Skip to content

A.X 4.0 Light VRAM requirements

How much VRAM A.X 4.0 Light needs
4-bit weights: 4.1 GBWith 8K context: 5.3 GBWith 16K context: 5.7 GB8-bit, 8K context: 8.4 GB
Parameters
7.3B
Longest context
16K
Released
Jul 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM A.X 4.0 Light needs

Memory A.X 4.0 Light needs at each precision and context length
PrecisionWeightsWith 8KWith 16K
16-bit13.5 GB14.7 GB15.2 GB
8-bit7.2 GB8.4 GB8.8 GB
6-bit5.5 GB6.7 GB7.2 GB
5-bit4.8 GB6.0 GB6.4 GB
4-bit4.1 GB5.3 GB5.7 GB
3-bit3.3 GB4.5 GB4.9 GB
2-bit2.5 GB3.7 GB4.2 GB
1.6-bit1.4 GB2.6 GB3.0 GB

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

GPUs and Macs that run A.X 4.0 Light

The best way A.X 4.0 Light runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40606-bit16K33 tok/s1.0 sPDF: A.X 4.0 Light on RTX 4060 report
RTX 50606-bit16K53 tok/s0.8 sPDF: A.X 4.0 Light on RTX 5060 report
RTX 30706-bit16K53 tok/s0.7 sPDF: A.X 4.0 Light on RTX 3070 report
RTX 3080 10 GB8-bit16K68 tok/s0.5 sPDF: A.X 4.0 Light on RTX 3080 10 GB report
RTX 3060 12 GB8-bit16K33 tok/s1.1 sPDF: A.X 4.0 Light on RTX 3060 12 GB report
Arc B5808-bit16K30 tok/s0.8 sPDF: A.X 4.0 Light on Arc B580 report
RTX 40708-bit16K46 tok/s0.5 sPDF: A.X 4.0 Light on RTX 4070 report
RTX 4070 Super8-bit16K46 tok/s0.4 sPDF: A.X 4.0 Light on RTX 4070 Super report
RTX 50708-bit16K61 tok/s0.5 sPDF: A.X 4.0 Light on RTX 5070 report
RTX 4060 Ti 16 GB8-bit16K27 tok/s0.7 sPDF: A.X 4.0 Light on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit16K41 tok/s0.6 sPDF: A.X 4.0 Light on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit16K36 tok/s0.7 sPDF: A.X 4.0 Light on Arc A770 16 GB report
Radeon RX 9070 XT8-bit16K49 tok/s0.4 sPDF: A.X 4.0 Light on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit16K61 tok/s0.3 sPDF: A.X 4.0 Light on RTX 4070 Ti Super report
RTX 4080 Super8-bit16K66 tok/s0.3 sPDF: A.X 4.0 Light on RTX 4080 Super report
RTX 5070 Ti8-bit16K80 tok/s0.3 sPDF: A.X 4.0 Light on RTX 5070 Ti report
RTX 50808-bit16K85 tok/s0.3 sPDF: A.X 4.0 Light on RTX 5080 report
Radeon RX 7900 XTX8-bit16K71 tok/s0.6 sPDF: A.X 4.0 Light on Radeon RX 7900 XTX report
RTX 30908-bit16K83 tok/s0.4 sPDF: A.X 4.0 Light on RTX 3090 report
RTX 40908-bit16K89 tok/s0.2 sPDF: A.X 4.0 Light on RTX 4090 report
Radeon AI PRO R97008-bit16K49 tok/s0.4 sPDF: A.X 4.0 Light on Radeon AI PRO R9700 report
RTX 50908-bit16K150 tok/s0.1 sPDF: A.X 4.0 Light on RTX 5090 report
2× RTX 30908-bit16K83 tok/s0.2 sPDF: A.X 4.0 Light on 2× RTX 3090 report
2× RTX 40908-bit16K89 tok/s0.1 sPDF: A.X 4.0 Light on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit16K116 tok/s0.2 sPDF: A.X 4.0 Light on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit16K85 tok/s0.2 sPDF: A.X 4.0 Light on RTX 6000 Ada report
L40S8-bit16K77 tok/s0.3 sPDF: A.X 4.0 Light on L40S report
2× RTX 50908-bit16K150 tok/s0.1 sPDF: A.X 4.0 Light on 2× RTX 5090 report
4× RTX 30908-bit16K83 tok/s0.1 sPDF: A.X 4.0 Light on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit16K150 tok/s0.1 sPDF: A.X 4.0 Light on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit16K13 tok/s9.2 sPDF: A.X 4.0 Light on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit16K13 tok/s9.2 sPDF: A.X 4.0 Light on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit16K13 tok/s9.2 sPDF: A.X 4.0 Light on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit16K29 tok/s4.6 sPDF: A.X 4.0 Light on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit16K29 tok/s4.6 sPDF: A.X 4.0 Light on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit16K29 tok/s4.6 sPDF: A.X 4.0 Light on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit16K50 tok/s2.3 sPDF: A.X 4.0 Light on M4 Max 128 GB report
M5 Max 128 GB8-bit16K55 tok/s0.7 sPDF: A.X 4.0 Light on M5 Max 128 GB report
M3 Ultra 96 GB8-bit16K63 tok/s1.3 sPDF: A.X 4.0 Light on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit16K63 tok/s1.3 sPDF: A.X 4.0 Light on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit16K63 tok/s1.3 sPDF: A.X 4.0 Light on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit16K86 tok/s0.4 sPDF: A.X 4.0 Light on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit16K20 tok/s1.3 sPDF: A.X 4.0 Light on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit16K20 tok/s1.3 sPDF: A.X 4.0 Light on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit16K28 tok/s0.5 sPDF: A.X 4.0 Light on RTX Spark report
DGX Spark8-bit16K26 tok/s0.5 sPDF: A.X 4.0 Light on DGX Spark report
Dual DGX Spark8-bit16K26 tok/s0.3 sPDF: A.X 4.0 Light on Dual DGX Spark report
DGX Station GB3008-bit16K432 tok/s0.1 sPDF: A.X 4.0 Light on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit16K150 tok/s0.1 sPDF: A.X 4.0 Light on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit16K150 tok/s0.1 sPDF: A.X 4.0 Light on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit16K150 tok/s0.1 sPDF: A.X 4.0 Light on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit16K167 tok/s0.3 sPDF: A.X 4.0 Light on A100 80 GB SXM report
H100 SXM8-bit16K252 tok/s0.1 sPDF: A.X 4.0 Light on H100 SXM report
4× H200 NVL8-bit16K331 tok/s0.1 sPDF: A.X 4.0 Light on 4× H200 NVL report
8× H200 NVL8-bit16K331 tok/s0.1 sPDF: A.X 4.0 Light 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 mykor/A.X-4.0-Light-gguf
vLLM
vllm serve skt/A.X-4.0-Light

Questions

How much VRAM does A.X 4.0 Light need?

A.X 4.0 Light needs 4.1 GB for its weights at 4-bit, 5.3 GB with an 8K context and 5.7 GB with 16K. At 8-bit, 8.4 GB.

What is the minimum VRAM for A.X 4.0 Light?

About 4.5 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 A.X 4.0 Light?

7.3B parameters: 13.5 GB of weights at 16-bit, 4.1 GB at 4-bit.

Can A.X 4.0 Light run on an RTX 4090?

Yes. At 8-bit it holds 16K of context and writes about 89 tokens a second.

More models