VQV Signal

ROBOTS & HARDWARE WATCH TECHNICAL

Base Compute says its BaseRT local inference engine now supports AMD chips

According to a post on X by @basecmpt, Base Compute says its BaseRT local inference engine now supports AMD chips, using both the GPU and NPU. The company claims on-device LLM performance rises by up to 664% on Ryzen AI Max+ 395 and Ryzen AI 9 chips, after work with AMD and Dell. It says the NPU alone added 41% while using 2–3x less energy per token. In one example, an on-device email agent handling 50 emails with attachments reportedly takes 17 seconds instead of 2–3 minutes.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-20T02:45:00+00:00 Detected 2026-11-20T19:02:45+00:00
View original source

According to a post on X by @basecmpt, Base Compute says its BaseRT local inference engine now supports AMD chips, using both the GPU and NPU. The company claims on-device LLM performance rises by up to 664% on Ryzen AI Max+ 395 and Ryzen AI 9 chips, after work with AMD and Dell. It says the NPU alone added 41% while using 2–3x less energy per token. In one example, an on-device email agent handling 50 emails with attachments reportedly takes 17 seconds instead of 2–3 minutes.

AI-assisted summary based on the listed source.

Base Compute says BaseRT now supports AMD chips, using the GPU and NPU to raise on-device LLM performance by up to 664% on Ryzen AI Max+ 395 and Ryzen AI 9 after work with AMD and Dell; the NPU alone added 41% while using 2–3x less energy per token. In its example, an on-device email agent handling 50 emails with...

For developers and businesses running LLMs on AMD Ryzen AI hardware, the claims point to faster local inference with lower energy use per token. These figures come from Base Compute and its own example, so independent benchmarks would be the thing to watch. The source text cuts off before the post's full explanation of why it matters for companies, so that wider case is not covered here.

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

Signal Strength 88% Technical label WATCH Public Interest 57 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 56 Novelty Interest Score 70 Consequence Score 34 Curiosity Score 16 Shareability Score 67

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