I continue to be impressed with the utility of the ASUS ROG Flow Z13 (Strix Halo) laptops. The following is a copy of a conversation with ChatGPT, which walked through the successful clustering of 2 of those machines:
https://com-pute.com/nick/Clustering_Strix_Halo.mhtml
The clustering process was much simpler than it was with the Asus GX10s, which I covered here:
https://aibynick.com/thread/38
For the Strix Halo laptops, I simply connected a generic $17 Acer 40Gb/s USB/Thunderbolt cable from Amazon. That was much less expensive and less time consuming to configure than the $179 ASUS Ascent GX10 0.4m 400G QSFP112 DAC Cable used to cluster the GX10 machines.
Of course the GX10 solution is faster, especially at loading models and for time to first token, but for my personal needs, the Strix Halo laptops perform really well, and I absolutely love having a portable laptop solution for inference, which can handle some genuinely serious work loads - and as I've posted elsewhere, the output generation speeds of some select models are actually faster on Strix Halo than on the GB10 architecture.
For clients who need a serious permanent setup, especially for clustering, the GX10 will still be my first choice, but those Strix Halos are no slouch for personal work.
Unfortunately, both machines went up in price by about $500 during the past week. The Strix Halo laptops now cost $2,972.73 for used/like-new models on Amazon. The GX10s now cost $3,999.00 new - so there's still about $1000 dollars difference.