Skip to content

Phi 3.5 Mini Instruct VRAM requirements

How much VRAM Phi 3.5 Mini Instruct needs
4-bit weights: 2.2 GBWith 8K context: 5.8 GBWith 32K context: 14.8 GB8-bit, 8K context: 7.4 GB
Parameters
3.8B
Longest context
128K
Released
Aug 2024
License
MIT
Strongest at
Not rated

How much VRAM Phi 3.5 Mini Instruct needs

Memory Phi 3.5 Mini Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit7.1 GB10.7 GB19.7 GB55.7 GB
8-bit3.8 GB7.4 GB16.4 GB52.4 GB
6-bit2.9 GB6.5 GB15.5 GB51.5 GB
5-bit2.5 GB6.1 GB15.1 GB51.1 GB
4-bit2.2 GB5.8 GB14.8 GB50.8 GB
3-bit1.7 GB5.3 GB14.3 GB50.3 GB
2-bit1.3 GB4.9 GB13.9 GB49.9 GB
1.6-bit752 MB4.3 GB13.3 GB49.3 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Phi 3.5 Mini Instruct

The best way Phi 3.5 Mini Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Phi 3.5 Mini Instruct on RTX 4060 report
RTX 5060Too bigPDF: Phi 3.5 Mini Instruct on RTX 5060 report
RTX 3070Too bigPDF: Phi 3.5 Mini Instruct on RTX 3070 report
RTX 3080 10 GB5-bit16K168 tok/s0.3 sPDF: Phi 3.5 Mini Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit16K60 tok/s0.6 sPDF: Phi 3.5 Mini Instruct on RTX 3060 12 GB report
Arc B5808-bit16K52 tok/s0.4 sPDF: Phi 3.5 Mini Instruct on Arc B580 report
RTX 40708-bit16K82 tok/s0.3 sPDF: Phi 3.5 Mini Instruct on RTX 4070 report
RTX 4070 Super8-bit16K82 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on RTX 4070 Super report
RTX 50708-bit16K107 tok/s0.3 sPDF: Phi 3.5 Mini Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit28K49 tok/s0.4 sPDF: Phi 3.5 Mini Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit28K74 tok/s0.3 sPDF: Phi 3.5 Mini Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit28K62 tok/s0.4 sPDF: Phi 3.5 Mini Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit28K84 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit28K107 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit28K116 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit28K138 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on RTX 5070 Ti report
RTX 50808-bit28K146 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit48K118 tok/s0.4 sPDF: Phi 3.5 Mini Instruct on Radeon RX 7900 XTX report
RTX 30908-bit48K143 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on RTX 3090 report
RTX 40908-bit48K152 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on RTX 4090 report
Radeon AI PRO R97008-bit64K84 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on Radeon AI PRO R9700 report
RTX 50908-bit64K243 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on RTX 5090 report
2× RTX 30908-bit104K143 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 2× RTX 3090 report
2× RTX 40908-bit104K152 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit104K194 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit104K146 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on RTX 6000 Ada report
L40S8-bit104K133 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on L40S report
2× RTX 50908-bit128K243 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 2× RTX 5090 report
4× RTX 30908-bit128K143 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K243 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit16K30 tok/s5.2 sPDF: Phi 3.5 Mini Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit28K23 tok/s5.2 sPDF: Phi 3.5 Mini Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit40K23 tok/s5.2 sPDF: Phi 3.5 Mini Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit28K49 tok/s2.6 sPDF: Phi 3.5 Mini Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit80K49 tok/s2.6 sPDF: Phi 3.5 Mini Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit104K49 tok/s2.6 sPDF: Phi 3.5 Mini Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K82 tok/s1.3 sPDF: Phi 3.5 Mini Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit128K90 tok/s0.4 sPDF: Phi 3.5 Mini Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K100 tok/s0.7 sPDF: Phi 3.5 Mini Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K100 tok/s0.7 sPDF: Phi 3.5 Mini Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K100 tok/s0.7 sPDF: Phi 3.5 Mini Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K131 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit104K37 tok/s0.8 sPDF: Phi 3.5 Mini Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K37 tok/s0.8 sPDF: Phi 3.5 Mini Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K51 tok/s0.3 sPDF: Phi 3.5 Mini Instruct on RTX Spark report
DGX Spark8-bit128K46 tok/s0.3 sPDF: Phi 3.5 Mini Instruct on DGX Spark report
Dual DGX Spark8-bit128K46 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on Dual DGX Spark report
DGX Station GB3008-bit128K570 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K243 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K243 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K243 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K268 tok/s0.2 sPDF: Phi 3.5 Mini Instruct on A100 80 GB SXM report
H100 SXM8-bit128K378 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on H100 SXM report
4× H200 NVL8-bit128K468 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 4× H200 NVL report
8× H200 NVL8-bit128K468 tok/s0.1 sPDF: Phi 3.5 Mini Instruct on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf bartowski/Phi-3.5-mini-instruct-GGUF
vLLM
vllm serve microsoft/Phi-3.5-mini-instruct

Questions

How much VRAM does Phi 3.5 Mini Instruct need?

Phi 3.5 Mini Instruct needs 2.2 GB for its weights at 4-bit, 5.8 GB with an 8K context and 14.8 GB with 32K. At 8-bit, 7.4 GB.

What is the minimum VRAM for Phi 3.5 Mini Instruct?

About 5.3 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 3080 10 GB.

How big is Phi 3.5 Mini Instruct?

3.8B parameters: 7.1 GB of weights at 16-bit, 2.2 GB at 4-bit.

Can Phi 3.5 Mini Instruct run on an RTX 4090?

Yes. At 8-bit it holds 48K of context and writes about 152 tokens a second.

More models