Microsoft on NVIDIA Ada

Can I run Phi-4 14B on an RTX 4070 Super?

Not at Q4_K_M — it needs 10.8 GB against 10.5 GB available. Drop to Q4_K_S and it fits, at about 34.3 tokens per second.

Won't fit

10.8 GB of 10.5 GB · 103%
011 GB
Weights 8.4 GB
KV cache 1.6 GB
Runtime overhead 0.8 GB
Over the limit 0.3 GB

Short by 0.3 GB. You can run it with 38 of 40 layers on the RTX 4070 Super and the rest in system RAM, at roughly 24.4 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation24.4tok/s
Prompt processing1014tok/s
Max context6Ktokens
KV per 1K tokens0GB

Every quantisation of Phi-4 14B on a RTX 4070 Super

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 4.87
Q8_0 (GGUF) 8.50 14.3 GB 16.7 GB Won't fit 4.87
FP8 (E4M3) 8.00 13.7 GB 16.1 GB Won't fit 5.31
Q6_K 6.56 11.2 GB 13.6 GB Won't fit 8.19
Q5_K_M 5.67 9.7 GB 12.1 GB Won't fit 13.0
Q5_K_S 5.52 9.4 GB 11.8 GB Won't fit 1K 14.5
AWQ 4-bit 4.25 8.7 GB 11.1 GB Won't fit 5K 21.0
GPTQ 4-bit 4.25 8.7 GB 11.1 GB Won't fit 5K 21.0
MXFP4 4.25 8.7 GB 11.1 GB Won't fit 5K 21.0
Q4_K_M 4.85 8.4 GB 10.8 GB Won't fit 6K 24.4
Q4_K_S 4.58 7.9 GB 10.3 GB Fits, but tight 9K 34.3
Q4_0 4.55 7.9 GB 10.3 GB Fits, but tight 9K 34.5
IQ4_XS 4.25 7.4 GB 9.8 GB Fits, but tight 11K 36.5
Q3_K_M 3.91 6.9 GB 9.3 GB Runs comfortably 14K 39.1
IQ3_M 3.70 6.5 GB 8.9 GB Runs comfortably 16K 40.8
IQ3_XXS 3.06 5.5 GB 7.9 GB Runs comfortably 16K 47.4
Q2_K 2.63 4.8 GB 7.2 GB Runs comfortably 16K 53.0
IQ2_XXS 2.06 3.9 GB 6.3 GB Runs comfortably 16K 63.1
IQ1_M 1.75 3.4 GB 5.8 GB Runs comfortably 16K 70.3

Also worth checking