Skip to content

InternVL3 5 1B HF VRAM requirements

How much VRAM InternVL3 5 1B HF needs
4-bit weights: 613 MBWith 8K context: 2.2 GBWith 32K context: 4.9 GB8-bit, 8K context: 2.7 GB
Parameters
1.1B
Longest context
40K
Released
Aug 2025
License
Apache 2.0
Strongest at
Text and images

How much VRAM InternVL3 5 1B HF needs

Memory InternVL3 5 1B HF needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 40K
16-bit2.0 GB3.6 GB6.2 GB7.1 GB
8-bit1.0 GB2.7 GB5.3 GB6.2 GB
6-bit830 MB2.5 GB5.1 GB6.0 GB
5-bit720 MB2.3 GB5.0 GB5.8 GB
4-bit613 MB2.2 GB4.9 GB5.7 GB
3-bit494 MB2.1 GB4.8 GB5.6 GB
2-bit379 MB2.0 GB4.6 GB5.5 GB
1.6-bit209 MB1.8 GB4.5 GB5.3 GB

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

GPUs and Macs that run InternVL3 5 1B HF

The best way InternVL3 5 1B HF runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit40K170 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 4060 report
RTX 50608-bit40K256 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 5060 report
RTX 30708-bit40K256 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 3070 report
RTX 3080 10 GB8-bit40K378 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 3080 10 GB report
RTX 3060 12 GB8-bit40K215 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 3060 12 GB report
Arc B5808-bit40K170 tok/s0.1 sPDF: InternVL3 5 1B HF on Arc B580 report
RTX 40708-bit40K280 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 4070 report
RTX 4070 Super8-bit40K280 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 4070 Super report
RTX 50708-bit40K347 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 5070 report
RTX 4060 Ti 16 GB8-bit40K178 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit40K256 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit40K195 tok/s0.1 sPDF: InternVL3 5 1B HF on Arc A770 16 GB report
Radeon RX 9070 XT8-bit40K256 tok/s0.1 sPDF: InternVL3 5 1B HF on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit40K347 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 4070 Ti Super report
RTX 4080 Super8-bit40K369 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 4080 Super report
RTX 5070 Ti8-bit40K421 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 5070 Ti report
RTX 50808-bit40K440 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 5080 report
Radeon RX 7900 XTX8-bit40K326 tok/s0.1 sPDF: InternVL3 5 1B HF on Radeon RX 7900 XTX report
RTX 30908-bit40K433 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 3090 report
RTX 40908-bit40K454 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 4090 report
Radeon AI PRO R97008-bit40K256 tok/s0.1 sPDF: InternVL3 5 1B HF on Radeon AI PRO R9700 report
RTX 50908-bit40K622 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 5090 report
2× RTX 30908-bit40K433 tok/s0.1 sPDF: InternVL3 5 1B HF on 2× RTX 3090 report
2× RTX 40908-bit40K454 tok/s0.1 sPDF: InternVL3 5 1B HF on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit40K537 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit40K440 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX 6000 Ada report
L40S8-bit40K411 tok/s0.1 sPDF: InternVL3 5 1B HF on L40S report
2× RTX 50908-bit40K622 tok/s0.1 sPDF: InternVL3 5 1B HF on 2× RTX 5090 report
4× RTX 30908-bit40K433 tok/s0.1 sPDF: InternVL3 5 1B HF on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit40K622 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit40K83 tok/s1.1 sPDF: InternVL3 5 1B HF on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit40K83 tok/s1.1 sPDF: InternVL3 5 1B HF on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit40K83 tok/s1.1 sPDF: InternVL3 5 1B HF on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit40K153 tok/s0.6 sPDF: InternVL3 5 1B HF on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit40K153 tok/s0.6 sPDF: InternVL3 5 1B HF on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit40K153 tok/s0.6 sPDF: InternVL3 5 1B HF on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit40K218 tok/s0.3 sPDF: InternVL3 5 1B HF on M4 Max 128 GB report
M5 Max 128 GB8-bit40K231 tok/s0.1 sPDF: InternVL3 5 1B HF on M5 Max 128 GB report
M3 Ultra 96 GB8-bit40K247 tok/s0.2 sPDF: InternVL3 5 1B HF on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit40K247 tok/s0.2 sPDF: InternVL3 5 1B HF on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit40K247 tok/s0.2 sPDF: InternVL3 5 1B HF on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit40K288 tok/s0.1 sPDF: InternVL3 5 1B HF on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit40K131 tok/s0.2 sPDF: InternVL3 5 1B HF on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit40K131 tok/s0.2 sPDF: InternVL3 5 1B HF on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit40K184 tok/s0.1 sPDF: InternVL3 5 1B HF on RTX Spark report
DGX Spark8-bit40K170 tok/s0.1 sPDF: InternVL3 5 1B HF on DGX Spark report
Dual DGX Spark8-bit40K170 tok/s0.1 sPDF: InternVL3 5 1B HF on Dual DGX Spark report
DGX Station GB3008-bit40K968 tok/s0.1 sPDF: InternVL3 5 1B HF on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit40K622 tok/s0.1 sPDF: InternVL3 5 1B HF on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit40K622 tok/s0.1 sPDF: InternVL3 5 1B HF on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit40K622 tok/s0.1 sPDF: InternVL3 5 1B HF on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit40K661 tok/s0.1 sPDF: InternVL3 5 1B HF on A100 80 GB SXM report
H100 SXM8-bit40K800 tok/s0.1 sPDF: InternVL3 5 1B HF on H100 SXM report
4× H200 NVL8-bit40K888 tok/s0.1 sPDF: InternVL3 5 1B HF on 4× H200 NVL report
8× H200 NVL8-bit40K888 tok/s0.1 sPDF: InternVL3 5 1B HF 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 lewiswatson/Frame2KG-InternVL3.5-1b-JSON
vLLM
vllm serve OpenGVLab/InternVL3_5-1B-HF

Questions

How much VRAM does InternVL3 5 1B HF need?

InternVL3 5 1B HF needs 613 MB for its weights at 4-bit, 2.2 GB with an 8K context and 4.9 GB with 32K. At 8-bit, 2.7 GB.

What is the minimum VRAM for InternVL3 5 1B HF?

About 2.1 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 4060.

How big is InternVL3 5 1B HF?

1.1B parameters: 2.0 GB of weights at 16-bit, 613 MB at 4-bit.

Can InternVL3 5 1B HF run on an RTX 4090?

Yes. At 8-bit it holds 40K of context and writes about 454 tokens a second.

More models