Topic

#Rocm

7 articles on Rocm — news, releases, guides and analysis from the SourceFeed engine.

DeepSeek V4 Flash on One AMD GPU Took Nine Patches

A single 192 GB MI300X now serves the full checkpoint — proof of AMD's hardware edge and its lingering software gap.

AMD and Meta port the single-controller runtime to ROCm, and the fault-tolerance demo is the real story.

Drop-in translators, HIP, and SYCL solve different problems. Choose by binaries versus long-term portability.

The Ryzen AI Halo dev kit offers 128GB of unified memory for local LLMs, but the price tag hurts.

MI355X hits about 80% of B200 throughput at over 2x lower cost, but only after hand-tuning the ROCm stack.

The translation layer can run unmodified CUDA binaries on AMD and Intel GPUs, but its new direction changes the enterprise calculus.

AMD's native ROCm serving stack splits prefill and decode to eliminate head-of-line blocking on Instinct hardware.

The best developer & AI content, delivered. No spam.