Skip to content

Llm Jp 4.1 8B Thinking VRAM requirements

How much VRAM Llm Jp 4.1 8B Thinking needs
4-bit weights: 4.9 GBWith 8K context: 6.8 GBWith 32K context: 9.8 GB8-bit, 8K context: 10.5 GB
Parameters
8.6B
Longest context
64K
Released
Sep 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM Llm Jp 4.1 8B Thinking needs

Memory Llm Jp 4.1 8B Thinking needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 64K
16-bit16.0 GB18.0 GB21.0 GB25.0 GB
8-bit8.5 GB10.5 GB13.5 GB17.5 GB
6-bit6.6 GB8.5 GB11.5 GB15.5 GB
5-bit5.7 GB7.6 GB10.6 GB14.6 GB
4-bit4.9 GB6.8 GB9.8 GB13.8 GB
3-bit3.9 GB5.9 GB8.9 GB12.9 GB
2-bit3.0 GB5.0 GB8.0 GB12.0 GB
1.6-bit1.7 GB3.6 GB6.6 GB10.6 GB

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

GPUs and Macs that run Llm Jp 4.1 8B Thinking

The best way Llm Jp 4.1 8B Thinking runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40603-bit20K48 tok/s1.0 sPDF: Llm Jp 4.1 8B Thinking on RTX 4060 report
RTX 50603-bit20K76 tok/s0.8 sPDF: Llm Jp 4.1 8B Thinking on RTX 5060 report
RTX 30703-bit20K76 tok/s0.8 sPDF: Llm Jp 4.1 8B Thinking on RTX 3070 report
RTX 3080 10 GB5-bit20K88 tok/s0.5 sPDF: Llm Jp 4.1 8B Thinking on RTX 3080 10 GB report
RTX 3060 12 GB6-bit28K38 tok/s1.2 sPDF: Llm Jp 4.1 8B Thinking on RTX 3060 12 GB report
Arc B5806-bit28K33 tok/s0.8 sPDF: Llm Jp 4.1 8B Thinking on Arc B580 report
RTX 40706-bit28K52 tok/s0.5 sPDF: Llm Jp 4.1 8B Thinking on RTX 4070 report
RTX 4070 Super6-bit28K52 tok/s0.4 sPDF: Llm Jp 4.1 8B Thinking on RTX 4070 Super report
RTX 50706-bit28K69 tok/s0.5 sPDF: Llm Jp 4.1 8B Thinking on RTX 5070 report
RTX 4060 Ti 16 GB8-bit44K24 tok/s0.7 sPDF: Llm Jp 4.1 8B Thinking on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit44K37 tok/s0.7 sPDF: Llm Jp 4.1 8B Thinking on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit44K32 tok/s0.8 sPDF: Llm Jp 4.1 8B Thinking on Arc A770 16 GB report
Radeon RX 9070 XT8-bit44K43 tok/s0.4 sPDF: Llm Jp 4.1 8B Thinking on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit44K54 tok/s0.4 sPDF: Llm Jp 4.1 8B Thinking on RTX 4070 Ti Super report
RTX 4080 Super8-bit44K59 tok/s0.3 sPDF: Llm Jp 4.1 8B Thinking on RTX 4080 Super report
RTX 5070 Ti8-bit44K71 tok/s0.4 sPDF: Llm Jp 4.1 8B Thinking on RTX 5070 Ti report
RTX 50808-bit44K75 tok/s0.3 sPDF: Llm Jp 4.1 8B Thinking on RTX 5080 report
Radeon RX 7900 XTX8-bit64K63 tok/s0.7 sPDF: Llm Jp 4.1 8B Thinking on Radeon RX 7900 XTX report
RTX 30908-bit64K74 tok/s0.4 sPDF: Llm Jp 4.1 8B Thinking on RTX 3090 report
RTX 40908-bit64K79 tok/s0.2 sPDF: Llm Jp 4.1 8B Thinking on RTX 4090 report
Radeon AI PRO R97008-bit64K43 tok/s0.4 sPDF: Llm Jp 4.1 8B Thinking on Radeon AI PRO R9700 report
RTX 50908-bit64K132 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on RTX 5090 report
2× RTX 30908-bit64K74 tok/s0.3 sPDF: Llm Jp 4.1 8B Thinking on 2× RTX 3090 report
2× RTX 40908-bit64K79 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit64K103 tok/s0.2 sPDF: Llm Jp 4.1 8B Thinking on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit64K75 tok/s0.2 sPDF: Llm Jp 4.1 8B Thinking on RTX 6000 Ada report
L40S8-bit64K68 tok/s0.3 sPDF: Llm Jp 4.1 8B Thinking on L40S report
2× RTX 50908-bit64K132 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on 2× RTX 5090 report
4× RTX 30908-bit64K74 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit64K132 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit20K15 tok/s9.9 sPDF: Llm Jp 4.1 8B Thinking on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit44K12 tok/s9.9 sPDF: Llm Jp 4.1 8B Thinking on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit64K12 tok/s9.9 sPDF: Llm Jp 4.1 8B Thinking on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit44K25 tok/s5.0 sPDF: Llm Jp 4.1 8B Thinking on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit64K25 tok/s5.0 sPDF: Llm Jp 4.1 8B Thinking on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit64K25 tok/s5.0 sPDF: Llm Jp 4.1 8B Thinking on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit64K44 tok/s2.5 sPDF: Llm Jp 4.1 8B Thinking on M4 Max 128 GB report
M5 Max 128 GB8-bit64K49 tok/s0.7 sPDF: Llm Jp 4.1 8B Thinking on M5 Max 128 GB report
M3 Ultra 96 GB8-bit64K55 tok/s1.4 sPDF: Llm Jp 4.1 8B Thinking on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit64K55 tok/s1.4 sPDF: Llm Jp 4.1 8B Thinking on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit64K55 tok/s1.4 sPDF: Llm Jp 4.1 8B Thinking on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit64K76 tok/s0.4 sPDF: Llm Jp 4.1 8B Thinking on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit64K18 tok/s1.4 sPDF: Llm Jp 4.1 8B Thinking on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit64K18 tok/s1.4 sPDF: Llm Jp 4.1 8B Thinking on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit64K25 tok/s0.5 sPDF: Llm Jp 4.1 8B Thinking on RTX Spark report
DGX Spark8-bit64K23 tok/s0.5 sPDF: Llm Jp 4.1 8B Thinking on DGX Spark report
Dual DGX Spark8-bit64K23 tok/s0.3 sPDF: Llm Jp 4.1 8B Thinking on Dual DGX Spark report
DGX Station GB3008-bit64K381 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit64K132 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit64K132 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit64K132 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit64K148 tok/s0.3 sPDF: Llm Jp 4.1 8B Thinking on A100 80 GB SXM report
H100 SXM8-bit64K223 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on H100 SXM report
4× H200 NVL8-bit64K292 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking on 4× H200 NVL report
8× H200 NVL8-bit64K292 tok/s0.1 sPDF: Llm Jp 4.1 8B Thinking 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 masahiroid/llm-jp-4.1-8b-thinking-gguf
vLLM
vllm serve llm-jp/llm-jp-4.1-8b-thinking

Questions

How much VRAM does Llm Jp 4.1 8B Thinking need?

Llm Jp 4.1 8B Thinking needs 4.9 GB for its weights at 4-bit, 6.8 GB with an 8K context and 9.8 GB with 32K. At 8-bit, 10.5 GB.

What is the minimum VRAM for Llm Jp 4.1 8B Thinking?

About 5.9 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 Llm Jp 4.1 8B Thinking?

8.6B parameters: 16.0 GB of weights at 16-bit, 4.9 GB at 4-bit.

Can Llm Jp 4.1 8B Thinking run on an RTX 4090?

Yes. At 8-bit it holds 64K of context and writes about 79 tokens a second.

More models