Skip to content

K2 Horizon MoVA 36B A4B VRAM requirements

How much VRAM K2 Horizon MoVA 36B A4B needs
4-bit weights: 21.1 GBWith 8K context: 23.6 GBWith 32K context: 28.1 GB8-bit, 8K context: 39.6 GB
Parameters
37.4B
Active per token
4B
Longest context
512K
Released
Sep 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM K2 Horizon MoVA 36B A4B needs

Memory K2 Horizon MoVA 36B A4B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit69.7 GB72.3 GB76.8 GB94.8 GB
8-bit37.1 GB39.6 GB44.1 GB62.1 GB
6-bit28.6 GB31.1 GB35.6 GB53.6 GB
5-bit24.8 GB27.3 GB31.8 GB49.8 GB
4-bit21.1 GB23.6 GB28.1 GB46.1 GB
3-bit17.0 GB19.6 GB24.1 GB42.1 GB
2-bit13.1 GB15.6 GB20.1 GB38.1 GB
1.6-bit7.2 GB9.7 GB14.2 GB32.2 GB

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

GPUs and Macs that run K2 Horizon MoVA 36B A4B

The best way K2 Horizon MoVA 36B A4B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: K2 Horizon MoVA 36B A4B on RTX 4060 report
RTX 5060Too bigPDF: K2 Horizon MoVA 36B A4B on RTX 5060 report
RTX 3070Too bigPDF: K2 Horizon MoVA 36B A4B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 3060 12 GB report
Arc B580Too bigPDF: K2 Horizon MoVA 36B A4B on Arc B580 report
RTX 4070Too bigPDF: K2 Horizon MoVA 36B A4B on RTX 4070 report
RTX 4070 SuperToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 4070 Super report
RTX 5070Too bigPDF: K2 Horizon MoVA 36B A4B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: K2 Horizon MoVA 36B A4B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: K2 Horizon MoVA 36B A4B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: K2 Horizon MoVA 36B A4B on RTX 5070 Ti report
RTX 5080Too bigPDF: K2 Horizon MoVA 36B A4B on RTX 5080 report
Radeon RX 7900 XTX3-bit24K130 tok/s0.9 sPDF: K2 Horizon MoVA 36B A4B on Radeon RX 7900 XTX report
RTX 30903-bit24K171 tok/s0.6 sPDF: K2 Horizon MoVA 36B A4B on RTX 3090 report
RTX 40903-bit24K177 tok/s0.3 sPDF: K2 Horizon MoVA 36B A4B on RTX 4090 report
Radeon AI PRO R97005-bit24K85 tok/s0.6 sPDF: K2 Horizon MoVA 36B A4B on Radeon AI PRO R9700 report
RTX 50905-bit24K194 tok/s0.2 sPDF: K2 Horizon MoVA 36B A4B on RTX 5090 report
2× RTX 30908-bit36K107 tok/s0.3 sPDF: K2 Horizon MoVA 36B A4B on 2× RTX 3090 report
2× RTX 40908-bit36K112 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit40K135 tok/s0.3 sPDF: K2 Horizon MoVA 36B A4B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit40K109 tok/s0.2 sPDF: K2 Horizon MoVA 36B A4B on RTX 6000 Ada report
L40S8-bit40K101 tok/s0.4 sPDF: K2 Horizon MoVA 36B A4B on L40S report
2× RTX 50908-bit112K160 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on 2× RTX 5090 report
4× RTX 30908-bit256K107 tok/s0.2 sPDF: K2 Horizon MoVA 36B A4B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K160 tok/s0.2 sPDF: K2 Horizon MoVA 36B A4B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: K2 Horizon MoVA 36B A4B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: K2 Horizon MoVA 36B A4B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: K2 Horizon MoVA 36B A4B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: K2 Horizon MoVA 36B A4B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB6-bit24K47 tok/s6.8 sPDF: K2 Horizon MoVA 36B A4B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit40K39 tok/s6.8 sPDF: K2 Horizon MoVA 36B A4B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K58 tok/s3.4 sPDF: K2 Horizon MoVA 36B A4B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K63 tok/s1.0 sPDF: K2 Horizon MoVA 36B A4B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit160K68 tok/s1.9 sPDF: K2 Horizon MoVA 36B A4B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit512K68 tok/s1.9 sPDF: K2 Horizon MoVA 36B A4B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit512K68 tok/s1.9 sPDF: K2 Horizon MoVA 36B A4B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit512K82 tok/s0.6 sPDF: K2 Horizon MoVA 36B A4B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit40K31 tok/s1.9 sPDF: K2 Horizon MoVA 36B A4B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K31 tok/s1.9 sPDF: K2 Horizon MoVA 36B A4B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit384K43 tok/s0.8 sPDF: K2 Horizon MoVA 36B A4B on RTX Spark report
DGX Spark8-bit384K40 tok/s0.8 sPDF: K2 Horizon MoVA 36B A4B on DGX Spark report
Dual DGX Spark8-bit512K40 tok/s0.5 sPDF: K2 Horizon MoVA 36B A4B on Dual DGX Spark report
DGX Station GB3008-bit512K268 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit512K160 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit512K160 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit512K160 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit200K171 tok/s0.4 sPDF: K2 Horizon MoVA 36B A4B on A100 80 GB SXM report
H100 SXM8-bit200K213 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on H100 SXM report
4× H200 NVL8-bit512K241 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B on 4× H200 NVL report
8× H200 NVL8-bit512K241 tok/s0.1 sPDF: K2 Horizon MoVA 36B A4B 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 IFM/K2-Horizon-MoVA-36B-A4B-GGUF
vLLM
vllm serve IFM/K2-Horizon-MoVA-36B-A4B
MLX (Mac)
mlx_lm.chat --model mlx-community/K2-Horizon-MoVA-36B-A4B-4bit

Questions

How much VRAM does K2 Horizon MoVA 36B A4B need?

K2 Horizon MoVA 36B A4B needs 21.1 GB for its weights at 4-bit, 23.6 GB with an 8K context and 28.1 GB with 32K. At 8-bit, 39.6 GB.

What is the minimum VRAM for K2 Horizon MoVA 36B A4B?

About 19.6 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 K2 Horizon MoVA 36B A4B?

37.4B parameters (4B active per token): 69.7 GB of weights at 16-bit, 21.1 GB at 4-bit.

Can K2 Horizon MoVA 36B A4B run on an RTX 4090?

Yes. At 3-bit it holds 24K of context and writes about 177 tokens a second.

More models