A developer released rGPU, an open-source project that lets PyTorch programs run GPU work on a remote NVIDIA machine via a custom `rgpu` device or a CUDA shim covering libcuda, CUDA Runtime, cuBLAS, c…
ComfyUI v0.36.0 ships optimizations that make the MiniMax H3 video VAE encode up to ~2.2x faster and decode ~1.4-2.7x faster, cutting a 1344x768, 129-frame encode-and-decode round trip from 24.3 to 12…
NVIDIA's cuDNN Frontend graph API lets developers express a computation as a graph of operations, have cuDNN select an execution engine, and then override that engine choice, according to a Marktechpo…
NVIDIA's Agentic Variation Operators (AVO) architecture achieved a 100% score on the ARC-AGI-3 benchmark, demonstrating a frontier-level general-purpose architecture for long-horizon autonomous agents…
Torch.compile is not a traditional compiler but a system that intercepts Python bytecode at runtime, extracts compilable regions through a multi-stage pipeline, and stitches them back with eager Pytho…
TornadoVM now natively integrates NVIDIA CUDA libraries cuBLAS, cuFFT, and cuDNN, allowing Java applications to call GPU-accelerated linear algebra and deep learning primitives directly from JIT-compi…
TornadoVM, a Java heterogeneous programming framework, now supports CUDA Graphs, multiple streams, and vendor libraries like cuBLAS, cuFFT, and cuDNN, enabling Java applications to leverage NVIDIA's t…
Photoroom reduced inference latency of its convolutional segmentation backbone by 22% on A100 GPUs by switching to PyTorch's channels-last memory format. The change required only four lines of code an…
Researchers at arXiv analyzed the performance of Med-DDPM, a 3D generative diffusion model for MRI synthesis, across three generations of NVIDIA GPUs. They identified inefficiencies in memory access a…
A developer built a 3-node cluster using NVIDIA Jetson Orin Nano Super 8GB developer kits, achieving ~759 Mbps per link, peak 58.3°C under full load, and zero throttling at 1728 MHz. The cluster is de…
A developer implemented FlashAttention's forward and backward passes from scratch in pure CUDA C++, achieving O(N) memory complexity through manual SRAM tiling and online softmax recurrence. A rejecte…