Skip to content

Kanana 2 3B Instruct VRAM requirements

How much VRAM Kanana 2 3B Instruct needs
4-bit weights: 2.0 GBWith 8K context: 3.8 GBWith 32K context: 6.8 GB8-bit, 8K context: 5.2 GB
Parameters
3.5B
Longest context
32K
Released
Jul 2026
License
Kanana Open License
Strongest at
Not rated

Its licence (Kanana Open License) is not a standard open-source one: check the terms before commercial use.

How much VRAM Kanana 2 3B Instruct needs

Memory Kanana 2 3B Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit6.5 GB8.3 GB11.3 GB
8-bit3.5 GB5.2 GB8.2 GB
6-bit2.7 GB4.5 GB7.5 GB
5-bit2.3 GB4.1 GB7.1 GB
4-bit2.0 GB3.8 GB6.8 GB
3-bit1.6 GB3.4 GB6.4 GB
2-bit1.2 GB3.0 GB6.0 GB
1.6-bit690 MB2.4 GB5.4 GB

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

GPUs and Macs that run Kanana 2 3B Instruct

The best way Kanana 2 3B Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit24K49 tok/s0.5 sPDF: Kanana 2 3B Instruct on RTX 4060 report
RTX 50608-bit24K78 tok/s0.4 sPDF: Kanana 2 3B Instruct on RTX 5060 report
RTX 30708-bit24K78 tok/s0.4 sPDF: Kanana 2 3B Instruct on RTX 3070 report
RTX 3080 10 GB8-bit32K126 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K63 tok/s0.6 sPDF: Kanana 2 3B Instruct on RTX 3060 12 GB report
Arc B5808-bit32K55 tok/s0.4 sPDF: Kanana 2 3B Instruct on Arc B580 report
RTX 40708-bit32K87 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX 4070 report
RTX 4070 Super8-bit32K87 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX 4070 Super report
RTX 50708-bit32K113 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K51 tok/s0.3 sPDF: Kanana 2 3B Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K78 tok/s0.3 sPDF: Kanana 2 3B Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K66 tok/s0.3 sPDF: Kanana 2 3B Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K88 tok/s0.2 sPDF: Kanana 2 3B Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K113 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K122 tok/s0.1 sPDF: Kanana 2 3B Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit32K145 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX 5070 Ti report
RTX 50808-bit32K154 tok/s0.1 sPDF: Kanana 2 3B Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit32K124 tok/s0.3 sPDF: Kanana 2 3B Instruct on Radeon RX 7900 XTX report
RTX 30908-bit32K150 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX 3090 report
RTX 40908-bit32K160 tok/s0.1 sPDF: Kanana 2 3B Instruct on RTX 4090 report
Radeon AI PRO R97008-bit32K88 tok/s0.2 sPDF: Kanana 2 3B Instruct on Radeon AI PRO R9700 report
RTX 50908-bit32K254 tok/s0.1 sPDF: Kanana 2 3B Instruct on RTX 5090 report
2× RTX 30908-bit32K150 tok/s0.1 sPDF: Kanana 2 3B Instruct on 2× RTX 3090 report
2× RTX 40908-bit32K160 tok/s0.1 sPDF: Kanana 2 3B Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K203 tok/s0.1 sPDF: Kanana 2 3B Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K154 tok/s0.1 sPDF: Kanana 2 3B Instruct on RTX 6000 Ada report
L40S8-bit32K140 tok/s0.1 sPDF: Kanana 2 3B Instruct on L40S report
2× RTX 50908-bit32K254 tok/s0.1 sPDF: Kanana 2 3B Instruct on 2× RTX 5090 report
4× RTX 30908-bit32K150 tok/s0.1 sPDF: Kanana 2 3B Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K254 tok/s0.1 sPDF: Kanana 2 3B Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K25 tok/s4.6 sPDF: Kanana 2 3B Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K25 tok/s4.6 sPDF: Kanana 2 3B Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K25 tok/s4.6 sPDF: Kanana 2 3B Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K52 tok/s2.3 sPDF: Kanana 2 3B Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K52 tok/s2.3 sPDF: Kanana 2 3B Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K52 tok/s2.3 sPDF: Kanana 2 3B Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K86 tok/s1.1 sPDF: Kanana 2 3B Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit32K94 tok/s0.3 sPDF: Kanana 2 3B Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K105 tok/s0.6 sPDF: Kanana 2 3B Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K105 tok/s0.6 sPDF: Kanana 2 3B Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K105 tok/s0.6 sPDF: Kanana 2 3B Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K136 tok/s0.2 sPDF: Kanana 2 3B Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K39 tok/s0.7 sPDF: Kanana 2 3B Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K39 tok/s0.7 sPDF: Kanana 2 3B Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K53 tok/s0.2 sPDF: Kanana 2 3B Instruct on RTX Spark report
DGX Spark8-bit32K49 tok/s0.2 sPDF: Kanana 2 3B Instruct on DGX Spark report
Dual DGX Spark8-bit32K49 tok/s0.1 sPDF: Kanana 2 3B Instruct on Dual DGX Spark report
DGX Station GB3008-bit32K585 tok/s0.1 sPDF: Kanana 2 3B Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K254 tok/s0.1 sPDF: Kanana 2 3B Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K254 tok/s0.1 sPDF: Kanana 2 3B Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K254 tok/s0.1 sPDF: Kanana 2 3B Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K280 tok/s0.1 sPDF: Kanana 2 3B Instruct on A100 80 GB SXM report
H100 SXM8-bit32K392 tok/s0.1 sPDF: Kanana 2 3B Instruct on H100 SXM report
4× H200 NVL8-bit32K483 tok/s0.1 sPDF: Kanana 2 3B Instruct on 4× H200 NVL report
8× H200 NVL8-bit32K483 tok/s0.1 sPDF: Kanana 2 3B 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 mradermacher/kanana-2-3b-instruct-GGUF
vLLM
vllm serve kakaocorp/kanana-2-3b-instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/kanana-2-3b-instruct-4bit

Questions

How much VRAM does Kanana 2 3B Instruct need?

Kanana 2 3B Instruct needs 2.0 GB for its weights at 4-bit, 3.8 GB with an 8K context and 6.8 GB with 32K. At 8-bit, 5.2 GB.

What is the minimum VRAM for Kanana 2 3B Instruct?

About 3.4 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 Kanana 2 3B Instruct?

3.5B parameters: 6.5 GB of weights at 16-bit, 2.0 GB at 4-bit.

Can Kanana 2 3B Instruct run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 160 tokens a second.

More models