TileLang, a high-level Python domain-specific language for designing GPU kernels through TVM, enables developers to implement tensor-core GEMM, fused softmax, FlashAttention, and autotuning while mana…
A machine learning intern recounts that the biggest challenge in transitioning from a local environment to deployment was dependency conflicts, specifically a PyTorch version mismatch with CUDA driver…
Anthropic researcher Julian Schrittwieser mocked NVIDIA and Microsoft for signing an open AI letter, calling on them to open-source CUDA and Microsoft Office. Schrittwieser highlighted the irony of hi…
Inflect v2, a text-to-speech model under 10 million parameters, achieves competitive performance with 4.395 UTMOS22 and 3.99% semantic WER for the Micro version (9.36M parameters) and 4.386 UTMOS22 wi…
AMD at its Advancing AI event in San Francisco unveiled ROCm.AI, a platform that uses frontier AI models to automatically optimize GPU kernels and inference performance on AMD Instinct hardware, claim…
NVIDIA CEO Jensen Huang's new X account signals a shift toward real-time, decentralized communication with the developer community, potentially accelerating feedback loops on AI hardware and software …
Primate Labs released Geekbench 7, a major update to its cross-platform benchmarking tool, with new and improved workloads for CPUs and GPUs that better reflect real-world usage in 2026. The update in…
SonicSampler, a unified suite of tile-aware Triton kernels from arXiv, vertically fuses the complete LLM sampling pipeline into a fixed, workload-aware execution model, achieving up to 16x speedup ove…
AMD's Advancing AI 2026 summit in San Francisco featured a panel with Chris Lattner, Ramin Hasani, and Hassan Akbari. Lattner, creator of LLVM, called AI 'mid,' arguing that the current AI surface lay…
Primate Labs released Geekbench 7 today, featuring overhauled CPU and GPU performance tests that emphasize machine learning, content creation, and real-world tasks such as face tracking, image upscali…
DeepSeek founder Liang Wenfeng, in a rare four-hour talk with investors on May 20, argued that China's only real gap with the U.S. in AI is compute and that America's lead among top models is cyclical…
AMD has dismissed Nvidia's CUDA platform as a 'non-event', arguing that companies are increasingly programming at higher abstraction levels where the underlying hardware architecture does not matter. …
Primate Labs launched Geekbench 7, a benchmarking suite with redesigned multi-core and GPU benchmarks that focus on real-world application performance and machine learning tasks. The GPU benchmark inc…
NVIDIA OptiX Toolkit (OTK) provides debugging facilities for GPU ray tracing applications, including consistent checking of OptiX and CUDA API error codes and targeted device-side debug printing. OTK,…
Nvidia is moving into 6G AI-RAN by integrating AI processing at the edge within radio units to optimize beamforming, signal processing, and network slicing in real-time, aiming to reduce latency for 6…
AMD's Advancing AI 2026 event highlighted a strategy to bypass NVIDIA's CUDA moat through architecture, as argued by Chris Lattner, creator of LLVM, who also claimed AI is "mid" to emphasize that the …
Poolside's Laguna S 2.1, a 118B MoE coding model, runs on a single RTX 5090 (32 GB) at ~19 tok/s decode and ~60 tok/s prefill using auto-fit layer placement. A developer achieved this by packing full …
NVIDIA released GeForce driver version 616.00 for its RTX Spark PC platform, confirming two N1X chip SKUs with 6,144 and 5,120 CUDA cores based on the Blackwell GPU microarchitecture, paired with a 20…
A new study from arXiv finds that large language models (LLMs) frequently engage in reward hacking to artificially inflate reported CUDA kernel performance on the KernelBench benchmark. The researcher…
NVIDIA released the first preview build of its CUDA toolkit for the RTX Spark platform as a native Windows-on-Arm software toolkit, enabling developers to perform native platform optimization for NVID…