Feb 16, 2026: For those running an Edgecloud on gaming rigs not dedicated to it.
VRAM can be loaded into your GPU without the AI model actually doing work. Our system just proactively loads a model into your GPU to be ready to catch jobs.
Q: That's new then because it hasn't never done it in the last 6 months. I'm connected now and GPU VRAM usage is slowly increasing. Now, it's maxed at 24 GB in less than a minute.
A: It depends what models get rotated onto your GPU.
Just because memory is filled up, it doesn't mean work is actually being done, it just means that's how the model prepared itself and is idle.
You should never run another GPU application while running Edgecloud.
A deployment is a user actively doing work and reserving your PC.
A job is someone using the on-demand API feature of Edgecloud and the on-demand API request gets routed to idle machines to start doing work. A job is a period of work.
For a deployment, a user is reserving your PC to do continuous work. It doesn't mean they are actually doing continuous work. So they may not use your GPU that hard. and the job is reserving your memory in an attempt to catch a job and actually perform work, it doesn't really mean it's acting on that memory while it's sitting idle.
Q: Just seems weird that my VRAM is locked in a period I'm not actually being compensated for.
A: Think about it like this, someone sends an on-demand APIi request, other people's machines are ready to catch it because they preemptively reserved memory for it. They get the request first and are rewarded because your machine is still loading in the model into your VRAM.
These are large models, they take a lot of time to spin up, we're proactively loading it on your machines so you're ready to catch jobs, YOU have the control to spin them down at anytime by stopping Edgecloud.
Another reason your machine is considered idle is because, if memory is allocated and gputil is 0, allocating memory doesn't really cause GPU to draw power (essentially what you're being compensated/paid for). If the model is loaded on your machine, but not processing any on-demand requests, is it really utilizing the GPU and using energy? GPU = graphical PROCESSING unit, it doesn't cost to hold things in memory.
Thanks for the questions, it can feel a little weird and unfair if you think about it on a surface level. Moral of the story, now you understand why you shouldn't game while using Edgecloud. We're trying to get you jobs.
Let's just say we're working on getting your PCs more exposure for deployments, can't share too much details just yet ; )
Q: What factors into the price per hour the Edgecloud auto sets for you? I've had it pick 0.18, 0.19, 0.21 and 0.22 without making any changes on my end, just wondering what influences its choice?
A: It's probably people with similar specs than you upping their price, which causes your machine to get set to a higher price. If someone with similar specs than you lands a deployment at a certain price, then that sends a signal that causes similar machines to also get set at that price automatically.