Skip to content

K2 Horizon 7B VRAM requirements

How much VRAM K2 Horizon 7B needs
4-bit weights: 5.1 GBWith 8K context: 7.3 GBWith 32K context: 10.6 GB8-bit, 8K context: 11.1 GB
Parameters
9B
Longest context
512K
Released
Sep 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM K2 Horizon 7B needs

Memory K2 Horizon 7B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit16.8 GB18.9 GB22.3 GB35.8 GB
8-bit8.9 GB11.1 GB14.5 GB28.0 GB
6-bit6.9 GB9.1 GB12.4 GB25.9 GB
5-bit6.0 GB8.1 GB11.5 GB25.0 GB
4-bit5.1 GB7.3 GB10.6 GB24.1 GB
3-bit4.1 GB6.3 GB9.6 GB23.1 GB
2-bit3.1 GB5.3 GB8.7 GB22.2 GB
1.6-bit1.7 GB3.9 GB7.3 GB20.8 GB

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

GPUs and Macs that run K2 Horizon 7B

The best way K2 Horizon 7B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40603-bit16K46 tok/s1.0 sPDF: K2 Horizon 7B on RTX 4060 report
RTX 50603-bit16K74 tok/s0.8 sPDF: K2 Horizon 7B on RTX 5060 report
RTX 30703-bit16K74 tok/s0.8 sPDF: K2 Horizon 7B on RTX 3070 report
RTX 3080 10 GB5-bit16K85 tok/s0.5 sPDF: K2 Horizon 7B on RTX 3080 10 GB report
RTX 3060 12 GB6-bit24K37 tok/s1.2 sPDF: K2 Horizon 7B on RTX 3060 12 GB report
Arc B5806-bit24K32 tok/s0.8 sPDF: K2 Horizon 7B on Arc B580 report
RTX 40706-bit24K51 tok/s0.5 sPDF: K2 Horizon 7B on RTX 4070 report
RTX 4070 Super6-bit24K51 tok/s0.4 sPDF: K2 Horizon 7B on RTX 4070 Super report
RTX 50706-bit24K67 tok/s0.5 sPDF: K2 Horizon 7B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit36K23 tok/s0.7 sPDF: K2 Horizon 7B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit36K36 tok/s0.7 sPDF: K2 Horizon 7B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit36K31 tok/s0.8 sPDF: K2 Horizon 7B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit36K42 tok/s0.4 sPDF: K2 Horizon 7B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit36K52 tok/s0.4 sPDF: K2 Horizon 7B on RTX 4070 Ti Super report
RTX 4080 Super8-bit36K57 tok/s0.3 sPDF: K2 Horizon 7B on RTX 4080 Super report
RTX 5070 Ti8-bit36K69 tok/s0.4 sPDF: K2 Horizon 7B on RTX 5070 Ti report
RTX 50808-bit36K73 tok/s0.3 sPDF: K2 Horizon 7B on RTX 5080 report
Radeon RX 7900 XTX8-bit88K61 tok/s0.7 sPDF: K2 Horizon 7B on Radeon RX 7900 XTX report
RTX 30908-bit88K71 tok/s0.4 sPDF: K2 Horizon 7B on RTX 3090 report
RTX 40908-bit88K76 tok/s0.2 sPDF: K2 Horizon 7B on RTX 4090 report
Radeon AI PRO R97008-bit144K42 tok/s0.4 sPDF: K2 Horizon 7B on Radeon AI PRO R9700 report
RTX 50908-bit144K128 tok/s0.1 sPDF: K2 Horizon 7B on RTX 5090 report
2× RTX 30908-bit248K71 tok/s0.3 sPDF: K2 Horizon 7B on 2× RTX 3090 report
2× RTX 40908-bit248K76 tok/s0.1 sPDF: K2 Horizon 7B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit248K99 tok/s0.2 sPDF: K2 Horizon 7B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit248K73 tok/s0.2 sPDF: K2 Horizon 7B on RTX 6000 Ada report
L40S8-bit248K66 tok/s0.3 sPDF: K2 Horizon 7B on L40S report
2× RTX 50908-bit352K128 tok/s0.1 sPDF: K2 Horizon 7B on 2× RTX 5090 report
4× RTX 30908-bit512K71 tok/s0.1 sPDF: K2 Horizon 7B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit512K128 tok/s0.1 sPDF: K2 Horizon 7B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit16K14 tok/s9.9 sPDF: K2 Horizon 7B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit36K11 tok/s9.9 sPDF: K2 Horizon 7B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit72K11 tok/s9.9 sPDF: K2 Horizon 7B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit36K25 tok/s5.0 sPDF: K2 Horizon 7B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit168K25 tok/s5.0 sPDF: K2 Horizon 7B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit248K25 tok/s5.0 sPDF: K2 Horizon 7B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit512K42 tok/s2.5 sPDF: K2 Horizon 7B on M4 Max 128 GB report
M5 Max 128 GB8-bit512K47 tok/s0.7 sPDF: K2 Horizon 7B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit416K53 tok/s1.4 sPDF: K2 Horizon 7B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit512K53 tok/s1.4 sPDF: K2 Horizon 7B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit512K53 tok/s1.4 sPDF: K2 Horizon 7B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit512K73 tok/s0.4 sPDF: K2 Horizon 7B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit248K18 tok/s1.4 sPDF: K2 Horizon 7B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit512K18 tok/s1.4 sPDF: K2 Horizon 7B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit512K24 tok/s0.5 sPDF: K2 Horizon 7B on RTX Spark report
DGX Spark8-bit512K22 tok/s0.5 sPDF: K2 Horizon 7B on DGX Spark report
Dual DGX Spark8-bit512K22 tok/s0.3 sPDF: K2 Horizon 7B on Dual DGX Spark report
DGX Station GB3008-bit512K359 tok/s0.1 sPDF: K2 Horizon 7B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit512K128 tok/s0.1 sPDF: K2 Horizon 7B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit512K128 tok/s0.1 sPDF: K2 Horizon 7B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit512K128 tok/s0.1 sPDF: K2 Horizon 7B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit448K143 tok/s0.3 sPDF: K2 Horizon 7B on A100 80 GB SXM report
H100 SXM8-bit448K213 tok/s0.1 sPDF: K2 Horizon 7B on H100 SXM report
4× H200 NVL8-bit512K278 tok/s0.1 sPDF: K2 Horizon 7B on 4× H200 NVL report
8× H200 NVL8-bit512K278 tok/s0.1 sPDF: K2 Horizon 7B 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 OsaurusAI/Diplodocus-preview-GGUF
vLLM
vllm serve IFM/K2-Horizon-7B
MLX (Mac)
mlx_lm.chat --model mlx-community/K2-Horizon-7B-oQ6e

Questions

How much VRAM does K2 Horizon 7B need?

K2 Horizon 7B needs 5.1 GB for its weights at 4-bit, 7.3 GB with an 8K context and 10.6 GB with 32K. At 8-bit, 11.1 GB.

What is the minimum VRAM for K2 Horizon 7B?

About 6.3 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 K2 Horizon 7B?

9B parameters: 16.8 GB of weights at 16-bit, 5.1 GB at 4-bit.

Can K2 Horizon 7B run on an RTX 4090?

Yes. At 8-bit it holds 88K of context and writes about 76 tokens a second.

More models