Skip to content
Hot storyPast story

WorldPager proposes single-GPU tiered KV storage for long-horizon world models

1 report1 sourceUpdated 27 d ago

What happened

AI overview

Game AI researcher Bingnan Liu proposes WorldPager, which lets interactive world models generate video continuously for long stretches on a single consumer GPU. According to the paper, the method stores int4-quantized KV memory in tiers across GPU, DRAM and NVMe, and prefetches it based on camera pose, so generation does not pause to wait for data. The test setup was one RTX 4090, 32 GiB of host memory and one NVMe SSD. The system continuously generated 37.5 minutes of video, with host memory staying constant and a speed of about 9.5 frames per second. The paper says int4 storage showed no measurable quality loss on a closed test set. In long-interval revisit scenarios, image quality dropped by about 1 MUSIQ point. The authors claim this is the first scheme to place the KV memory of interactive world models at the NVMe tier with zero-stall prefetching. This claim comes from the authors' own description.

AI-written from the coverage · updated 26 d ago

Coverage timeline

Follow the coverage to see the story from different sides.

Oct 26
  1. arXiv Game World Models & GenerationPick
    WorldPager: zero-stall tiered memory for long-horizon world models on a single consumer GPU

    WorldPager stores the int4 KV memory of interactive world models across GPU, DRAM and NVMe tiers and prefetches it based on camera pose. On a single RTX 4090, it continuously generates 37.5 minutes of video without stalls, with host memory staying constant and a speed of about 9.5 frames per second. On a closed test set, int4 storage showed no measurable quality loss. On long-interval revisits, image quality dropped by about 1 MUSIQ point.