PulseAugur
EN
LIVE 19:43:02

llama.cpp optimizes CUDA top-k algorithm for significant speedup

By PulseAugur Editorial · [1 sources] ·

The llama.cpp project has released an update (b11513) that significantly optimizes the CUDA implementation of the top-k algorithm. This update replaces the per-row DeviceTopKKernel with a more efficient grid-over-rows radix select for large row counts, drastically reducing processing time. For instance, on the qwen4exp model with 34,816 tokens, the top-k operation speed improved from over 5.7 seconds to approximately 941 milliseconds. The release also refines the selection of the optimal top-k implementation based on shape, utilizing bitonic, radix select, or DeviceTopK/CUB argsort depending on row length and configuration. AI

IMPACT Improves inference performance for models running on CUDA-enabled hardware via llama.cpp.

RANK_REASON This is a software update for an open-source project that improves performance of a specific algorithm on certain hardware, not a frontier release or significant industry event.

Read on llama.cpp — Releases →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

llama.cpp optimizes CUDA top-k algorithm for significant speedup

How we ranked this

Signal score
13 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
This is a software update for an open-source project that improves performance of a specific algorithm on certain hardware, not a frontier release or significant industry event.
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
infra
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. llama.cpp — Releases TIER_1 English(EN) · praneshgo ·

    b11513: CUDA: improve top-k algorithm selection (#28713)

    <ul> <li>CUDA: radix top-k for large row counts</li> </ul> <p>Replaces CUB's per-row DeviceTopKKernel with a grid-over-rows radix select,<br /> gated on GGML_CUDA_TOPK_RADIX_MIN_ROWS. On qwen4exp at 34,816 tokens this cuts<br /> top-k from 1,671,253 launches / 5,761.8 ms to 2,329…

RELATED ENTITIES

RELATED TOPICS