Nvidia’s Project G-Assist is one of the few times in the last year or so that any of the big gaming companies have name-dropped AI and I’ve actually seen the vision. A chatbot/diagnostic tool that reads your system and gives advice on how to better optimise it seems intuitively like something that could actually be useful
A lot of the problem with games having larger and larger settings menus is that it can all feel a tad too granular. Do you flip one option to get 1 fps higher, or do you go for a slightly nicer texture? Are all your components running at the speeds and temps they should be, and how would you know if not? Tinkering is one of the best parts of owning a PC, but that joy lies in the option to do so—not the obligation.
Well, if you have a desktop RTX 20-, 30-, or 40-series card, you can download G-Assist straight from the Nvidia app to help out. Or at least that’s the hope. The program is in version 0.1.9, and notes that it’s pretty experimental in both the Nvidia app and in G-Assist itself. Should Nvidia continue to work on it, which it surely will, it will likely get better with time. The company has already planned a hackathon to work on plugins for the software, but ended up extending the deadline.
As it is technically available to install, I thought I’d put it to the test.
Once installed, I need only press ‘Alt + G’ to bring up a chatbox in the bottom right of my screen. The chatbox popped up on my main monitor and becomes transparent when playing games.
While opening G-Assist is fairly easy, I did run into some trouble closing it. Minimising removes it from the screen, but it has no icon on the taskbar, and it’s pretty much always running. You can’t shut it off easily, and even closing the app via Task Manager didn’t see it gone for good. It automatically opened itself up again right after.
The Nvidia App regularly takes up between 100 MB to 150 MB of RAM usage, which is only about a third of what Slack demands right now, and about 15% of what Google Chrome demands (don’t judge me, I enjoy having lots of tabs open). However, this spiked up to 1000 MB and around 10% of CPU in operation. Built on Meta’s Llama 3.1 LLM, it runs locally (on device), which would explain some of how resource-intensive it is. Though you would expect it to be harder on your GPU than your CPU, seeing who made it. It just has to function well enough to justify the usage.
I thought Black Myth: Wukong to be the best first test of G-Assist, being a good benchmark for showing off Nvidia tech with DLSS and a game that benefits from smooth frames per second.
After asking G-Assist to optimise my rig to run Wukong, it assured me that “Black Myth: Wukong has been optimised successfully. This will take effect from the next game launch”. Finding this answer a bit vague, I asked it what optimisations it did, to which it replied, “Based on the provided information, I did not perform any optimisations”.
Oh, okay. Great. Thanks, G-Assist.
What the chatbot did do, however, was tell me to update to Nvidia GeForce driver 576.40. This is normally solid advice—up-to-date drivers make for a happy gaming PC—except for the fact that I had installed these exact drivers at the start of May. This outdated guidance was inadvertently a pretty good reminder to get driver 577.00, which launched a few days ago. So, I suppose a happy reminder in a roundabout way..
As a test, I told G-Assist that Wukong was running choppy, but it refused to fix my imaginary problem. G-Assist kept telling me I wasn’t running a game at all. After repeated failed attempts at convincing the bot that I was, in fact, playing Black Myth: Wukong, my spirit was broken and I gave up momentarily.
Third times the charm, I thought. After wandering around a bit and playing the game a little, I said I was running Black Myth Wukong, and it told me, “Sorry I missed that! Can you please repeat?” The idea of a bot having an excuse like ‘missing’ something I’ve typed makes my head spin, but I skipped over that for now and told it ‘My game is running choppy. Could you fix that?”
The sage advice G-Assist gave me? “Model load error. Can you please repeat?” Classic. Ironically, moments after this, the game did start to perform poorly, and I had to use Google instead.
Another reason I chose Black Myth: Wukong specifically is that it’s an Nvidia-optimised game. It has good Multi Frame Generation (MFG) support and was on the cover of a set of new Nvidia drivers when it launched. It’s also become a bit of a poster child for Nvidia because it’s both popular and looks great in motion. The speed of the game makes a justification for DLSS, specifically DLSS4.
In Red Dead Redemption 2, after failing to see that I was playing the game and refusing to give any advice, the ‘Alt+G’ shortcut simply stopped working. Zero for two now. This bug was so persistent that I had to shut down G-Assist in Task Manager and boot it up again.
This didn’t fix my problem as it got stuck ‘initialising G-Assist’, and it failed to initialise over the 15 minutes I spent waiting. It was hogging up 500 MB of memory and around 7% of my CPU this whole time. Immediately after booting up the chatbot while Playing Red Dead Redemption 2, I noticed some stuttering but I’m unsure if this is directly related. RDR2 can be a bit of a finicky game and it definitely doesn’t like when you move out of the game and into a different tab.
After resetting the Nvidia app, it prompted me to update to version 0.1.13, which seemed to fix my problem of it not initialising.
Booting up Baldur’s Gate 3, G-Assist once again told me it had made optimisations it hadn’t and recommended old Nvidia drivers. It didn’t at any point mention specific graphics settings in Baldur’s Gate 3, though it did say “adjust the in-game graphics settings to balance performance and visual quality”. Which settings to adjust is anyone’s guess. I had the audacity to ask it which settings I should change based on my rig, and it told me “to help you with this, you need to be running a game”. Zero for three.
So, I wouldn’t use it in games right now. However, it can access system information quickly and can chart CPU and GPU utilisation. This is neat, but once again, held back a little. This charting only happens over a small period of time and includes the ramping of resources to run G-Assist. That’s when it works at least. Also it’s not an entirely novel feature—a litany of other third-party applications can do this. Oh and Task Manager.
The chatbot offers up a few prompts on boot, like the charting option above or asking it to tell you more about your rig. One other option, “What is DLSS 4?”, took one minute and 56 seconds for the chatbot to answer. Its reply: “Sorry I missed that! Can you please repeat?”.
I asked this same question a few days prior and got a sufficient, if a little self-referential, answer. “DLSS 4 uses transformer models to generate new pixels by analysing the entire frame, not just localised context, unlike previous versions. This allows for more accurate and detailed image reconstruction, resulting in improved performance and image quality”. This answer requires you to know the basics of DLSS, but I guess it’s not wrong.
From my time testing, both when it worked as intended and not at all, all I can think is that almost all of my problems were more efficiently and accurately answered by Google. After resetting my PC, updating the app, turning everything off and on again, I just couldn’t get it to run stably. The idea of optimising your rig in a moment or diagnosing technical problems is pretty great, but all I got using G-Assist was more technical problems. I didn’t even get a lousy shirt with it.
Best gaming rigs 2025