GeForce RTX 4090 · Phi-4 · Q4_K_M
Yes. A GeForce RTX 4090 runs Phi-4 at Q4_K_M with room left at 8,192 tokens, roughly 52–89 tokens per second.
| Piece | Value | Note |
|---|---|---|
| Weights | 8.43 GiB | phi-4-Q4_K_M.gguf |
| KV cache | 1.56 GiB | 8,192 tokens |
| Engine overhead | 0.50 GiB | Flat 0.5 GiB allowance |
| 5% headroom | 0.52 GiB | On the sum above |
| Total | 11.0 GiB | Fits |
| Fast memory | 24.0 GiB | 24 GB VRAM |
| Speed | 52–89 tok/s | Dense bandwidth range |
ollama run phi4:14b
GeForce RTX 4090 · Phi-4 ·GGUF file · Methodology
FAQ
How was this fit calculated?
Weights 8.43 GiB, KV cache 1.56 GiB at 8,192 tokens, engine overhead 0.50 GiB, then 5% headroom, for 11.0 GiB. Fast memory available is 24.0 GiB.
What if I change the quant or the context?
This page is Q4_K_M at 8,192 tokens. A smaller quant or a shorter context lowers the total. The homepage calculator changes both without a new page.
Can I trust the tokens per second?
The range 52–89 tok/s is a bandwidth roofline, 55% to 80% efficient. This dense model reads the whole weight file each token. Measure your own box before you buy around it.
What do I type?
ollama run phi4:14b
Is GeForce RTX 4090 assumed to have extra RAM?
Yes. Discrete GPU pages assume 32 GB of system RAM. Offload uses whatever remains after 8 GB for Windows, and it is limited by the slower of DDR5-5600 and the PCIe link.