Mamba: New Selective State Space Model vs Transformer

Mamba: a selective SSM state-space model that generalizes S4 to enable linear long-context scaling, million-token sequences, and improved language modeling.

Global and China AI Server Market Size

Analysis of AI server market dynamics, NVIDIA dominance, and Huawei Ascend's role in domestic substitution of AI chips amid export controls and foundry capacity gains.

Why A100/H100 Beat RTX 4090 for Large-Model Training

Technical analysis comparing RTX 4090 and H100 GPUs: why the 4090 is impractical for large-model training but viable for inference with optimized batching and KV cache.

Inside Amazon's AI Chip Architecture

Analysis of AWS Trainium2 architecture and its relationship to Inferentia, with performance projections, core/memory scaling, NeuronLink bandwidth and instance implications.

Sora: OpenAI's Video Model Architecture and Use Cases

Overview of Sora, OpenAI's video diffusion model using a spatiotemporal autoencoder and DiT Transformer to generate high-resolution, minute-long videos.

Nvidia AI Chip Roadmap Explained

Technical analysis of Nvidia's GPU roadmap: annual cadence with H200/B100/X100, One Architecture, SuperChip design and NVLink interconnect evolution.

AI Server PCB Hardware Breakdown

Teardown analysis of NVIDIA DGX A100 AI server PCBs: PCB types, area and per-system value breakdown for GPU board assembly, CPU motherboard, substrates and accessories.

ORB-SLAM3 Architecture and Workflow Analysis

Overview of ORB-SLAM3 architecture and visual-inertial SLAM: tracking, local mapping, loop/map merging, Atlas and IMU-camera fusion for pose estimation and optimization.

Precision Comparison: FP64 FP32 FP16 TF32 BF16 INT8

Concise overview of numeric precision formats, FP64, FP32, FP16, TF32, BF16 and int8, comparing bit widths, accuracy trade-offs and use cases for AI training and inference.

NVIDIA's Three Core Technologies: CUDA, NVLink, InfiniBand

Technical overview of CUDA and NVLink for GPU-accelerated AI: architecture, interconnect bandwidth, and scalable multi-GPU networking.