PulseAugur
EN
LIVE 05:09:53

Together AI optimizes ThunderKittens for NVIDIA Vera Rubin Blackwell GPUs

By PulseAugur Editorial · [1 sources] ·

Together AI has gained access to NVIDIA's Vera Rubin NVL72 platform, which is based on the Blackwell architecture. Their team has updated their ThunderKittens software to leverage new features of the Vera Rubin chip, specifically focusing on optimizing NVFP4 and FP8 GEMMs (General Matrix Multiply operations). Initial tests show that while the new platform doubles the K dimension capacity for tensor cores and increases tensor memory, existing Blackwell-optimized kernels were not feeding the cores fast enough, achieving only about 42-44% of theoretical performance. Together AI's optimizations aim to push performance closer to the theoretical ceiling, targeting over 22 PFLOPS and competitive speeds with existing libraries like cuBLAS. AI

IMPACT Optimizations for NVIDIA's Blackwell architecture could improve AI training and inference performance on compatible hardware.

RANK_REASON Software optimization for existing hardware architecture.

Read on Together AI blog →

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

Together AI optimizes ThunderKittens for NVIDIA Vera Rubin Blackwell GPUs

How we ranked this

Signal score
1 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Software optimization for existing hardware architecture.
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, product
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
1 days old
Coverage has settled into its steady-state source set.

Full methodology in our editorial standards.

COVERAGE [1]

  1. Together AI blog TIER_1 English(EN) ·

    To Infinity and Beyond: ThunderKittens Now on NVIDIA Vera Rubin NVL72!

    We ported ThunderKittens to NVIDIA's Vera Rubin NVL72 and rebuilt our NVFP4 GEMM around the new hardware, taking it from 42% of roofline to over 22 PFLOPS — competitive with cuBLAS and CuTe DSL. Here is what changed in the ISA and how we used it.

RELATED ENTITIES

RELATED TOPICS