RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

Can I run Phi 3.5 vision instruct locally?

Phi 3.5 vision instruct by microsoft has 4.1B parameters. At 4-bit with an 8K context it needs about 5.8 GB of GPU memory. It fits on 65 of 65 GPUs and Macs we track.

5.8 GBQ4_K_M · 8K context
RTX 4060smallest consumer GPU that fits
M1 16GBsmallest Mac that fits
128Kmax context
MHAattention type

Memory needed by quant and context length

GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use parameter count × bits per weight. How we calculate.

QuantQuality4K ctx8K ctx32K ctx128K ctx
2-bit (Q2_K)rough3.4 GB4.9 GB13.9 GB49.9 GB
3-bit (Q3_K_M)usable3.8 GB5.3 GB14.3 GB50.3 GB
4-bit (Q4_K_M)good, most popular4.3 GB5.8 GB14.8 GB50.8 GB
5-bit (Q5_K_M)very good4.7 GB6.2 GB15.2 GB51.2 GB
6-bit (Q6_K)near lossless5.1 GB6.6 GB15.6 GB51.6 GB
8-bit (Q8_0)effectively lossless6.1 GB7.6 GB16.6 GB52.6 GB
16-bit (F16/BF16)original weights9.7 GB11.2 GB20.2 GB56.2 GB

Which GPUs and Macs run Phi 3.5 vision instruct?

Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).

HardwareMemoryFits?Best quantNeedsMax tok/s (est.)
RTX 3060 12GB12 GBFitsQ8_07.6 GB82details
RTX 4060 Ti 16GB16 GBFitsF1611.2 GB35details
RTX 309024 GBFitsF1611.2 GB113details
RTX 409024 GBFitsF1611.2 GB122details
RTX 509032 GBFitsF1611.2 GB216details
RTX 5070 Ti16 GBFitsF1611.2 GB108details
RX 7900 XTX24 GBFitsF1611.2 GB116details
Mac M4 Pro 48GB48 GBFitsF1611.2 GB33details
Mac M4 Max 128GB128 GBFitsF1611.2 GB66details
Mac M3 Ultra 512GB512 GBFitsF1611.2 GB99details
DGX Spark128 GBFitsF1611.2 GB33details
RTX 407012 GBFitsQ8_07.6 GB114details
RTX 5060 Ti 16GB16 GBFitsF1611.2 GB54details
RTX 40608 GBFitsQ6_K6.6 GB80details
Arc B58012 GBFitsQ8_07.6 GB104details
RTX 3060 Ti8 GBFitsQ6_K6.6 GB132details
RTX 30708 GBFitsQ6_K6.6 GB132details
RTX 3080 10GB10 GBFitsQ8_07.6 GB173details
RTX 3080 12GB12 GBFitsQ8_07.6 GB207details
RTX 3090 Ti24 GBFitsF1611.2 GB122details
RTX 4060 Ti 8GB8 GBFitsQ6_K6.6 GB85details
RTX 4070 Super12 GBFitsQ8_07.6 GB114details
RTX 4070 Ti12 GBFitsQ8_07.6 GB114details
RTX 4070 Ti Super16 GBFitsF1611.2 GB81details
RTX 408016 GBFitsF1611.2 GB86details
RTX 4080 Super16 GBFitsF1611.2 GB89details
RTX 50608 GBFitsQ6_K6.6 GB132details
RTX 507012 GBFitsQ8_07.6 GB153details
RTX 508016 GBFitsF1611.2 GB116details
2x RTX 30902×24 GBFitsF1611.2 GB113details
2x RTX 40902×24 GBFitsF1611.2 GB122details
2x RTX 50902×32 GBFitsF1611.2 GB216details
RTX A600048 GBFitsF1611.2 GB93details
RTX 6000 Ada48 GBFitsF1611.2 GB116details
RTX PRO 6000 Blackwell96 GBFitsF1611.2 GB216details
L40S48 GBFitsF1611.2 GB104details
A100 40GB40 GBFitsF1611.2 GB188details
A100 80GB80 GBFitsF1611.2 GB246details
H100 PCIe80 GBFitsF1611.2 GB241details
H100 SXM80 GBFitsF1611.2 GB404details
H200141 GBFitsF1611.2 GB579details
RX 76008 GBFitsQ6_K6.6 GB85details
RX 7700 XT12 GBFitsQ8_07.6 GB98details
RX 7800 XT16 GBFitsF1611.2 GB75details
RX 7900 GRE16 GBFitsF1611.2 GB69details
RX 7900 XT20 GBFitsF1611.2 GB96details
RX 907016 GBFitsF1611.2 GB77details
RX 9070 XT16 GBFitsF1611.2 GB77details
Ryzen AI Max+ 395 128GB128 GBFitsF1611.2 GB31details
Arc A770 16GB16 GBFitsF1611.2 GB68details
Mac M1 16GB16 GBFitsQ8_07.6 GB15details
Mac M2 24GB24 GBFitsF1611.2 GB12details
Mac M3 24GB24 GBFitsF1611.2 GB12details
Mac M4 16GB16 GBFitsQ8_07.6 GB27details
Mac M4 32GB32 GBFitsF1611.2 GB14details
Mac M1 Pro 32GB32 GBFitsF1611.2 GB24details
Mac M3 Pro 36GB36 GBFitsF1611.2 GB18details
Mac M4 Pro 24GB24 GBFitsF1611.2 GB33details
Mac M4 Pro 64GB64 GBFitsF1611.2 GB33details
Mac M1 Max 64GB64 GBFitsF1611.2 GB48details
Mac M2 Max 96GB96 GBFitsF1611.2 GB48details
Mac M3 Max 128GB128 GBFitsF1611.2 GB48details
Mac M4 Max 64GB64 GBFitsF1611.2 GB66details
Mac M2 Ultra 192GB192 GBFitsF1611.2 GB96details
Mac M3 Ultra 256GB256 GBFitsF1611.2 GB99details

Questions

How much VRAM does Phi 3.5 vision instruct need?

About 5.8 GB at 4-bit (Q4_K_M) with an 8K context: 2.4 GB of weights (estimated), 3.0 GB of KV cache and 0.5 GB of runtime overhead. At 8-bit it needs about 7.6 GB.

What is the smallest GPU that runs Phi 3.5 vision instruct?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 4060 (8 GB).

Can I run Phi 3.5 vision instruct on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M1 16GB or larger, using the default macOS GPU memory limit.

Model facts

Hugging Face repomicrosoft/Phi-3.5-vision-instruct
Parameters4,146,621,440 (4.1B)
Layers / hidden size32 / 3072
AttentionMHA; 32 heads, 32 KV heads, head dim 96
KV cache per token (FP16)384.0 KiB
Max context131,072 tokens
Licensemit
Released on HFAug 16, 2024
Downloads (30d) / likes270,035 / 750

Sources: Hugging Face model API · config.json (microsoft/Phi-3.5-vision-instruct) · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.