I'm looking for a graphics card for a modern gaming PC that will also handle local LLM workloads. I'm not loyal to AMD, Intel, or NVIDIA—value, availability, features, and reliable driver support matter most. My budget is preferably below RM2,000, or roughly US$400 before considering import taxes.
I've used a GTX 1050 Ti, an R7 260X, and an RTX 3050 8GB. The older cards still work reasonably well for general use, but newer games and software are increasingly rejecting them or running poorly. The RTX 3050 is acceptable for gaming, but I'd like higher settings and don't want to buy another card that may become troublesome in a few years.
AMD cards seem widely available and competitively priced where I live. NVIDIA has appealing features and may be better supported for some AI software, while Intel could be worth considering for generative workloads, although I'm unsure about driver and application compatibility. Between 8GB and 16GB models, which options offer the best overall value for gaming and local LLM use?
3 Answers
For newer 16GB options, the RTX 5060 and Radeon RX 9060 XT are the kinds of cards to compare. The best choice will depend heavily on local pricing, but 16GB gives more room for newer games and larger models than an 8GB card. For local LLMs, also check software compatibility and memory requirements rather than judging purely by gaming benchmarks.
A used RTX 3060 could be a practical choice if you can find one in good condition. It offers 12GB of VRAM, NVIDIA’s CUDA ecosystem, and generally broad support for games and local AI software. Just check the card’s condition, temperatures, and warranty situation before buying.
The RX 6600 XT is worth considering if you find it at a good price. It delivers performance around or above the RTX 3060 in many games and can be excellent value, especially for a mostly gaming-focused build. However, NVIDIA may still be the safer option for local LLM software because CUDA and application support are usually more straightforward.
The RX 6600-series pricing does look very competitive for the gaming performance. I’ll need to check whether the specific LLM tools I use support it well enough.