Can I Run Phi 3.5 Family on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM)?

Superseded model. Phi 3.5 Family has been superseded by Phi-4 Family. This page is kept for reference; the newer family is a better starting point. View Phi-4 Family →

Written by Jakub Rusinowski · Last updated August 20, 2024

Yes — comfortably

Yes, comfortably — Phi 3.5 Mini at Q8_0 needs about 5.6 GB of the 24 GB usable on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~18.4 GB spare and running at ~116.7 tok/s (estimated), with room for about 131,072 tokens of context.

Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~116.7 tok/s

RTX 3090 Desktop (24 GB VRAM, 64 GB RAM) — what it gives a model

Usable memory for models24 GB
Memory bandwidth936 GB/s
Form factorDesktop
Operating systemWindows or Linux
Memory upgradeableYes

Phi 3.5 Family on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM): memory by quantization

QuantMemory neededFits 24 GB?Max contextEst. speedDownload
F169.2 GB✓ Yes128K~73.3 tok/s7.6 GB
Q8_05.6 GB✓ Yes128K~116.7 tok/s4 GB
Q6_K4.7 GB✓ Yes128K~137.8 tok/s3.1 GB
Q5_K_M4.3 GB✓ Yes128K~150.2 tok/s2.7 GB
Q4_K_M3.9 GB✓ Yes128K~164.3 tok/s2.3 GB
Q3_K_M3.2 GB✓ Yes128K~195 tok/s1.6 GB
Q2_K2.9 GB✓ Yes128K~217.4 tok/s1.2 GB

RTX 3090 desktop limitations

Recommended setup

Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving

How these numbers are calculated

FAQ

Can I run Phi 3.5 Family on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM)?

Yes, comfortably — Phi 3.5 Mini at Q8_0 needs about 5.6 GB of the 24 GB usable on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~18.4 GB spare and running at ~116.7 tok/s (estimated), with room for about 131,072 tokens of context.

Which quantization of Phi 3.5 Family should I use on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM)?

Q8_0 — it needs about 5.6 GB of the 24 GB available, downloads as roughly 4 GB, and runs at an estimated 116.7 tokens/sec with up to 128K of context.

What limits Phi 3.5 Family on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM)?

Nothing binding — the model fits with headroom and generates at a usable speed on this hardware.

Which runtime should I use?

Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving

Other Computers

Other Models on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM)

Phi 3.5 Family on GPUs

What This Model Is Good At

Model & Tools

← Can I Run It? | Phi 3.5 Family model page | Check your hardware