amd

Triton 3.9.0 removes tl.make_block_ptr in favour of tensor descriptors, requires Python 3.11 or newer, and lists deprecations for 3.10 including a default change to floating-point fusion.

AMD and Embedded LLM tested five speculative-decoding drafting methods in vLLM on AMD Instinct GPUs, with some configurations reaching up to 2.87x throughput over the non-speculative baseline.

AMD's open-source local AI agent framework for Ryzen AI NPUs gains a terminal-native agent hub, cross-surface confirmation gates, and MCP security fixes.

A ServeTheHome deep dive breaks down CDNA 5, the architecture behind AMD's Instinct MI455X, alongside the ROCm.ai tooling meant to let AI coding agents program AMD GPUs natively.

AMD is buying Toronto-based Taalas, whose chips etch model weights directly into silicon instead of storing them in memory, to strengthen its AI inference lineup.

AMD's Adrenalin 26.6.2 driver brings FSR 4.1 machine-learning upscaling to RDNA 3 Radeon RX 7000 cards using an INT8 model, but frame generation remains an RDNA 4 exclusive.

AMD's 6th Gen EPYC 'Venice' is ramping on TSMC 2nm — the first HPC chip on the node — with 256 Zen 6 cores and 1.6 TB/s memory bandwidth.

Mercury Research data show AMD's EPYC processors claimed a record 46.2 percent of x86 server CPU revenue in Q1 2026, as AMD's data center segment hit $5.8 billion and overtook Intel's comparable unit.

Zyphra's open-weight ZAYA1-8B uses 760M active parameters out of 8.4B total and was trained on a 1,024-GPU AMD Instinct MI300X cluster, narrowing the gap to frontier reasoning models on math benchmarks.

AMD and Intel released the AI Compute Extensions specification, marking the x86 Ecosystem Advisory Group's first joint matrix-multiplication standard since the rivals formed the consortium in October 2024 to fend off Arm.

Cloudflare's Gen 13 platform pairs AMD EPYC 9965 chips with FL2, a Rust rewrite of its NGINX-based proxy, doubling edge throughput while trading L3 cache for cores.

MLCommons' April 2026 inference benchmark round adds text-to-video and vision-language tests while NVIDIA posts a 2.7x jump on DeepSeek-R1 through software alone.