Got it during the sale. Tried to vibecode a clean UI personal AI agent. looks good but Oof In function. Want to use that 48gb of fast ram but my brain just isn’t clicking. what are you guys running? I’m not a programmer by any means. CLI intimidates me.
This will be my first Windows on ARM machine, I saw the ram and the price and the brain shut down. I started playing around with LLMs, I'm assuming that this can fit a model but obviously won't be as fast as discrete GPUs. What else can I expect? Coming from a MacBook 14 M3 Pro 18GB. Would like to do a bit of everything on this machine, but also make use of that delicious RAM. Also why so (relatively) cheap?
EDIT: it’s here! will do all the windows stuff. Gonna try steam, KCD2, and try to run a less compressed 27b model (any suggestions?)
This possible? It's in a SFF case, so would be like an AI buddy (hopefully can easily upgrade and not wipe it)
wish I had the time to verify, but using the LACT app, which you can find on the desktop App Store, you can enable AMD overclocking (it’ll show an error message but the app will work on restart) and in the app slightly overclock and undervolt. mine is at -69. Games would crash any higher so it tells me that it’s doing something.
when I hit Apply, the screen flickers a bit (on a VRR OLED display if that has any effect) but goes away on restart.
Ive noticed in KCD, the VRAM actually does hold the new value (it reports 1/2 the total, so 1200) and I gained 3-4 fps, which somewhat helps LSFG stay at 60fps or 40s without and on ultra high at “3K” ( 2880 × 1620) very unscientific and ymmv
just FYI. one thing I loved about the steamdeck was the tinkering community. Hoping that the steamcube gets some of that same treatment.
(and I know it’s kinda funny to squeeze performance out of machine that is more expensive than faster PCs, but I like the cube)