12gb required VRAM, eh.

well don't *I* feel the fool for becoming at all interested in this. ah, well.

Someone posted on Reddit about G-Assist automatically undervolting their GPU: https://www.reddit.com/r/nvidia/comments/1jkfkib/project_gassist_undervolt/

I don't know if it turned out to be stable, however I'm impressed that not only was it able to do that, but suggest it to the user as part of a general exchange about their power and temps. It even provided a chart showing the difference in the frequency/voltage curve before and after the change.

This comment was edited on Mar 27, 2025, 02:19.

Beamer wrote on Mar 26, 2025, 11:22:
As someone that's long past the years where I'd do any changing of graphics card settings outside of in-game, and who doesn't even bother with in-game half the time, it feels like this would probably be helpful to me.

Brainstorming with it is going to be the most useful part.
Yeah you could do that with a general AI like right now
but it doesn't have access to the performance metrics and it isn't a model specifically trained for that task.

This is going to be great for when I have 5 options of anti-aliasing technologies and I have no idea what they do or mean.

As someone that's long past the years where I'd do any changing of graphics card settings outside of in-game, and who doesn't even bother with in-game half the time, it feels like this would probably be helpful to me.

AI will help you find the perfect G-spot.

GeForce Experience, but with AI? No thanks.

Amazing how they can make suggested default settings, something that has been around for a long time, suddenly feel toxic by adding the word 'AI'

Not that I ever used the Geforce Experience defaults, as I didn't like the app to begin with and they didn't seem to prioritize what I considered to be the right things in game settings, so I could get a better looking and playing combination of settings by starting with the game developer's suggested settings and modifying from there.

jacobvandy wrote on Mar 25, 2025, 20:01:
As mentioned in the blurb, this is a custom small language model (SLM) designed to run locally on your GPU, so it does not require an internet connection to function. Whether or not they're collecting data related to how it's interacting with their other software is another matter, but I highly doubt they're capturing every input and response like a cloud-based AI service would.

They mention it's based on an 8B-sized large language model (LLM) which is normally operated on about 4-6GB of VRAM, but since this is meant to be run alongside your game and they don't need all of that 'knowledge' a generalized model would include, I'm guessing they trimmed that down to 1-2GB or possibly even less. So that gets loaded in and waits for you to enter a query, at which point it will momentarily divert GPU resources to generating the response.

Apparently that can include charts or graphs of frametimes, system latency, power consumption, and other diagnostics like a lot of folks use third-party tools to monitor, which actually sounds pretty nifty... Except with this you can just look at that and then tell it to make changes for you, such as adjusting the frame limiter or power target or enabling/disabling V-Sync or G-Sync (perhaps even while still in-game?). If they can expand that to include checking or changing DLSS status, presets, and overrides, that would be a HUGE quality of life change compared to what is usually involved with that now (i.e. toggling the dev mode on-screen indicator via registry key and re-launching the game). It'd also be pretty great to be able to change graphics settings on the fly, without accessing the menu, but that might require some plugin support from the game developer side.

It's never about the model used, it's about what's wrapped around it. Sure, the smaller the model (and quantization), the lower the mem and cycles required. But that's not the issue here. It's how they'll use it going forward. I've worked on some agentic stuff that can take a screenshot of your desktop (many times per second), feed it to a model, and it will understand everything that you're doing and actually know how to reproduce it. And this is all happening in the background without the user's knowledge. But if what they're pitching sounds good to you, go for it. It might be a bit convenient for a while, but keep an eye on those updates and the EULA

jacobvandy wrote on Mar 25, 2025, 20:01:
As mentioned in the blurb, this is a custom small language model (SLM) designed to run locally on your GPU, so it does not require an internet connection to function. Whether or not they're collecting data related to how it's interacting with their other software is another matter, but I highly doubt they're capturing every input and response like a cloud-based AI service would.

They mention it's based on an 8B-sized large language model (LLM) which is normally operated on about 4-6GB of VRAM, but since this is meant to be run alongside your game and they don't need all of that 'knowledge' a generalized model would include, I'm guessing they trimmed that down to 1-2GB or possibly even less. So that gets loaded in and waits for you to enter a query, at which point it will momentarily divert GPU resources to generating the response.

Apparently that can include charts or graphs of frametimes, system latency, power consumption, and other diagnostics like a lot of folks use third-party tools to monitor, which actually sounds pretty nifty... Except with this you can just look at that and then tell it to make changes for you, such as adjusting the frame limiter or power target or enabling/disabling V-Sync or G-Sync (perhaps even while still in-game?). If they can expand that to include checking or changing DLSS status, presets, and overrides, that would be a HUGE quality of life change compared to what is usually involved with that now (i.e. toggling the dev mode on-screen indicator via registry key and re-launching the game). It'd also be pretty great to be able to change graphics settings on the fly, without accessing the menu, but that might require some plugin support from the game developer side.

Yeah it's actually a really nice use of the technology, which is relatively easy to see if your first reaction isn't uninformed pantshitting.

As mentioned in the blurb, this is a custom small language model (SLM) designed to run locally on your GPU, so it does not require an internet connection to function. Whether or not they're collecting data related to how it's interacting with their other software is another matter, but I highly doubt they're capturing every input and response like a cloud-based AI service would.

They mention it's based on an 8B-sized large language model (LLM) which is normally operated on about 4-6GB of VRAM, but since this is meant to be run alongside your game and they don't need all of that 'knowledge' a generalized model would include, I'm guessing they trimmed that down to 1-2GB or possibly even less. So that gets loaded in and waits for you to enter a query, at which point it will momentarily divert GPU resources to generating the response.

Apparently that can include charts or graphs of frametimes, system latency, power consumption, and other diagnostics like a lot of folks use third-party tools to monitor, which actually sounds pretty nifty... Except with this you can just look at that and then tell it to make changes for you, such as adjusting the frame limiter or power target or enabling/disabling V-Sync or G-Sync (perhaps even while still in-game?). If they can expand that to include checking or changing DLSS status, presets, and overrides, that would be a HUGE quality of life change compared to what is usually involved with that now (i.e. toggling the dev mode on-screen indicator via registry key and re-launching the game). It'd also be pretty great to be able to change graphics settings on the fly, without accessing the menu, but that might require some plugin support from the game developer side.

Not being a power user, some of it sounds like it could be useful to me. My main question is: does this require phoning home to NVidia at any point? If so, then it's a "NO" with a capital "FUCK".

Probably a kewl tewl for n00bs but useless as tits on a bull for pros like me!

Hmmmm...a buzzword laden piece of bloatware controlled by a corporation with system level access to your machine.

Sounds totes legit and that it could never be compromised or used against users.

.

G spot assist??

Who is this for? Is it going to tell you to keep your drivers up to date, run antivirus, update windows, don't download strange files from the internet, and tell you for best performance in games turn off all other programs?

It sounds like your typical "give an inch, take a mile" scenario. It would be tweaking your settings today; keylogging and browser/app use tomorrow. With the added bonus of spewing ads at you where no ad blockers exist

El Pit wrote on Mar 25, 2025, 12:19:
G-Assist - Nvidia helps you to find the right spot where "gaming" becomes real fun!

Still waiting for G-Dominate, a service where the AI plays the games for you. Maybe in a few months, Nvidia?

After ya said G-Dominate I went in a totalllllllly different direction then you did.

My safe word is Pineapple.

I guess somebody over there looked at GeForce Experience and said "hold my beer".

Disk Space Required:

› System Assistant: 6.5 GB
› Voice Commands: 3 GB

GPU:

› GeForce RTX 30, 40, and 50 Series Desktop GPUs with 12GB VRAM or Higher

12GB of the their precious VRAM ha! sorry 5060's.

and who the hell needs voice commands for their video card?!? I can't wait to see what nefarious shiz this really does.