Hardware2026-06-18
Blackwell B200 and GB200 NVL72 ramp across clouds
NVIDIA's Blackwell data-centre parts are now bookable by the hour, with 192 GB of HBM3e per GPU pushing single-node model sizes higher.
Source: nvidia.com
GPU & AI news
A hand-picked digest of what's moving in GPUs and accelerated AI — new silicon, model releases and the research that matters. Every item has a short write-up and a link straight to the source. Browse the latest below, or pick a topic from the menu.
Latest
Image2026-06-12
Nano Banana Pro brings Gemini-3 image generation
Google DeepMind's Nano Banana Pro sets a new bar for in-image text and precise editing, intensifying the race with GPT Image and FLUX.
Source: blog.google
Image2026-06-05
Z-Image: efficient single-stream diffusion goes open
Tongyi-MAI's Z-Image shows high-quality generation from a compact single-stream architecture that runs comfortably on consumer GPUs.
Source: github.com
Research2026-05-30
3D silicon stacking aims to extend Moore's Law
Researchers report a chip-stacking advance that could keep transistor density — and GPU throughput — climbing for years.
Source: sciencedaily.com
Text2026-05-28
DeepSeek V4 ships a million-token context agents can use
DeepSeek's open V4 release pairs a very large context window with strong reasoning at low cost, keeping open weights within reach of the frontier.
Source: huggingface.co
Video2026-05-20
Waypoint-1.5 runs real-time AI worlds on everyday GPUs
Interactive, generated game worlds now run locally rather than in a datacentre — a milestone for real-time video models on consumer cards.
Source: over.world
Audio2026-04-22
Pocket TTS gives your CPU a high-quality voice
Kyutai's Pocket TTS shows that natural speech synthesis no longer needs a GPU at all, while Kokoro TTS keeps shrinking the footprint further.
Source: kyutai.org
Hardware2026-04-15
AMD pushes Instinct MI325X with 256 GB HBM3e
AMD's CDNA-3 refresh widens the memory lead for large-model inference, sharpening the open-ROCm alternative to NVIDIA in the data centre.
Source: amd.com
Video2026-03-30
Self-Forcing demonstrates real-time autoregressive video
A new training recipe generates video frame-by-frame in real time, a key step toward interactive generative video on a single GPU.
Source: self-forcing.github.io
Text2026-01-07
nanochat: a full-stack ChatGPT clone you can read in a weekend
Andrej Karpathy's minimal, hackable training-and-serving stack is a fast way to understand how modern LLMs are actually built.
Source: github.com
Browse by topic
Hardware
New silicon and accelerators — GPUs, memory and the systems they go into.
2 articles
Text
Large language models, reasoning and agents.
2 articles
Image
Text-to-image generation and editing.
2 articles
Video
Generative and real-time video.
2 articles
Audio
Music, speech and sound.
1 article
Research
Papers and breakthroughs shaping what GPUs will do next.
1 article