ASUS Ascent GX10 adds 128GB config for local inference of up to 200B models
gnukeith · x · 2026-10-02
ASUS announced new 64GB and 128GB variants of the Ascent GX10 desktop AI supercomputer, both powered by the NVIDIA GB10 Grace Blackwell chip with up to 1 PFLOP of AI performance.
- The 64GB model runs 30-35B models for agentic workloads and can infer up to 100B locally
- The 128GB model can run inference on models up to 200B parameters
- ConnectX-7 200GbE networking enables clustering: two 64GB units pool to 128GB / 2 PFLOP, while four 128GB units scale to 512GB / 4 PFLOP
- Ships with OpenShell secured containers and sandboxes for private, on-device inference of long-running autonomous agents
More from Infra
- Cloudflare Durable Objects now survive client disconnects for long-running agents — threepointone · 2026-10-03
- SemiAnalysis: Nvidia's custom NVHBM frees ~25% more compute die area on Feynman — zephyr_z9 · 2026-10-03
- llama.cpp PR Halves Indexer Score Memory for Qwen Flash, Cutting VRAM Use — jacek2023 · 2026-10-03
- gufo-Qwen3.6-35B hits 3095 tok/s prefill, 190 tok/s decode on Strix Halo — nubela · 2026-10-03
- Garage server farms return: GPU and power shortages reverse the AWS era in Palo Alto — bookwormengr · 2026-10-03
- AI maxi calls GPU price hike a bubble peak, plans to buy cheap cards after burst — AIFlow_ML · 2026-10-03