Faster AI at Home: New Hack Makes Chatbots Talk Quickly

If you’ve ever felt frustrated waiting for an AI chatbot to catch up while you type, a new trick could make it feel much faster. A Reddit user figured out how to speed up the Qwen3.8 AI model—used for coding and writing help—by about 50% on older gaming PCs with two graphics cards. The hack works like a smart closet organizer: instead of stuffing all your shoes into one pile, it keeps the ones you wear most often right at the front. For AI, that means caching (saving) the most frequently used parts of the model in the graphics card’s memory, so it doesn’t have to dig through the whole thing every time you ask a question.

The trick works best for short to medium conversations—like asking for help with an email or a quick coding fix. Long documents or deep technical queries still slow down a bit, but the speed boost is noticeable. The user tested this on two older RTX 3090 GPUs (the kind gamers used years ago) and got the AI to respond 25-29 times per second instead of 17. For context, that’s the difference between waiting a few seconds and a few milliseconds for a reply.

Here’s the catch: this isn’t something your average person can just download and use. You’d need to be comfortable tweaking software settings and have a very specific setup—like two graphics cards and a lot of RAM. But for tech enthusiasts or small businesses running AI locally, it’s a game-changer. It’s like upgrading from dial-up to broadband, but for your AI assistant.