Skip to content

Nanbeige4.3 33B VRAM requirements

How much VRAM Nanbeige4.3 33B needs
4-bit weights: 18.7 GBWith 8K context: 21.3 GBWith 32K context: 26.2 GB8-bit, 8K context: 35.4 GB
Parameters
33.1B
Longest context
512K
Released
Oct 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM Nanbeige4.3 33B needs

Memory Nanbeige4.3 33B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit61.7 GB64.3 GB69.2 GB88.7 GB
8-bit32.8 GB35.4 GB40.3 GB59.8 GB
6-bit25.3 GB27.9 GB32.8 GB52.3 GB
5-bit21.9 GB24.6 GB29.4 GB48.9 GB
4-bit18.7 GB21.3 GB26.2 GB45.7 GB
3-bit15.1 GB17.7 GB22.6 GB42.1 GB
2-bit11.6 GB14.2 GB19.1 GB38.6 GB
1.6-bit6.4 GB9.0 GB13.8 GB33.3 GB

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

GPUs and Macs that run Nanbeige4.3 33B

The best way Nanbeige4.3 33B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Nanbeige4.3 33B on RTX 4060 report
RTX 5060Too bigPDF: Nanbeige4.3 33B on RTX 5060 report
RTX 3070Too bigPDF: Nanbeige4.3 33B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Nanbeige4.3 33B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Nanbeige4.3 33B on RTX 3060 12 GB report
Arc B580Too bigPDF: Nanbeige4.3 33B on Arc B580 report
RTX 4070Too bigPDF: Nanbeige4.3 33B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Nanbeige4.3 33B on RTX 4070 Super report
RTX 5070Too bigPDF: Nanbeige4.3 33B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Nanbeige4.3 33B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Nanbeige4.3 33B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Nanbeige4.3 33B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Nanbeige4.3 33B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Nanbeige4.3 33B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Nanbeige4.3 33B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Nanbeige4.3 33B on RTX 5070 Ti report
RTX 5080Too bigPDF: Nanbeige4.3 33B on RTX 5080 report
Radeon RX 7900 XTX3-bit32K34 tok/s3.0 sPDF: Nanbeige4.3 33B on Radeon RX 7900 XTX report
RTX 30903-bit32K39 tok/s1.9 sPDF: Nanbeige4.3 33B on RTX 3090 report
RTX 40903-bit32K42 tok/s0.8 sPDF: Nanbeige4.3 33B on RTX 4090 report
Radeon AI PRO R97006-bit20K14 tok/s1.9 sPDF: Nanbeige4.3 33B on Radeon AI PRO R9700 report
RTX 50906-bit20K45 tok/s0.6 sPDF: Nanbeige4.3 33B on RTX 5090 report
2× RTX 30908-bit52K19 tok/s1.1 sPDF: Nanbeige4.3 33B on 2× RTX 3090 report
2× RTX 40908-bit52K20 tok/s0.5 sPDF: Nanbeige4.3 33B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit56K27 tok/s1.0 sPDF: Nanbeige4.3 33B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit56K19 tok/s0.7 sPDF: Nanbeige4.3 33B on RTX 6000 Ada report
L40S8-bit56K17 tok/s1.2 sPDF: Nanbeige4.3 33B on L40S report
2× RTX 50908-bit128K35 tok/s0.4 sPDF: Nanbeige4.3 33B on 2× RTX 5090 report
4× RTX 30908-bit256K19 tok/s0.6 sPDF: Nanbeige4.3 33B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K35 tok/s0.5 sPDF: Nanbeige4.3 33B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Nanbeige4.3 33B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Nanbeige4.3 33B on Mac mini M4 24 GB report
Mac mini M4 32 GB3-bit20K6.1 tok/s43 sPDF: Nanbeige4.3 33B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Nanbeige4.3 33B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB6-bit36K8.3 tok/s22 sPDF: Nanbeige4.3 33B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit56K6.4 tok/s22 sPDF: Nanbeige4.3 33B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K11 tok/s11 sPDF: Nanbeige4.3 33B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K13 tok/s3.1 sPDF: Nanbeige4.3 33B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit168K15 tok/s6.0 sPDF: Nanbeige4.3 33B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit512K15 tok/s6.0 sPDF: Nanbeige4.3 33B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit512K15 tok/s6.0 sPDF: Nanbeige4.3 33B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit512K21 tok/s1.9 sPDF: Nanbeige4.3 33B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit56K4.5 tok/s6.2 sPDF: Nanbeige4.3 33B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K4.5 tok/s6.2 sPDF: Nanbeige4.3 33B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit384K6.1 tok/s2.2 sPDF: Nanbeige4.3 33B on RTX Spark report
DGX Spark8-bit384K5.6 tok/s2.2 sPDF: Nanbeige4.3 33B on DGX Spark report
Dual DGX Spark8-bit512K5.6 tok/s1.2 sPDF: Nanbeige4.3 33B on Dual DGX Spark report
DGX Station GB3008-bit512K119 tok/s0.2 sPDF: Nanbeige4.3 33B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit512K35 tok/s0.3 sPDF: Nanbeige4.3 33B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit512K35 tok/s0.2 sPDF: Nanbeige4.3 33B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit512K35 tok/s0.1 sPDF: Nanbeige4.3 33B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit200K39 tok/s1.4 sPDF: Nanbeige4.3 33B on A100 80 GB SXM report
H100 SXM8-bit200K62 tok/s0.4 sPDF: Nanbeige4.3 33B on H100 SXM report
4× H200 NVL8-bit512K86 tok/s0.2 sPDF: Nanbeige4.3 33B on 4× H200 NVL report
8× H200 NVL8-bit512K86 tok/s0.1 sPDF: Nanbeige4.3 33B 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 bartowski/Nanbeige_Nanbeige4.3-33B-GGUF
vLLM
vllm serve Nanbeige/Nanbeige4.3-33B

Questions

How much VRAM does Nanbeige4.3 33B need?

Nanbeige4.3 33B needs 18.7 GB for its weights at 4-bit, 21.3 GB with an 8K context and 26.2 GB with 32K. At 8-bit, 35.4 GB.

What is the minimum VRAM for Nanbeige4.3 33B?

About 17.7 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 Radeon RX 7900 XTX.

How big is Nanbeige4.3 33B?

33.1B parameters: 61.7 GB of weights at 16-bit, 18.7 GB at 4-bit.

Can Nanbeige4.3 33B run on an RTX 4090?

Yes. At 3-bit it holds 32K of context and writes about 42 tokens a second.

More models