Install
CUDA & Compute
CUDA, creator workloads, AI frameworks, and encoder/decoder blocks.
- 17 Tracked terms
- Last 30 days Feed window
What this topic collects on
An article joins this feed when it matches these terms. Each one is also a search of its own.
Related topics
Latest in CUDA & Compute
Holo3.1-0.8B vs Holo3.1-35B-A3B-FP8: Benchmarks & Cost
16+ hour, 58+ min ago (90+ words) Updated October 10, 2026. We do not rank this pair: at least one has no public score. Public scores include evidence status and uncertainty. This is a same-family comparison, so migration details appear when the source data supports them. The documented context…...
PyTorch Hardware Enablement and the Accelerator Integration Working Group Updates
10+ hour, 51+ min ago (207+ words) While compute diversity drives efficiency and cost reductions across the AI supply chain, it presents a massive software engineering hurdle: standardizing framework integration. Without unified abstractions, hardware vendors are forced to maintain fragmented forks of PyTorch, resulting in delayed feature…...
NPU vs. GPU: Key Differences, Uses, and Which One Do You Need?
9+ hour, 57+ min ago (444+ words) AI is changing how we use computers, from real-time video effects to training advanced machine learning models. Two processors powering these capabilities are Neural Processing Units (NPUs) and Graphics Processing Units (GPUs). Although both accelerate computing, they serve different purposes....
NVIDIA cuPhoton accelerates scientific image analysis
2+ day, 21+ hour ago (298+ words) The idea is to keep image data resident on the GPU from ingestion through processing, fitting and classification rather than repeatedly transferring it between host and accelerator memory. NVIDIA’s GitHub project repository includes modules for loading FITS and HDF5 data, reprojection,…...
AES-256-GCM vs Ascon: NIST Standard Cuts Code 2 Ways [2026]
1+ day, 4+ hour ago (1253+ words) AES-256-GCM’s security rests on 256-bit key strength and a strict requirement: never reuse a nonce with the same key. Nonce reuse under GCM doesn’t just weaken the cipher, it can expose the authentication key entirely, which is why implementations…...
AWS Graviton5 R9g: 25% Faster, No Sydney Yet [2026]
1+ day, 2+ hour ago (317+ words) Based on how AWS has staged every previous Graviton rollout, and on the scale of its current Australian infrastructure commitments, a few outcomes look likely over the next year. What is AWS Graviton5?Graviton5 is AWS’s fifth-generation custom Arm-based server processor, built…...
Scaling Decision Optimization to 100 Million Variables and Beyond with mPDLP in NVIDIA cuOpt
3+ day, 1+ hour ago (1167+ words) Supply chain problems are expanding across more SKUs, lanes, and constraints than ever before, while energy grids are balancing more distributed sources in real time. To meet these challenges, teams need to evaluate larger models and more uncertainty within a…...
NVIDIA Introduces Multi-GPU Solver for 100M-Variable Models
2+ day, 10+ hour ago (336+ words) The cuOpt solver uses NVLink-connected GPUs to handle models with up to 2.1 billion nonzero matrix entries while cutting per-GPU memory use by up to 6x. NVIDIA introduced a multi-GPU solver for linear-programming models with more than 100 million variables, bringing distributed GPU…...
Building Spyre as a Native PyTorch Device
2+ day, 4+ hour ago (1624+ words) Spyre is IBM’s dataflow AI accelerator, optimized for inference. It is designed for enterprise teams running AI alongside their applications and data on IBM Z, LinuxONE, and Power systems. Its reduced-precision compute is well suited to the matrix-heavy work in…...
Integrating Spyre as a Native PyTorch Device
1+ day, 10+ hour ago (148+ words) The evolution of the PyTorch ecosystem is defined by its ability to integrate diverse hardware backends. By implementing Spyre as a native PyTorch device, developers can now unlock specialized performance without abandoning the familiar PyTorch API. This integration, facilitated by…...