB
calculated from real file sizes plus working memory. Not yet measured on this setup. How this works.

01What to download

PartFileFolderSize
ModelMiniMax-H3-FL2VA-Q4_K_M.gguf
Q4_K_M · Abiray/MiniMax-H3-GGUF
models/unet19.9 GBDownload →
Text encoderqwen3vl_32b_minimax_h3_int8_convrot.safetensors
Qwen3-VL 32B INT8 (convrot) · Comfy-Org/MiniMax-H3
models/text_encoders27.1 GBDownload →
VAEminimax_h3_video_vae_fp16.safetensors
MiniMax H3 video VAE FP16 · Comfy-Org/MiniMax-H3
models/vae5.2 GBDownload →
Also neededminimax_h3_audio_vae_fp32.safetensors
MiniMax H3 audio VAE FP32 · Comfy-Org/MiniMax-H3
models/vae0.6 GBDownload →
Total download · keep about the same free on disk52.8 GB

Alternatives: Qwen3-VL 32B NVFP4 (AWQ) (15.7 GB); Qwen3-VL 32B BF16 (51.5 GB). The 16-bit text encoder is a little more faithful; the smaller one is chosen here because it loads faster and needs less RAM. FP4 files are listed only as alternatives unless the GPU is an RTX 50 (Blackwell). Sizes read from Hugging Face (2026-09-25). “Download” links start the file directly; the file name links are the same files the official ComfyUI workflows use.

System RAM: 64 GB or more. ComfyUI keeps the model file, the text encoder and the VAE in system RAM and moves them to the GPU as needed. With this set of files that is about 52.8 GB, plus roughly 6 GB for Windows, ComfyUI and a browser: 58.8 GB in total. With less RAM it still runs, but Windows starts swapping to disk and loading gets very slow. calculated

02Every MiniMax H3 file on 16 GB

FileSizeNeededOn this cardQualityDownload
16-bit
SAFETENSORS · Comfy-Org
66.3 GB72.1 GB56.1 GB too bigthe original weightsHugging Face →
Q8_0
GGUF · Abiray
36.0 GB41.8 GB25.8 GB too bigpractically identical to the originalHugging Face →
INT8
SAFETENSORS · Comfy-Org
34.0 GB39.8 GB23.8 GB too bigpractically identical to the originalHugging Face →
Q6_K
GGUF · Abiray
28.2 GB34.0 GB18.0 GB too bigvery close to the originalHugging Face →
Q5_K_M
GGUF · Abiray
23.9 GB29.7 GB13.7 GB too bigclose; small differences in fine detailHugging Face →
Q4_K_M ←
GGUF · Abiray
19.9 GB25.7 GB9.7 GB too biggood; some loss in fine detail and textHugging Face →
Q3_K_M
GGUF · Abiray
15.6 GB21.4 GB5.4 GB too bignoticeable loss of detailHugging Face →

Sizes from the Hugging Face file listing, read 2026-09-25. “Needed” = file + 5 GB working memory for this model + 0.8 GB kept free for the system.

03The text encoder

Qwen3-VL 32B: 51.5 GB as 16-bit, 27.1 GB as INT8, 15.7 GB as NVFP4. ComfyUI encodes the prompt first and can push the encoder out of VRAM before sampling, so it does not have to fit together with the model. On 16 GB the encoder itself is too big for VRAM, so let it run from system RAM: slower prompt encoding, same images.

04About this GPU

ComfyUI supports AMD GPUs on Windows officially since v0.7.0 (January 2026), through ROCm. Memory works the same as on NVIDIA, so the fit verdicts apply. Some custom nodes are NVIDIA-only. RX 9000 (RDNA 4) GPUs have FP8 hardware, and ComfyUI computes FP8 natively on them with PyTorch 2.7+ and ROCm 6.4+.

RX 9070 XT 16 GB: 16 GB GDDR6 · 256-bit · 640 GB/s · RDNA 4. Everything that runs on the RX 9070 XT 16 GB →

05What more VRAM would change

With 48 GB you could run MiniMax H3 at 8-bit or better (Q8_0, 36.0 GB) with no offloading.

06Measured and reported results

Nobody has sent measured numbers for this pair yet. If you run MiniMax H3 on a RX 9070 XT 16 GB, send your time per image and peak VRAM and it will appear here, credited.

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), counting working memory and a small system reserve. The full 16-bit file needs about 72.1 GB.

Which MiniMax H3 file should I download for the RX 9070 XT 16 GB?

Q4_K_M (19.9 GB) from Abiray/MiniMax-H3-GGUF. Load it with Unet Loader (GGUF) from the ComfyUI-GGUF node pack.

Is FP8 faster than GGUF on the RX 9070 XT 16 GB?

It can be. This GPU has FP8 hardware, and ComfyUI computes FP8 natively for files made for it (Comfy-Org's fp8_scaled files), or for any FP8 file with the --fast fp8_matrix_mult option. GGUF files are unpacked on the fly, which costs some speed.

How much do I need to download for MiniMax H3 on the RX 9070 XT 16 GB?

About 52.8 GB for the model file, text encoder and VAE listed on this page, and the same again free on disk. With a 16 GB card, 64 GB of system RAM or more is recommended.