RTX 4070 12GB for MiniMax H3: Speed, VRAM & Recommended Setup

12GB GDDR6X · bandwidth 504 GB/s · Ada · sm_89 · estimated with 32GB RAM / fast-disk · updated 2026-09-28

Verdict: Recommended: 864 × 480 · 8s, ~3 min 2 s per clip (VRAM peak ~11GB). See the table for higher tiers.

Size (16:9)	5s	8s	10s	15s
416 × 256 0.1MP · flicker-prone	1 min 14 s VRAM 11G	1 min 24 s VRAM 11G	1 min 32 s VRAM 11G	1 min 51 s VRAM 11G
864 × 480 0.4MP	2 min 10 s VRAM 11G	3 min 2 s VRAM 11G	⚠ 3 min 45 s VRAM 11G	⚠ 5 min 55 s VRAM 11G
960 × 544 0.5MP	2 min 32 s VRAM 11G	⚠ 3 min 47 s VRAM 11G	⚠ 4 min 48 s VRAM 11G	⚠ 8 min VRAM 11G
1152 × 640 0.7MP	⚠ 3 min 22 s VRAM 11G	⚠ 5 min 33 s VRAM 11G	⚠ 7 min 22 s VRAM 11G	⚠ 13 min 25 s VRAM 11G
1344 × 768 1.0MP · official max	⚠ 4 min 47 s VRAM 11G	⚠ 8 min 39 s VRAM 11G	⚠ 12 min 3 s VRAM 11G	⚠ 24 min 7 s VRAM 11G
1376 × 768 1.0MP · above native	⚠ 4 min 55 s VRAM 11G	⚠ 8 min 57 s VRAM 11G	⚠ 12 min 31 s VRAM 11G	⚠ 25 min 13 s VRAM 11G
1920 × 1088 2.0MP · above native	⚠ 12 min 30 s VRAM 11G	⚠ 28 min 13 s VRAM 11G	⚠ 43 min 55 s VRAM 11G	✕ 1.8 h VRAM 11G

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 864 × 480 · 8s at ~3 min 2 s per clip. 1 high-end combinations are not recommended (pagefile pressure or >90 min per clip) — see the ✕ marks below. For comfortable 768p: add RAM to 64GB, or move to a card with 16GB+.

Recommended combo (Balanced tier (12-16GB VRAM))

Main model: minimax_h3_fl2va_pruned_int8_convrot.safetensors（20.97GB）
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

540p-768p 5s; 4-8 steps; 1.5-6 min per clip

Official minimum recommended combination (42.5GB); with 32GB RAM the TE+DiT cannot both stay resident, use --fast-disk

FAQ

Can the RTX 4070 12GB run MiniMax H3?

Yes. Estimated with 32GB RAM + fast-disk: 864×480·8s at ~3 min 2 s per clip. Configs that exceed VRAM fall back to weight offloading (slower, but they still render).

How fast is H3 on the RTX 4070 12GB?

About 3 min 2 s per 864×480·8s 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 32GB 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 4070 12GB pre-filled)

Same tier: RTX 3070 Ti 8GB · RTX 3080 10GB · RTX 3080 12GB · RTX 3080 20GB (modded) · RTX 3080 Ti 12GB · RTX 4060 Ti 16GB · RTX 4070 Super 12GB · RTX 4070 Ti 12GB