# NVIDIA Finds Key Scoring Takes 47% of Layer Prefill Time

> NVIDIA's AI model scoring uses 47% of prefill time, slowing long prompt processing.
- Title: NVIDIA Finds Key Scoring Takes 47% of Layer Prefill Time
- Summary: NVIDIA's AI model scoring uses 47% of prefill time, slowing long prompt processing.
- Keywords: nvidia, ai model, prefill time, deepseek v3.2, scoring, top-k selection, gpu processing, crypto, TokenPost
- URL: https://shortify.news/TokenPost/nvidia-finds-key-scoring-takes-47-of-layer-prefill-time
- Source: TokenPost — https://tokenpost.com/news/technology/60107
- Published: 2026-11-18T04:03:48+00:00
- Read time: 1 min
- Topics: nvidia, ai model, prefill time, deepseek v3.2, scoring, top-k selection, gpu processing, crypto, finance
## NVIDIA's AI Scoring Efficiency
NVIDIA found scoring and selecting what an AI model attends to consumes 47% of per-layer prefill time in a 128k token DeepSeek V3.2 example.
## Prefill Time Breakdown
Prefill is when a model processes input before generating a response, with scoring taking 47% of layer prefill time.
## Indexer Performance Impact
Increasing indexer key heads from one to eight made the indexer run 7.6 times longer during decode.
## Block Size Effects
Halving contiguous selection blocks nearly doubled core-attention runtime in both prefill and decode.
## Key takeaway

NVIDIA discovered scoring and selection processes consume 47% of layer prefill time in AI models, impacting efficiency for long inputs.

---

Summary: Shortify — https://shortify.news/TokenPost/nvidia-finds-key-scoring-takes-47-of-layer-prefill-time
Reporting: TokenPost — https://tokenpost.com/news/technology/60107

When you use this summary, link the Shortify URL for the summary and credit the reporting to its newsroom.