VQV Signal

ROBOTS & HARDWARE WATCH TECHNICAL

Post reports GPU and NPU speed and power figures for Liquid AI's d1-2b on AMD Strix Halo

According to a post on X by @jetha, the author ported Liquid AI's d1-2b model to AMD's Strix Halo platform. The post reports 86 ms at about 120 W on the Radeon 8060S GPU, compared with 1.7 s at 5.8 W on the XDNA2 NPU. The same post mentions @KrakowiakK adding an Engines section to his website, asking whether 24 local inference engines running the same weights really behave like the same model. The figures are self-reported in the post, and the snippet is truncated, so test conditions are not shown.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-19T14:51:00+00:00 Detected 2026-11-20T01:05:34+00:00
View original source

According to a post on X by @jetha, the author ported Liquid AI's d1-2b model to AMD's Strix Halo platform. The post reports 86 ms at about 120 W on the Radeon 8060S GPU, compared with 1.7 s at 5.8 W on the XDNA2 NPU. The same post mentions @KrakowiakK adding an Engines section to his website, asking whether 24 local inference engines running the same weights really behave like the same model. The figures are self-reported in the post, and the snippet is truncated, so test conditions are not shown.

AI-assisted summary based on the listed source.

@jetha ported Liquid AI's d1-2b to AMD's Strix Halo, reporting 86 ms at about 120 W on the Radeon 8060S GPU versus 1.7 s at 5.8 W on the XDNA2 NPU. @KrakowiakK added an Engines section to his website asking whether 24 local inference engines running the same weights really behave like the same model. Why it...

Developers shipping on-device AI get a concrete view of the trade-off between speed and power, with the GPU much faster but drawing far more power in this one test. Teams choosing between GPU and NPU paths on AMD hardware, or comparing local inference engines, will want to check whether results hold under their own workloads. Watch for the full write-up and benchmark details, since the snippet alone does not explain the methodology.

Hardware and robotics watchers may want to track whether this becomes a product, benchmark, or deployment signal.

Signal Strength 95% Technical label WATCH Public Interest 45 Category ROBOTS & HARDWARE Reader Depth TECHNICAL

Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.

Public Interest components
Recognizable Entity Score 73 Practical Impact Score 20 Novelty Interest Score 70 Consequence Score 18 Curiosity Score 0 Shareability Score 61

VQV surfaced this signal because it is recent, relevant to LLM Inference, connected to Scobleizer News Pulse (X).