@KardinalZyn@Venalis21 In a time of AI and rise of IT-Security this will cost streamers money. You dont need my channel ownership for giving someone 1$.
@gospaceport Hey you, hope you are well! Wendel of @Level1Techs just released a video you might be interested in - See: https://t.co/uXRqY1h1i9 - Its very interesting for PCIe-Signaling / Extension reasons and for using multiple GPUs with solid performance.
@__mharrison__@arvidkahl I dont still dont understand why anyone, ever, would hype or recommend a DGX Spark. Its literally the worst price to performance ratio and not even designed to run local llms ... Its a Dev-Box-Kit/DEV-Platform for science and research for the GB300 (some architecture) ....
@arvidkahl Grab a @JetKvm and a @FrameworkPuter Desktop and build it yourself - The performance for the price and open-ness is really good. There is also no Vendor-locking and Updates that break everything or payment-model per month. Just what you want on it, running it.
@jakkuh_t Last-Video-Answer: Yes replace your Alexa with a local LLM, you can follow @NetworkChuck 's video on this topic with LLM+HOAS, works really good. In about 40 minutes you learn, that there are many levels to achieving this too - https://t.co/735N4zlXSr
@gospaceport You might have better performance with AMD Drivers or Vulkan with an 9060XT, as with ROCm, could be possible. But my script should install the drivers with the ROCm deps with. As the drivers are part of ROCm.
Hey @gospaceport just discovered your content about Proxmox9 and passing nvidia cards into LXC-CTs, its awesome, love it! Question: Do you plan on making content on the Ryzen AI APUs? I would love to see if you use, CPU or NPU or iGPU with Raedon or ROCm drivers in a CT (1/2)
@gospaceport Hey @gospaceport - Im sorry for the delayed answer, my APU is classified as Strix Point and yes this works also to my knowledge for Strix Halo (Max+ 395), though i dont have any here to test this myself, the architecture and software-stack is identical.
@gospaceport (2/3) version was for the old stable ROCm 6.4.4, after i managed to get that working i tried the "then 7.0.2 Preview" only to realize it took me that long, that 7.0.2 got out of preview-state and 7.1.0 was released. Tried that too, worked and now i have fully automated it ...
@gospaceport (1/3) Hey @gospaceport thanks for the hint with Donato, sadly it wasnt that helpful for Proxmox+LXC and mixed gpu setup. BUT i finally, after weeks of research and testing, solved it! It was an very exhausting experience until I finally had a breakthrough. My first working ...
@E3DOnline@OfSecondson@BambulabGlobal@DBNozzles 4/4 You have to understand what you are doing, this looks like actively lying and deceiving customers for a nozzle that costs 127Euros, while i can get a COMPLETE SET of .2+4+6+8 nozzles in hardened steel for 118,xx€ from the same Bambu Lab Store ... @E3DOnline@BambulabGlobal
@E3DOnline@OfSecondson@BambulabGlobal@DBNozzles 3/4 The setup of profiles is since June a nightmare, you only read "later" included, WHEN is later, define a clear ETA and deliver on this. You also cant just go around and tell customers "in theory its almost 50% faster" when in real-world there is no proof of that.
@E3DOnline@OfSecondson@BambulabGlobal@DBNozzles 2/4 I tested a lot of models just in the slicer to validate the number and even the model in the video. Most of the time the ObXidian nozzle is only 10-20m slower at a 2-5 hour print or at most 35 MINUTES fast on a 25,8 HOURS print, thats far from 50% faster and you know this...