AI isn't a one-platform future. As models, frameworks, and infrastructure evolve, developers need more choice. AMD VP of AI Software @AnushElangovan and AI futurist Zack Kass discuss why open software, developer flexibility, and ecosystem choice will shape what's next.

🎉 Congrats to @Alibaba_Qwen on Qwen3.8-2.4T-A95B, one of the largest open-weight models released to date. 2.4T params, 95B active, 512 experts. Day-0 support in vLLM, verified on @NVIDIA and @AMD hardware. A ready-made 4-bit checkpoint per vendor, both out of the box: Inferact/Qwen3.8-2.4T-A95B-NVFP4, 1.32 TiB, one NVIDIA 8xB300 node Inferact/Qwen3.8-2.4T-A95B-MXFP4, 1.45 TiB, one AMD 8xMI355X node No conversion, no calibration on your side. Just vllm serve. Thanks to @Alibaba_Qwen for the weights and the collaboration, @NVIDIAAI and @AIatAMD for the joint kernel engineering, @inferact for the quantized checkpoints and vLLM integration, @digitalocean and @togethercompute for early testing, and the vLLM community. 🙌 🔗

AI isn't a one-platform future. As models, frameworks, and infrastructure evolve, developers need more choice, not more lock-in. Watch AMD VP of AI Software @AnushElangovan and AI futurist Zack Kass discuss why open software, developer flexibility, and ecosystem choice will shape what's next. Watch the full fireside chat on 8/13.

Real operators in AMD Strix Halo local AI: @Italianclownz — ROCmFPX formats/kernels + Hermes on Strix @pupposandro / @luceboxai — DFlash/PFlash/DSpark speculative engines @ciruai — gfx1151 Laguna quant recipes that hold up @dcapitella — toolboxes, benches, host tuning (the on-ramp; aka kyuz0) @hec_ovi — Compose packs with measured prefill/decode + ROCmFPX/vLLM/3D @NathanW1014 — strix-halo-vulkan llama.cpp fork @1bitlabs — multi-box research; Hermes + Laguna agent economics @PromptInjection — Strix fine-tune guides (Win/Linux) @soloish90 — ops: drivers, toolbox freeze, prefill bottleneck honesty @DevOuterReaches — Moonshine: Kimi K3 expert-streaming on one 128GB box @lhl — measurement science / Strix-vs-Spark honesty (quiet on X) @Level1Techs — Proxmox + ROCmFP4 + MTP depth guides Not really on X — GitHub instead (sorry if you are actually on X): deseven — + Homelab Discord · Hal0 — full Strix home AI appliance · julianmb — DeepSeek Flash ROCmFPX+DSpark packaging · daimonionnn — Windows ROCmFPX server · ayysasha — dual Strix vLLM over USB4 · ianbarber — Claude Code gfx1151 setup skills · urbanswelt — IOMMU-off host tuning benches · lhl repos —

spur your compute: . Multi-thousand GPU clusters now run on SPUR it can also control k0s so workloads can use k8s or spur on unified management plane. We will be ramping up on very large CPU fleets too. Give it a try, submit an issue or PR.

my team figured out how to run Kimi K3 on @AMD MI355X at 952 tok/s/node and 118 tok/s single stream. 3.8x the aggregate throughput/node and 1.3x the single stream decode of B200 and beat B300 on performance per dollar: 48 vs 33 tok/s/$ details in reply

🚨 BREAKING: these engineers figured out how to serve Kimi K3 on @AMD MI355X at 952 tok/s/node and 118 tok/s single stream! this crushes B200 by 3.8x in aggregate throughput/node and 1.3x in single stream decode + beats B300 on performance per dollar (48 vs 33 tok/s/$) See how in the thread.