vLLM-Omni adds Day-0 H3-Image support on Vera Rubin
Start here
The vLLM-Omni team added Day-0 serving support for MiniMax's H3-Image model on NVIDIA Vera Rubin. In its short-prompt serving benchmark, native 4096×4096 text-to-image generation reached a median of 0.87 seconds on four GPUs; a separate throughput configuration delivered 6,115 images per hour, which the team projects to 146,755 images per day. Those figures exclude operating costs, and the daily capacity is an extrapolation of the measured rate. H3-Image is an open-weight model that handles both text-to-image and editing, pairing a Qwen3-VL encoder with a 33.1B dense DiT and an image VAE, and completing either task in six sampling steps with up to 16 reference images. The team attributes the speed to MXFP8 DiT GEMMs, Ulysses SP4, AdaLN caching and a VAE decoding change that cut decode time from 2.635s to 0.213s. The Day-0 integration currently uses MiniMax's hosted image API, and recipes are listed for AMD, Intel, Huawei Ascend and Moore Threads hardware.
Generated from reports · updated 15 h ago
Report timeline
Follow the reports to see the event from different sides.
- vLLM BlogPicksvLLM-Omni benchmarks MiniMax H3-Image on Vera Rubin: 0.87s 4K generation and 6,115 images/hour
vLLM-Omni brought MiniMax H3-Image to NVIDIA Vera Rubin and measured 0.87-second median latency for native 4096×4096 text-to-image on four GPUs, plus 6,115 images/hour in a separate throughput configuration. H3-Image is an open-weight text-to-image and editing model that pairs a Qwen3-VL encoder with a 33.1B dense DiT and an image VAE, handling both tasks in six sampling steps with up to 16 reference images. The team reports MXFP8 DiT GEMMs, Ulysses SP4, AdaLN caching and a VAE decoding change that cut decode time from 2.635s to 0.213s, and lists recipes for AMD, Intel, Huawei Ascend and Moore Threads hardware.
Heat for this event
Not enough continuous observations yet to draw a trend.