Faster and lighter
This update focuses on performance, especially on AMD graphics cards.
- Much faster replies on AMD graphics cards, especially 8 GB cards: answers about your screen arrive about twice as fast, and long conversations no longer slow to a crawl.
- Better reply quality on NVIDIA graphics cards: the AI now keeps a more precise memory of the conversation, for more accurate and consistent answers.
- Fixed: on some NVIDIA systems with older drivers, the AI could silently fall back to the processor and become very slow. It now automatically switches to a faster mode.
- Faster safety filter that uses less video memory.
Clearer choices
- Tools now show a short notice the first time you turn them on, because they load a second AI model that uses extra video memory.
- One of the GPU-heavy voices is now unavailable on graphics cards with less than 12 GB of video memory, to avoid running out of memory. If you already use it, nothing changes.
- Options that can't run on your PC are now greyed out, with the reason shown.
Better support
- A new local diagnostic log helps me understand performance issues when you contact me. It stays on your PC and is never sent anywhere.
