Since there is a good chance that you won't get a support from these guyes, I am writing my experience. I have tested quite a bit of models when I had Evo-X2, which is returned for a defect and their customer support went radio silent with my (alomost) 2,000USD.
I first had to install 1.04 BIOS iirc, which has native option in the BIOS to select 96B GPU memory allocation. I used Ollama and most 6~70B models in Ollama's website run without issues except the couple latest models including llama4:scout. That seems to be a bug in the engine. When something is claimed to be supported by a model runner(LM Studio, Ollam, etc), that usually means it is supported on CUDA(Nvidia) architecture, and AMD's driver for ROCm is also kind of garbage at this point for this chipset. Everything worked much better in Linux btw but not much else to say.
When I contacted them through the email for instability issues I was having, they asked me to use 64GB because that is the most balanced setting, whatever that means. I told them that I wouldn't have bought it knowing 64GB is the maximum I can use for GPU and asked for refund if they cannot guarantee that it will work with 96GB. Then they started a refund processing. Not a very confidence inspiring move if you ask me.
In any case, good luck. When the machine was working, it was pretty cool but I just had so many issues, and now figthing for my refund with the customer support after I did everything they asked for. Now I am out of both my money and the machine that I returned, which they claimed not received, even after I have provided a document from UPS that someone signed at their return location for the delievery. They may just be slow on responding so I am just frustrated at the moment, but that was enough for me to decide that I will not recommend anyone buying from them based on my experience.
