Tech news in 3 minutes

NVIDIA Announces Expanded Jetson Thor Lineup with Mid-Range T3000 and T2000 Modules

3 d ago

AMD Instinct MI350P PCIe GPU trades shows specs comparison NVIDIA H200 RTX Pro 6000 inference memory capacity FP4 FP6 performance. The AMD Instinct MI350P accelerator, leveraging CDNA 4 architecture, has been prominently featured at Dell Tech World, HPE Discover, and Computex 2026, drawing significant attention from the AI and data center markets. Key specs position the MI350P against NVIDIA’s H200 NVL (141GB vs 144GB HBM3E memory) and the RTX Pro 6000 Blackwell Server Edition (96GB GDDR7). The MI350P’s advantage lies in modern architecture supporting FP6 and FP4 numeric formats, delivering higher dense compute performance for inference workloads, while NVIDIA’s Blackwell card lacks published FP6 figures and uses lower memory capacity. The MI350P is essentially half the OAM MI350X in compute, memory, and power—a design choice to fit a 600W passive-cooled PCIe card without video outputs. Its decode capabilities also support video and image feeds. AMD’s published dense performance figures at FP4 and FP6 exceed those of both NVIDIA competitors, and the ability to use lower-precision formats (MXFP6) allows fitting more model parameters into the card’s memory footprint. The accelerator is already appearing in systems from Dell, HPE, and other OEMs, signaling strong ecosystem adoption for large-scale AI inference deployments.