Microsoft on NVIDIA Ampere

Can I run Phi-4 14B on an RTX A6000?

Yes. Phi-4 14B at Q4_K_M uses 10.8 GB of the 44.3 GB available on a RTX A6000, and runs at about 51.8 tokens per second. You can push the context to 16K.

Runs comfortably

10.8 GB of 44.3 GB · 24%
044 GB
Weights 8.4 GB
KV cache 1.6 GB
Runtime overhead 0.8 GB

Phi-4 14B at Q4_K_M leaves 33.5 GB spare on a RTX A6000. There is room to raise the context length or move up a quantisation level.

Generation51.8tok/s
Prompt processing1071tok/s
Max context16Ktokens
KV per 1K tokens0GB

Every quantisation of Phi-4 14B on a RTX A6000

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 Runs comfortably 16K 17.0
INT8 / W8A8 8.50 14.3 GB 16.7 GB Runs comfortably 16K 31.6
Q8_0 (GGUF) 8.50 14.3 GB 16.7 GB Runs comfortably 16K 31.6
FP8 (E4M3) 8.00 13.7 GB 16.1 GB Runs comfortably 16K 33.0
Q6_K 6.56 11.2 GB 13.6 GB Runs comfortably 16K 39.7
Q5_K_M 5.67 9.7 GB 12.1 GB Runs comfortably 16K 45.4
Q5_K_S 5.52 9.4 GB 11.8 GB Runs comfortably 16K 46.5
AWQ 4-bit 4.25 8.7 GB 11.1 GB Runs comfortably 16K 50.2
GPTQ 4-bit 4.25 8.7 GB 11.1 GB Runs comfortably 16K 50.2
MXFP4 4.25 8.7 GB 11.1 GB Runs comfortably 16K 50.2
Q4_K_M 4.85 8.4 GB 10.8 GB Runs comfortably 16K 51.8
Q4_K_S 4.58 7.9 GB 10.3 GB Runs comfortably 16K 54.4
Q4_0 4.55 7.9 GB 10.3 GB Runs comfortably 16K 54.6
IQ4_XS 4.25 7.4 GB 9.8 GB Runs comfortably 16K 57.8
Q3_K_M 3.91 6.9 GB 9.3 GB Runs comfortably 16K 61.8
IQ3_M 3.70 6.5 GB 8.9 GB Runs comfortably 16K 64.6
IQ3_XXS 3.06 5.5 GB 7.9 GB Runs comfortably 16K 74.8
Q2_K 2.63 4.8 GB 7.2 GB Runs comfortably 16K 83.7
IQ2_XXS 2.06 3.9 GB 6.3 GB Runs comfortably 16K 99.5
IQ1_M 1.75 3.4 GB 5.8 GB Runs comfortably 16K 111

Also worth checking

Phi-4 14B on every GPU
Full hardware tableEverything that runs on a RTX A6000
Full model tableWhy the KV cache is the part that bites
Guide