Home / GPU ladder / RTX 3070 16GB (modded)
RTX 3070 16GB (modded) for MiniMax H3: Speed, VRAM & Recommended Setup
| Size (16:9) | 5s | 8s | 10s | 15s |
|---|---|---|---|---|
| 416 × 256 0.1MP · flicker-prone | 1 min 19 s VRAM 15G | 1 min 32 s VRAM 15G | 1 min 41 s VRAM 15G | 2 min 5 s VRAM 15G |
| 864 × 480 0.4MP | 2 min 27 s VRAM 15G | 3 min 29 s VRAM 15G | 4 min 20 s VRAM 15G | 6 min 50 s VRAM 15G |
| 960 × 544 0.5MP | 2 min 53 s VRAM 15G | 4 min 22 s VRAM 15G | 5 min 33 s VRAM 15G | 9 min 14 s VRAM 15G |
| 1152 × 640 0.7MP | 3 min 52 s VRAM 15G | 6 min 25 s VRAM 15G | 8 min 31 s VRAM 15G | 15 min 25 s VRAM 15G |
| 1344 × 768 1.0MP · official max | 5 min 32 s VRAM 15G | 9 min 59 s VRAM 15G | 13 min 52 s VRAM 15G | 27 min 30 s VRAM 15G |
| 1376 × 768 1.0MP · above native | 5 min 41 s VRAM 15G | 10 min 19 s VRAM 15G | 14 min 23 s VRAM 15G | 28 min 43 s VRAM 15G |
| 1920 × 1088 2.0MP · above native | 14 min 22 s VRAM 15G | 32 min 6 s VRAM 15G | 49 min 41 s VRAM 15G | ✕ 2.0 h VRAM 15G |
Times are per whole clip (model load and text encoding included); the first run is slower due to kernel compilation. ⚠ = weight offloading / noticeably slower; ✕ = not recommended.
Reading the numbers
With the recommended combo, the green (no disk spill) ceiling is 1920 × 1088 · 10s at ~49 min 41 s per clip. 1 high-end combinations are not recommended (pagefile pressure or >90 min per clip) — see the ✕ marks below. To unlock 2.0MP and full durations: step up to a 32GB card (e.g. 5090), or use a smaller quant (GGUF Q3/Q4).
Recommended combo (Blackwell-specific (50-series 16GB+))
- Main model: MiniMax-H3_FL2VA-NVFP4-HQ.safetensors（DmitryDB）（14.6GB）
- Text encoder: qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors（15.69GB）
- Acceleration: LightX2V Turbo 4-step v1.1 768p · ~3.44× sampling
- Launch flags:
--fast-disk
768p 5s; 4-8 steps; 1-5 min per clip
FAQ
Can the RTX 3070 16GB (modded) run MiniMax H3?
Yes. Estimated with 32GB RAM + fast-disk: 1920×1088·10s at ~49 min 41 s per clip. Configs that exceed VRAM fall back to weight offloading (slower, but they still render).
How fast is H3 on the RTX 3070 16GB (modded)?
About 49 min 41 s per 1920×1088·10s clip (including model load and encoding). Longer or denser clips scale quadratically in the attention term.
How much RAM do I need?
At least 64GB of system RAM recommended (minimum 32GB plus a 16GB pagefile). With 32GB, always enable --fast-disk or weights spill into the pagefile.
How can I make it faster?
Stack step distillation (e.g. LightX2V Turbo 4-step v1.1 768p), enable SageAttention / Kitchen attention; for long clips try SLA / VSA sparse attention. All of it is switchable in step 2 of the calculator.
→ Run the numbers for your RAM and budget in the calculator (RTX 3070 16GB (modded) pre-filled)
Same tier: RTX 3050 8GB · RTX 3060 8GB · RTX 3060 12GB · RTX 3060 Ti 8GB · RTX 3070 8GB · RTX 4060 8GB · RTX 4060 Ti 8GB · RTX 5050 8GB
Related: Hardware guide · Acceleration compared · 8GB VRAM setups