Messing around with ASM2464PD symbol and footprint
-
Updated
Oct 6, 2025
Messing around with ASM2464PD symbol and footprint
Exact LLM inference engine written from scratch for AMD Strix Halo (gfx1151, ROCm/HIP). Qwen3.8-Flash-Next on one node or two replicas; DeepSeek V4.1 Flash across two nodes over USB4 with an optional RTX PRO. Speculative decoding bit-identical to greedy, up to 255 tok/s on coding chat. MiniMax and GLM next.
Two-rank tensor-parallel LLM serving over USB4v2 between AMD Strix Halo hosts: a custom Linux thunderbolt stream transport, self-healing link, and gufo TP2 integration
Linux 7.3 patch series that turns the native AMD USB4 controllers of two Strix Halo (Ryzen AI Max+ 395) machines into a direct DMA interconnect for distributed LLM inference: USB4STREAM, GPU-to-GPU DMA-BUF transfers and safe recovery. ~43 Gb/s, 8 µs round trip.
SerDes validation framework with advanced visualization, multi-protocol support (USB4, PCIe, Ethernet), and interactive dashboards. Production-ready platform for high-speed signal analysis and compliance testing. Open-source alternative to commercial validation tools.
Dynamic PCIe Gen4/Gen5 retrainer, clock stabilization & state manager for NVIDIA Blackwell (RTX 50xx) eGPUs on Linux (KDE Plasma 6).
Proxmox cluster using 3 servers interconnected by USB4/Thunderbolt
Plug-and-play eGPU for SteamOS handhelds: auto-switch Game Mode, per-GPU shader caches, safe disconnect. Decky plugin + Desktop app.
Optimized dual AMD Strix Halo (gfx1151) vLLM MoE inference: TP=2 over USB4, tuned int4 MoE kernel, ~8us interconnect, serialized serving
General RDMA transport between two Strix Halo hosts over Thunderbolt/USB4 NHI DMA rings: a native software-verbs provider with a zero-copy data plane, plus the USB4STREAM backport it builds on
Switch to your eGPU and unplug it safely on Linux handhelds and laptops. Game Mode panel, desktop app and CLI. NVIDIA and AMD, Thunderbolt/USB4 and OCuLink.
Fail-closed USB4 link qualification and paired Strix Halo model operations
Hardware-neutral orchestration for heterogeneous LLM inference: placement and routing across a discrete GPU and AMD Strix Halo nodes over USB4, measured with DeepSeek V4.1 Flash and Qwen3.8-Flash-Next. Early development.
Build your own local AI server with Proxmox, Ubuntu, NVIDIA GPUs, Thunderbolt/USB4 eGPU docks, Ollama, Open WebUI and AI agents.
Heterogeneous Compute Cascade (HCC) — distributed 400B-parameter MoE inference across dual AMD Ryzen AI MAX+ 395 'Strix Halo' workstations via USB4.
x86 open-source portable router for desktop environment remote access
Cross-platform file transfer over a wire with no setup
Automatic Handheld / Docked / eGPU mode switcher for Windows gaming handhelds: power plans, Xbox full screen, iGPU VRAM, apps & Game Mode eGPU throttle fix. | Windows 掌机模式自动切换(掌机/底座/外接显卡):电源计划、Xbox 全屏体验、核显显存、启动程序与显卡坞降频修复。
To associate your repository with the usb4 topic, visit your repo's landing page and select "manage topics."