Skip to content

Phi 3 Mini 4K Instruct VRAM requirements

How much VRAM Phi 3 Mini 4K Instruct needs
4-bit weights: 2.2 GBWith 4K context: 4.3 GB8-bit, 4K context: 5.9 GB
Parameters
3.8B
Longest context
4K
Released
Apr 2024
License
MIT
Strongest at
Not rated

How much VRAM Phi 3 Mini 4K Instruct needs

Memory Phi 3 Mini 4K Instruct needs at each precision and context length
PrecisionWeightsWith 4K
16-bit7.1 GB9.2 GB
8-bit3.8 GB5.9 GB
6-bit2.9 GB5.0 GB
5-bit2.5 GB4.6 GB
4-bit2.2 GB4.3 GB
3-bit1.7 GB3.8 GB
2-bit1.3 GB3.4 GB
1.6-bit752 MB2.8 GB

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

GPUs and Macs that run Phi 3 Mini 4K Instruct

The best way Phi 3 Mini 4K Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit4K46 tok/s0.5 sPDF: Phi 3 Mini 4K Instruct on RTX 4060 report
RTX 50608-bit4K74 tok/s0.4 sPDF: Phi 3 Mini 4K Instruct on RTX 5060 report
RTX 30708-bit4K74 tok/s0.4 sPDF: Phi 3 Mini 4K Instruct on RTX 3070 report
RTX 3080 10 GB8-bit4K119 tok/s0.3 sPDF: Phi 3 Mini 4K Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit4K60 tok/s0.6 sPDF: Phi 3 Mini 4K Instruct on RTX 3060 12 GB report
Arc B5808-bit4K52 tok/s0.4 sPDF: Phi 3 Mini 4K Instruct on Arc B580 report
RTX 40708-bit4K82 tok/s0.3 sPDF: Phi 3 Mini 4K Instruct on RTX 4070 report
RTX 4070 Super8-bit4K82 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on RTX 4070 Super report
RTX 50708-bit4K107 tok/s0.3 sPDF: Phi 3 Mini 4K Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit4K49 tok/s0.4 sPDF: Phi 3 Mini 4K Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit4K74 tok/s0.3 sPDF: Phi 3 Mini 4K Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit4K62 tok/s0.4 sPDF: Phi 3 Mini 4K Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit4K84 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit4K107 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit4K116 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit4K138 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on RTX 5070 Ti report
RTX 50808-bit4K146 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit4K118 tok/s0.4 sPDF: Phi 3 Mini 4K Instruct on Radeon RX 7900 XTX report
RTX 30908-bit4K143 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on RTX 3090 report
RTX 40908-bit4K152 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on RTX 4090 report
Radeon AI PRO R97008-bit4K84 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on Radeon AI PRO R9700 report
RTX 50908-bit4K243 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on RTX 5090 report
2× RTX 30908-bit4K143 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 2× RTX 3090 report
2× RTX 40908-bit4K152 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit4K194 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit4K146 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on RTX 6000 Ada report
L40S8-bit4K133 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on L40S report
2× RTX 50908-bit4K243 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 2× RTX 5090 report
4× RTX 30908-bit4K143 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit4K243 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit4K23 tok/s5.2 sPDF: Phi 3 Mini 4K Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit4K23 tok/s5.2 sPDF: Phi 3 Mini 4K Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit4K23 tok/s5.2 sPDF: Phi 3 Mini 4K Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit4K49 tok/s2.6 sPDF: Phi 3 Mini 4K Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit4K49 tok/s2.6 sPDF: Phi 3 Mini 4K Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit4K49 tok/s2.6 sPDF: Phi 3 Mini 4K Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit4K82 tok/s1.3 sPDF: Phi 3 Mini 4K Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit4K90 tok/s0.4 sPDF: Phi 3 Mini 4K Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit4K100 tok/s0.7 sPDF: Phi 3 Mini 4K Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit4K100 tok/s0.7 sPDF: Phi 3 Mini 4K Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit4K100 tok/s0.7 sPDF: Phi 3 Mini 4K Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit4K131 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit4K37 tok/s0.8 sPDF: Phi 3 Mini 4K Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit4K37 tok/s0.8 sPDF: Phi 3 Mini 4K Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit4K51 tok/s0.3 sPDF: Phi 3 Mini 4K Instruct on RTX Spark report
DGX Spark8-bit4K46 tok/s0.3 sPDF: Phi 3 Mini 4K Instruct on DGX Spark report
Dual DGX Spark8-bit4K46 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on Dual DGX Spark report
DGX Station GB3008-bit4K570 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit4K243 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit4K243 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit4K243 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit4K268 tok/s0.2 sPDF: Phi 3 Mini 4K Instruct on A100 80 GB SXM report
H100 SXM8-bit4K378 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on H100 SXM report
4× H200 NVL8-bit4K468 tok/s0.1 sPDF: Phi 3 Mini 4K Instruct on 4× H200 NVL report
8× H200 NVL8-bit4K468 tok/s0.1 sPDF: Phi 3 Mini 4K 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 legraphista/Phi-3-mini-4k-instruct-update2024_07_03-IMat-GGUF
vLLM
vllm serve microsoft/Phi-3-mini-4k-instruct

Questions

How much VRAM does Phi 3 Mini 4K Instruct need?

Phi 3 Mini 4K Instruct needs 2.2 GB for its weights at 4-bit, 4.3 GB with an 4K context and 4.3 GB with 4K. At 8-bit, 5.9 GB.

What is the minimum VRAM for Phi 3 Mini 4K Instruct?

About 3.8 GB at 3-bit with an 4K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is Phi 3 Mini 4K Instruct?

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

Can Phi 3 Mini 4K Instruct run on an RTX 4090?

Yes. At 8-bit it holds 4K of context and writes about 152 tokens a second.

More models