Microsoft on NVIDIA Ampere

Can I run Phi-4 14B on an RTX 3080 10GB?

Not at Q4_K_M — it needs 10.8 GB against 8.6 GB available. Drop to IQ3_XXS and it fits, at about 74.1 tokens per second.

Won't fit

10.8 GB of 8.6 GB · 125%
011 GB
Weights 8.4 GB
KV cache 1.6 GB
Runtime overhead 0.8 GB
Over the limit 2.2 GB

Short by 2.2 GB. You can run it with 29 of 40 layers on the RTX 3080 10GB and the rest in system RAM, at roughly 12.9 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation12.9tok/s
Prompt processing857tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of Phi-4 14B on a RTX 3080 10GB

Highlighted row is the highest quality that still fits at 8K context.

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 27.4 GB 29.8 GB Won't fit 1.79
INT8 / W8A8 8.50 14.3 GB 16.7 GB Won't fit 4.33
Q8_0 (GGUF) 8.50 14.3 GB 16.7 GB Won't fit 4.33
FP8 (E4M3) 8.00 13.7 GB 16.1 GB Won't fit 4.69
Q6_K 6.56 11.2 GB 13.6 GB Won't fit 6.70
Q5_K_M 5.67 9.7 GB 12.1 GB Won't fit 8.90
Q5_K_S 5.52 9.4 GB 11.8 GB Won't fit 9.63
AWQ 4-bit 4.25 8.7 GB 11.1 GB Won't fit 11.7
GPTQ 4-bit 4.25 8.7 GB 11.1 GB Won't fit 11.7
MXFP4 4.25 8.7 GB 11.1 GB Won't fit 11.7
Q4_K_M 4.85 8.4 GB 10.8 GB Won't fit 12.9
Q4_K_S 4.58 7.9 GB 10.3 GB Won't fit 15.7
Q4_0 4.55 7.9 GB 10.3 GB Won't fit 15.8
IQ4_XS 4.25 7.4 GB 9.8 GB Won't fit 2K 19.8
Q3_K_M 3.91 6.9 GB 9.3 GB Won't fit 5K 29.5
IQ3_M 3.70 6.5 GB 8.9 GB Won't fit 6K 35.4
IQ3_XXS 3.06 5.5 GB 7.9 GB Fits, but tight 11K 74.1
Q2_K 2.63 4.8 GB 7.2 GB Runs comfortably 15K 82.9
IQ2_XXS 2.06 3.9 GB 6.3 GB Runs comfortably 16K 98.4
IQ1_M 1.75 3.4 GB 5.8 GB Runs comfortably 16K 110

Also worth checking