Skip to content

Kanana 1.5 8B Instruct 2505 VRAM requirements

How much VRAM Kanana 1.5 8B Instruct 2505 needs
4-bit weights: 4.5 GBWith 8K context: 6.4 GBWith 32K context: 9.4 GB8-bit, 8K context: 9.8 GB
Parameters
8B
Longest context
32K
Released
May 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM Kanana 1.5 8B Instruct 2505 needs

Memory Kanana 1.5 8B Instruct 2505 needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit15.0 GB16.8 GB19.8 GB
8-bit7.9 GB9.8 GB12.8 GB
6-bit6.1 GB8.0 GB11.0 GB
5-bit5.3 GB7.1 GB10.1 GB
4-bit4.5 GB6.4 GB9.4 GB
3-bit3.7 GB5.5 GB8.5 GB
2-bit2.8 GB4.6 GB7.6 GB
1.6-bit1.5 GB3.4 GB6.4 GB

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

GPUs and Macs that run Kanana 1.5 8B Instruct 2505

The best way Kanana 1.5 8B Instruct 2505 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40604-bit16K40 tok/s1.0 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 4060 report
RTX 50604-bit16K65 tok/s0.8 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 5060 report
RTX 30704-bit16K65 tok/s0.8 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 3070 report
RTX 3080 10 GB6-bit20K80 tok/s0.5 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit20K31 tok/s1.2 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 3060 12 GB report
Arc B5808-bit20K27 tok/s0.8 sPDF: Kanana 1.5 8B Instruct 2505 on Arc B580 report
RTX 40708-bit20K42 tok/s0.5 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 4070 report
RTX 4070 Super8-bit20K42 tok/s0.4 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 4070 Super report
RTX 50708-bit20K56 tok/s0.5 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K25 tok/s0.7 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K38 tok/s0.7 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K33 tok/s0.8 sPDF: Kanana 1.5 8B Instruct 2505 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K45 tok/s0.4 sPDF: Kanana 1.5 8B Instruct 2505 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K56 tok/s0.4 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K61 tok/s0.3 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 4080 Super report
RTX 5070 Ti8-bit32K73 tok/s0.4 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 5070 Ti report
RTX 50808-bit32K78 tok/s0.3 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 5080 report
Radeon RX 7900 XTX8-bit32K65 tok/s0.7 sPDF: Kanana 1.5 8B Instruct 2505 on Radeon RX 7900 XTX report
RTX 30908-bit32K76 tok/s0.4 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 3090 report
RTX 40908-bit32K82 tok/s0.2 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 4090 report
Radeon AI PRO R97008-bit32K45 tok/s0.4 sPDF: Kanana 1.5 8B Instruct 2505 on Radeon AI PRO R9700 report
RTX 50908-bit32K137 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 5090 report
2× RTX 30908-bit32K76 tok/s0.3 sPDF: Kanana 1.5 8B Instruct 2505 on 2× RTX 3090 report
2× RTX 40908-bit32K82 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K106 tok/s0.2 sPDF: Kanana 1.5 8B Instruct 2505 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K78 tok/s0.2 sPDF: Kanana 1.5 8B Instruct 2505 on RTX 6000 Ada report
L40S8-bit32K71 tok/s0.3 sPDF: Kanana 1.5 8B Instruct 2505 on L40S report
2× RTX 50908-bit32K137 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on 2× RTX 5090 report
4× RTX 30908-bit32K76 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K137 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit24K15 tok/s9.9 sPDF: Kanana 1.5 8B Instruct 2505 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K12 tok/s9.9 sPDF: Kanana 1.5 8B Instruct 2505 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K12 tok/s9.9 sPDF: Kanana 1.5 8B Instruct 2505 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K26 tok/s5.0 sPDF: Kanana 1.5 8B Instruct 2505 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K26 tok/s5.0 sPDF: Kanana 1.5 8B Instruct 2505 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K26 tok/s5.0 sPDF: Kanana 1.5 8B Instruct 2505 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K45 tok/s2.5 sPDF: Kanana 1.5 8B Instruct 2505 on M4 Max 128 GB report
M5 Max 128 GB8-bit32K50 tok/s0.7 sPDF: Kanana 1.5 8B Instruct 2505 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K57 tok/s1.4 sPDF: Kanana 1.5 8B Instruct 2505 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K57 tok/s1.4 sPDF: Kanana 1.5 8B Instruct 2505 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K57 tok/s1.4 sPDF: Kanana 1.5 8B Instruct 2505 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K78 tok/s0.4 sPDF: Kanana 1.5 8B Instruct 2505 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K19 tok/s1.4 sPDF: Kanana 1.5 8B Instruct 2505 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K19 tok/s1.4 sPDF: Kanana 1.5 8B Instruct 2505 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K26 tok/s0.5 sPDF: Kanana 1.5 8B Instruct 2505 on RTX Spark report
DGX Spark8-bit32K23 tok/s0.5 sPDF: Kanana 1.5 8B Instruct 2505 on DGX Spark report
Dual DGX Spark8-bit32K23 tok/s0.3 sPDF: Kanana 1.5 8B Instruct 2505 on Dual DGX Spark report
DGX Station GB3008-bit32K390 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K137 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K137 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K137 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K153 tok/s0.3 sPDF: Kanana 1.5 8B Instruct 2505 on A100 80 GB SXM report
H100 SXM8-bit32K229 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on H100 SXM report
4× H200 NVL8-bit32K300 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 on 4× H200 NVL report
8× H200 NVL8-bit32K300 tok/s0.1 sPDF: Kanana 1.5 8B Instruct 2505 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 DevQuasar/kakaocorp.kanana-1.5-8b-instruct-2505-GGUF
vLLM
vllm serve kakaocorp/kanana-1.5-8b-instruct-2505

Questions

How much VRAM does Kanana 1.5 8B Instruct 2505 need?

Kanana 1.5 8B Instruct 2505 needs 4.5 GB for its weights at 4-bit, 6.4 GB with an 8K context and 9.4 GB with 32K. At 8-bit, 9.8 GB.

What is the minimum VRAM for Kanana 1.5 8B Instruct 2505?

About 5.5 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 1.5 8B Instruct 2505?

8B parameters: 15.0 GB of weights at 16-bit, 4.5 GB at 4-bit.

Can Kanana 1.5 8B Instruct 2505 run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 82 tokens a second.

More models