Say you go out and buy a couple 2080's and put them into an Ubuntu Server box. Could you use that box to provide GPU support to local computers (either in the same building or other buildings on the same ISP within, say, a mile)?<p>Of course, NVIDEA offers gaming as a service so it must be possible on some level (https://www.nvidia.com/object/cloud-gaming.html).<p>Is there software you could install on a Windows machine that would act like a GPU driver but would really send those commands over the internet to your local GSP (GPU Service Provider)?<p>You could use the same service for other things:<p>- crypto (of course)<p>- cracking hashes<p>- training machine learning models (probably won't get much use of that in your neighborhood, though).<p>Maybe you could even charge by the number of calculations done. Only five months ago someone created a related post on reddit about vectordash (https://www.reddit.com/r/gpumining/comments/86ofw2/rent_out_your_gpu_compute_to_ai_researchers_and/) but that was more for AI than gaming.<p>Now seems like the right time for this, right? GPU prices are crazy and GPUs are in demand for all sorts of applications. So if you could get the same performance by renting a 2080 for you know, $20 a month why wouldn't you?<p>Is there any existing software for this type of application?
This is already a thing:<p>- <a href="https://parsecgaming.com/" rel="nofollow">https://parsecgaming.com/</a><p>- <a href="https://www.paperspace.com/" rel="nofollow">https://www.paperspace.com/</a><p>Those two collaborate to do more or less exactly what you've described. It ends up looking a lot like RDP/VNC, but without the latency.<p>If you take the time to calculate Paperspace's profit margin, you'll see that it's a risky business to be in. You'd have to buy GPUs in bulk, and, even then, you're competing with the same company that's selling you the GPUs.<p>Nvidia doesn't license the most useful API, Nvidia GRID (<a href="https://en.wikipedia.org/wiki/Nvidia_GRID" rel="nofollow">https://en.wikipedia.org/wiki/Nvidia_GRID</a>), with consumer cards. You have to get their datacenter-grade cards (e.g., the Tesla line) if you want access to the most relevant APIs. It's possible to unlock consumer cards for this purpose, but, as you're lacking the respective license, you're liable to get sued. This effectively means Nvidia is two slices of your pie instead of just one. (They can do whatever they want with their own hardware an APIs.)