It would be interesting to revisit this review for the forthcoming Strix Halo. Particularly, for ML tasks on relative large memory sizes which are prohibitively expensive for GPUs.