Squidwrad

True, technically not made for video output, still capable of creating a virtual desktop and thus everything that ensues. this is how cloud gaming works. They take tesla gpus and run the created frames out of the Virtual machine and to your device. So if they can render, then they can technically be ran like gpus. It’s all about how, and I think we can find a way.

Hi, so long story short. I like to make builds that break all taboos and cost so weirdly cheap and perform so well. However, out of all my broken taboos of dual cpu gaming, ddr3 memory 2gbx16 build, purely usb gaming computer; the only thing I’ve never messed with is the gpu. Mainly cause the deals of used gpu are non existent. Gpu don’t scale that well in the used market the same way xeons do and ecc registered memory and even Chinese mother boards. Gpu usually are over priced, rare and stupidly hard to get imported when bought used. Except I forgot that tesla gpus exist. You know that server dedicated line of gpu that use to be paired with Xeon super computers. Yea those things. Guess what, they cheap as hell and they are plentiful. However, I know for a fact you can game on them cause there are videos of benchmarks, there’s vm machines that can use them with a little modding. There’s the fact they have CUDA cores. Only problem, no video output. Vm solves this with vgpu or pcie pass through. Except I’m not running this on a vm, I’m running this on a normal build. Which means I need to find a way to make this render a game, grab the frame buffer, pass it through the pcie and to either integrated gpu such as the p106 card in Linus’ video, or preferred to pass through a dedicated graphics card. Trust me when I say this, the next person to say this is not a gaming graphics card I will hit you with the might of cloud gaming and curse you to tech hell. Now let’s get started on solutions huh? First step, program running. Easy enough just tell it which gpu to run, Windows has a feature for this, second step, frame buffer. This is integrated into all gpus just need a way to access it and this is possible through Nvidia sdk. Now to transfer it to another i/o device. Hard part, see all solutions involve tricking an api into thinking it’s passing the frames through pcie pass through when in reality it’s just passing raw frames over same instance. The second solution involves running code that creates a vnc like server to grab the frames and then spit them out on the actual gpu output. Both solutions require tricking or writing code that there is zero support for understanding. Please someone has to help me solve this. If I can then imma get tech yes city mind blown straight to Valhalla. Email me if you wanna get direct with me squidwrad@gmail.com