GMKtec EVO-X3 AI Workstation: Lemonade, Benchmarks, and specs

GMKtec's EVO-X3 AI workstation, featuring AMD Strix Halo, is shipping with delays and questionable VAT documentation, according to forum user ewook on Level1Techs. The user also questions whether adding an R9700 GPU via OCuLink improves prefill speed for large AI models split between the GPU and system RAM, noting a lack of published benchmarks.

ewook https://forum.level1techs.com/u/ewook 21 Yes, seems to be. I guess there’s two kinds of coordinations happening, either from Germany that I doubt they have time to fill and from factory. Took almost a week for my fedex slip to be created, and additional three/ fours days for the package to be handed over to fedex. I guess the lead time is a bit shorter from Germany. There’s also a questionable shipping document attached for VAT/taxes.. yes, I saw that… not happy about how they are handling it… in the store it clearly said there was stock, and that the stock was from Germany So does adding a r9700 over oculink usb/pcie to the strix halo actually improve prefill speed when the model is split or not? say a large model that takes up ~110gb of memory total including context, split filling up the r9700 and spilling the rest to the halo’s ram. i havent been able to see anyone actually showing the benchmarks on this. what about a ~140-150gb model since the r9700 makes it a 160 total?