What AI can the GeForce RTX 5090 run?

Qwen3 32B (Q4_K_M, a 20 GB download) is the largest of the 42 models we track that fits entirely in the 32 GB the GeForce RTX 5090 can give a model at a 4K context, with 12 GB to spare and room for up to 40,960 tokens of context.

Not entirely in its own memory. DeepSeek-R1 Distill 70B (Q4_K_M) needs 42 GB at a 4K context, and the GeForce RTX 5090 can give a model 32 GB. With enough system RAM the rest can spill over into it, which works but is much slower; pick your RAM above to check.

All 32 GB of its VRAM, per NVIDIA GeForce graphics card comparison (checked Oct 2, 2026). The model file, its context and the program running it all have to fit in that.

Hardware makers and model publishers don’t publish tokens-per-second figures, so there is no source we could cite. We only show whether a model fits and how much context it leaves room for.