Remix.run Logo
▲ lrvick a day ago

> Hoping AMD steps up to the plate

Nvidia has nothing close to the Strix Point, Strix Halo, Gorgon Halo, etc.

My GPD pocket 4 handheld can run Qwen 3.8 27b only because of AMD making embedded mobile inference cheap.

▲bornfreddy a day ago | parent [-]

> Nvidia has nothing close to the Strix Point, Strix Halo, Gorgon Halo, etc.

Is there a reason why DGX Spark doesn't qualify? (genuinely curious)

▲9x39 20 hours ago | parent | next [-]

Having such a nice GPU setup they may just not have known.

The Nvidia DGX Spark is superior to the AMD equivalent (Strix Halo) but cost more per unit.

The Spark comes with 400GB networking which is huge if you can afford more than one and want to run a tensor.

https://build.nvidia.com/spark/connect-multiple-sparks

Servethehome did 8 of them:

https://www.servethehome.com/big-cluster-little-power-the-8x...

AMD Strix options had more low cost builds although not by much anymore, and some have eGPU Oculink ports but no 400Gbps networking.

https://www.bosgamepc.com/products/bosgame-m5-max-ai-mini-pc...

Spark can network, Strix optionally has 192GB ram and external GPU flavors.

Macs and GPU builds are better ROI at these prices, however.

▲lrvick a day ago | parent | prev [-]

The RTX spark is a comparable thing that runs in laptops now which is a TIL for me. Still requires proprietary drivers though which impact trust and sovereignty to a point that I would never buy one though.