Can I run Phi-3 mini 3.8B on NVIDIA RTX 4090 24GB?
Yes: Phi-3 mini 3.8B runs on NVIDIA RTX 4090 24GB with F16 at roughly 60 tokens/s. Estimated memory 10.9 GiB at 8,192 tokens of context.
Memory by quantisation (8,192-token context)
| Quantisation | Memory | Verdict | Est. speed |
|---|---|---|---|
| F16Unquantized reference | 10.9 GiB | Fits | ~60 tok/s |
| Q8_0Near-lossless | 7.5 GiB | Fits | ~90 tok/s |
| Q6_KVery close to F16 | 6.7 GiB | Fits | ~103 tok/s |
| Q5_K_MGreat quality/size | 6.3 GiB | Fits | ~110 tok/s |
| Q4_K_MBest default for most | 5.9 GiB | Fits | ~118 tok/s |
| IQ4_XSSmaller 4-bit (imatrix) | 5.6 GiB | Fits | ~125 tok/s |
| Q3_K_MNoticeable loss | 5.5 GiB | Fits | ~129 tok/s |
| Q2_KLast resort; big loss | 5.2 GiB | Fits | ~136 tok/s |
NVIDIA RTX 4090 24GB offers about 24 GiB of memory (94% usable for a model) at 1008 GB/s. Speed is a bandwidth-bound estimate of single-stream generation at 65% efficiency; real results vary by runtime, drivers and settings, so measure on your machine.
How this is calculated
Weights ≈ parameters × bits per weight ÷ 8. KV cache ≈ 2 × 32 layers × 32 KV heads × 96 head dim × 2 bytes per token. Decode speed ≈ efficiency × memory bandwidth ÷ bytes read per token. Learn the maths in VRAM math and memory bandwidth, or try the VRAM calculator.
Phi-3 mini 3.8B on other hardware
Similar models on NVIDIA RTX 4090 24GB
Running slowly or out of memory? See CUDA out of memory and the quantisation lesson.