Hands-on benchmarks of the M5 Ultra 256GB Mac Studio show 45.7% faster single-stream LLM decoding, 3.6x higher multi-user prefill throughput, 43.6% CPU multi-core gains, 67.8% GPU improvement, and 82.8% higher Cyberpunk 2077 4K ray-tracing frame rates versus M3 Ultra, with purchasing guidance.
DiffusionGemma, an experimental discrete text diffusion model built on the 26B MoE Gemma‑4 architecture, generates whole 256‑token blocks with bidirectional attention, moving the inference bottleneck from memory bandwidth to GPU compute, achieving up to four‑fold speed gains on H100 and RTX 5090 GPUs, though with lower output quality than standard autoregressive models.
This article compares six popular LLM inference frameworks—vLLM, TensorRT‑LLM, llama.cpp, ds4.c, Ollama, and Omlx—across performance, ease of use, and hardware compatibility, then provides a practical matrix to help users select the best fit for their GPU.
The article walks through a step‑by‑step optimization of a simple elementwise addition kernel (C = A + B) on HyperAI's RTX 5090 cloud instance, covering FP32 baseline, vectorized FP32, several FP16 variants, benchmark methodology, performance results, and the reasoning behind thread‑block sizing.
This article explains why GPUs have become the preferred platform for high‑performance computing, covering Dennard scaling, GPU speed advantages, theoretical FLOPS calculations, CUDA programming examples like SAXPY, the SIMT execution model, instruction pipelines, and modern techniques for handling branch divergence and register bank conflicts.
Geekbench scores show the iPhone 17 Pro and Pro Max delivering a 15% single‑core and 22% multi‑core CPU boost plus a 41% GPU performance jump compared with the iPhone 16 Pro, while the new models also feature up to 12 GB of RAM and improved thermal design.
This article examines the latest Nvidia GPU lineup—including A100, H100, A800, H800, and the upcoming H20—detailing their architectures, performance metrics for AI training and inference, cost considerations, and provides a step‑by‑step guide for building a high‑performance compute center.
This article provides a comprehensive, engineer‑focused breakdown of transformer architecture—including tensor fundamentals, matrix multiplication, GPU theoretical compute, attention and FFN mechanics, quantitative parameter and FLOP analysis, performance metrics like MFU, parallelism strategies, variant optimizations, and practical exercise questions—offering clear insight into large‑model efficiency and scaling.
In this talk from the 2024 China Generative AI Conference, Li Peng outlines the escalating computational demands of large‑model training and inference, identifies power, memory and communication walls, and presents Alibaba Cloud’s DeepGPU solutions and best‑practice strategies for scaling AI workloads on cloud GPUs.
In a live session, NVIDIA senior deep‑learning solutions architect Zhai Jian demonstrates how to use Nsight Systems and Nsight Compute to analyze a simple neural‑network training workload, accelerate BERT with mixed precision, and examine matrix‑transpose kernels, with registration via QR code and a detailed event schedule.
This technical walkthrough explains how Swin Transformer training and inference can be dramatically accelerated on NVIDIA GPUs by using Nsight Systems profiling, mixed‑precision tensor‑core kernels, Apex‑based and custom CUDA operator fusion, half2 vectorization, register‑array caching, and INT8 quantization, achieving up to 2.85× training and 7.34× inference speedups while preserving model accuracy.
The team enhanced real‑time recommendation by redesigning TensorFlow graphs—using constant‑folding, a custom CallGraphOP cache, a simplified dense layer, and CUDA‑Graph compatibility—boosting single‑machine throughput ~40%, raising GPU utilization from 30% to 43%, cutting latency and saving roughly 30% of hardware resources.
