cd /news/artificial-intelligence/minimax-s-new-inference-stack-render… · home topics artificial-intelligence article
[ARTICLE · art-123444] src=nvlabs.github.io ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

MiniMax's new inference stack renders 5 seconds of video with stereo audio in under 2 seconds, about 11 times faster than its own base model

NVIDIA Research and Singapore Lab introduced Sol-H3, an inference stack for MiniMax's H3 video generation model, that renders 5 seconds of video with stereo audio in 1.653 seconds on an 8× NVIDIA B300 system—11.04 times faster than the base model's 18.250 seconds. The stack combines Sol-Engine and Sol-Attn to unify sparse attention, fused kernels, and multi-GPU communication, achieving up to 15.05× speedup for 15-second outputs.

read5 min views2 publishedSep 8, 2026
MiniMax's new inference stack renders 5 seconds of video with stereo audio in under 2 seconds, about 11 times faster than its own base model
Image: source

Sol-H3Speed-of-Light MiniMax-H3 on an 8× NVIDIA B300 Blackwell System NVIDIA Research, Efficient AI Team & Singapore Lab.

5sof world.1.653sto infer.

Sol-H3 generates MiniMax-H3 video with native audio faster than playback on one 8× NVIDIA B300 system. Sol-Engine × Sol-Attn unifies sparse attention, fused kernels, fast multi-GPU communication, parallel video decoding, and other serving optimizations in one runtime.

Featured sample · MiniMax-H3 Base Ref2VA at 50 steps, generated through Sol-Engine on 8× NVIDIA B300.

Scroll to enter the engine

01

Benchmark · 8 / 4 / 1× NVIDIA B300

Latency across GPU counts.

Choose an output length, then compare Sol-H3 directly with Base H3 across 8, 4, and 1 B300 GPUs.

Measured generation time · seconds

Generation time across GPU counts.

Measured latency for 50-step Base H3 Dense and four-step Sol-H3 at 5, 10, and 15 second output lengths on one, four, and eight NVIDIA B300 GPUs.

GPU count

5s124 frames · 1344×768 · 24 FPS · stereo audio

10s243 frames · 1344×768 · 24 FPS · stereo audio

15s362 frames · 1344×768 · 24 FPS · stereo audio

8×NVIDIA B300

Base H3Dense · 50 steps18.250 s

Sol-H3Ours · 4 steps1.653 s11.04× faster

Base H3Dense · 50 steps50.660 s

Sol-H3Ours · 4 steps3.732 s13.57× faster

Base H3Dense · 50 steps99.513 s

Sol-H3Ours · 4 steps6.612 s15.05× faster

4×NVIDIA B300

Base H3Dense · 50 steps35.328 s

Sol-H3Ours · 4 steps2.918 s12.11× faster

Base H3Dense · 50 steps100.440 s

Sol-H3Ours · 4 steps6.993 s14.36× faster

Base H3Dense · 50 steps194.930 s

Sol-H3Ours · 4 steps12.542 s15.54× faster

1×NVIDIA B300

Base H3Dense · 50 steps129.898 s

Sol-H3Ours · 4 steps13.745 s9.45× faster

Base H3Dense · 50 steps376.942 s

Sol-H3Ours · 4 steps37.813 s9.97× faster

Base H3Dense · 50 steps746.885 s

Sol-H3Ours · 4 steps52.260 s14.29× faster

Measured scopeAll runs use 1344×768, 24 FPS, the same prompt and seed 20260903, and reference-free T2VA with stereo audio. The complete profiles differ: Base H3 uses 50 scheduler points (49 DiT forwards), while Sol-H3 uses four DiT forwards—not an attention-only runtime change. Sol-H3 uses Dense attention on 1× B300 and SOL with INT8 QKV / FP8 output transport on 4× / 8× B300. Each value is the median of three measured runs after one warmup. Timing includes text encoding, DiT denoising, and video/audio VAE decoding; model , compilation warmup, and final MP4 encoding are excluded.

02

Selected Sol-H3 generations

Sol-H3 video showcase.

Browse ten-second Sol-H3 videos with native audio, two examples at a time.

showcaseGenerated with Sol-H3

Selected examples.

Browse two videos at a time with the arrows; each video includes its exact prompt.

Showcase profileEach ten-second sample is generated by Sol-H3 at 1344×768 with native audio, using Sol-H3’s four-step Dense/Data-Free adapter from FastH3 Preview v1.

03

Two core systems

Sol-Engine × Sol-Attn. Sol-Engine speeds up the complete inference pipeline. Sol-Attn selects attention blocks while the model runs, with no retraining.

Sol-Engine keeps the model ready and coordinates GPU kernels, communication across eight GPUs, compilation, and video/audio decoding.

02 · Sol-Attn

Select the blocks that matter. Sol-Attn uses a query-dependent threshold to keep important attention blocks exact and cheaply approximate the rest. It works directly from the model’s Q/K/V tensors, so existing weights need no retraining.

04

Six optimizations · one runtime

How Sol-H3 gets faster.

Each card pairs one optimization with how it works, why it is faster, and the measured evidence.

Each 64-token query block sets its own threshold: mean + τ × standard deviation over lightweight query–key scores. Selected blocks use exact attention; skipped blocks receive an approximate correction from pooled K/V summaries.

Why it is faster

Routing, sparse attention, and approximate correction share one online-softmax pass. Sol-Attn writes no separate router output, full score map, or routing-index tensor, and needs no retraining.

GPU kernels

Fuse repeated operations

Norm · RoPE · MLP

02GPU kernels

Move less data between operations.

Mechanism

Sol-H3 fuses three repeated chains: residual + RMSNorm + modulation, QKNorm + partial RoPE, and SwiGLU. Their intermediate tensors stay in registers or on chip instead of returning to GPU memory between operations.

Why it is faster

These operations touch wide tensors but do relatively little math, so memory traffic is often the limit. Fewer reads, writes, and launches save traffic across all 50 transformer blocks.

Multi-GPU communication

Faster 8-GPU data exchange

Pack · all-to-all · merge 038-GPU exchange

Pack, exchange, and merge directly.

Mechanism

Ulysses redistributes sequence shards by attention head, then restores the original layout. Sol-H3 packs Q/K/V directly for the destination GPU, performs one all-to-all exchange, and merges returned heads without extra full-tensor layout copies.

Why it is faster

This removes separate stack, permute, and contiguous copies. The fastest profile sends QKV in INT8 and attention output in FP8; BF16 transport remains available when higher numerical fidelity is preferred.

Sparse-attention setup

Fuse the setup work

K/V summaries · threshold · metadata

BeforeFused

04Sparse setup

Build the sparse plan in fewer passes.

Mechanism

One preprocessing kernel computes each key centroid and value sum together. Thresholding also returns the pooled query for reuse; the final compaction step applies threshold, local-band, and exact-prefix rules while emitting the mask and compact block lists.

Why it is faster

The default SOL path feeds its fused K/V summaries, mask, and block list directly into cuDNN block-sparse attention. The optional native Sol-Attn kernel combines routing and sparse attention in one pass. Both reduce temporary traffic and launch overhead.

Video decoding

Parallel VAE decoding

Tile parallel · batch · compile

05VAE decoding

Decode spatial tiles in parallel.

Mechanism

A 5s output has seven temporal clips and 196 equal spatial tiles. The default path distributes the full tile set over eight GPUs, batches each GPU’s share into one compiled decoder call, and gathers once.

Why it is faster

Tile sharding cuts the work per GPU, while global batching replaces seven per-clip decode-and-gather rounds with one. A per-clip path remains available when matching the initial output exactly matters.

Conditioning

Precompute AdaLN

Precompute · cache · direct lookup

06AdaLN cache

Compute conditioning once, then reuse it.

Mechanism

All AdaLN values are knowable before denoising. Sol-H3 projects every block-and-step value up front, caches the full trajectory, and releases the projection weights before the denoising loop.

Why it is faster

Each block reads its cached slice directly instead of running two index selections. Across 50 blocks and four steps, that removes about 400 small launches and avoids roughly 26 GB of GPU-memory reads per step without changing output.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @nvidia research 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/minimax-s-new-infere…] indexed:0 read:5min 2026-09-08 ·