Releases·Tool Release
High-Throughput Structure Prediction with BioNeMo Inference Runtime
An NVIDIA blog post describes using the BioNeMo Inference Runtime (BioIR) to accelerate biomolecular structure-prediction models on NVIDIA GPUs at proteome scale, keeping a standard PyTorch workflow. The post targets practitioners moving entire prediction worklists through inference pipelines efficiently.
0 engagement1 sourceThu, Sep 10, 2026, 03:00 PM
AI Overview
- NVIDIA BioNeMo Inference Runtime (BioIR) accelerates supported structure-prediction models on NVIDIA GPUs within the PyTorch workflow
- BioIR uses optimized kernels and, where applicable, CUDA Graphs to speed proteome-scale prediction pipelines
Related
Developments·Breakthrough
Quantinuum and NVIDIA Validate Generative Quantum AI Framework for Pharmaceutical R&D
Quantinuum and NVIDIA announced validation of a generative quantum AI framework for pharmaceutical research and development. The collaboration combines Quantinuum's quantum computing with NVIDIA's AI platform to accelerate drug discovery workflows.
Quantinuum and NVIDIA announced validation of a generative quantum AI framework for pharmaceutical research and development. The collaboration combines Quantinuum's quantum computing with NVIDIA's AI platform to accelerate drug discovery workflows.
0 engagement1 sourceredditMon, Aug 3, 2026, 05:43 AM
Releases·Tool Release
New FP4 attention kernels for B300 achieve up to 1.69x speedup over FlashAttention-4
A new set of FP4 attention kernels optimized for NVIDIA's B300 GPU delivers up to 1.69x speedup compared to FlashAttention-4. The kernels leverage the B300's FP4 tensor core capabilities to accelerate attention computation, a key bottleneck in transformer inference.
A new set of FP4 attention kernels optimized for NVIDIA's B300 GPU delivers up to 1.69x speedup compared to FlashAttention-4. The kernels leverage the B300's FP4 tensor core capabilities to accelerate attention computation, a key bottleneck in transformer inference.
74 engagement1 sourceredditTue, Jul 14, 2026, 12:35 AM
Releases·Hardware Release
NVIDIA unveils Groq 3 LPX inference accelerator for Vera Rubin platform
NVIDIA announced Groq 3 LPX, an interactive AI inference accelerator designed for the NVIDIA Vera Rubin platform, paired with Vera Rubin NVL72. The combination aims to deliver high throughput and interactivity across a wide range of AI workloads, from small to large models, both open and closed.
NVIDIA announced Groq 3 LPX, an interactive AI inference accelerator designed for the NVIDIA Vera Rubin platform, paired with Vera Rubin NVL72. The combination aims to deliver high throughput and interactivity across a wide range of AI workloads, from small to large models, both open and closed.
29 engagement2 sourcesrss, redditMon, Aug 24, 2026, 03:00 PM
Releases·Tool Release
NVIDIA releases TensorRT Model Connect reference implementations for two-command model deployment
NVIDIA announced TensorRT Model Connect, an open collection of reference implementations for deploying open AI models with TensorRT in native C++ applications. It reduces deployment to two commands from checkpoint to inference, removing model-specific conversion, preprocessing, post-processing, and runtime code. This matters to practitioners shipping open models into native apps without hand-writing integration code.
NVIDIA announced TensorRT Model Connect, an open collection of reference implementations for deploying open AI models with TensorRT in native C++ applications. It reduces deployment to two commands from checkpoint to inference, removing model-specific conversion, preprocessing, post-processing, and runtime code. This matters to practitioners shipping open models into native apps without hand-writing integration code.
0 engagement1 sourcerssFri, Aug 28, 2026, 05:06 PM
Releases·Tool Release
Pulsar: SSD-streaming inference engine runs giant MoE models on consumer GPUs
Pulsar, an inference engine written in Rust and CUDA, streams MoE expert weights from NVMe SSDs per token, enabling models like GLM 5.2 743B at 2 tok/s and Hy3 295B at 7 tok/s on two consumer 16GB GPUs. It zero-config multi-GPU by measuring PCIe bandwidth and placing attention and hot experts optimally.
Pulsar, an inference engine written in Rust and CUDA, streams MoE expert weights from NVMe SSDs per token, enabling models like GLM 5.2 743B at 2 tok/s and Hy3 295B at 7 tok/s on two consumer 16GB GPUs. It zero-config multi-GPU by measuring PCIe bandwidth and placing attention and hot experts optimally.
29 engagement1 sourcegithubMon, Jul 13, 2026, 04:08 AM