Karnak 40B v1.0 — Hardware Requirements & GPU Compatibility
Chat
Karnak 40B v1.0 is a 40.7B-parameter open language model from Applied-Innovation-Center. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 24.84 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs on 24 GB (Q3_K_M · ~145 tok/s · 32K)Macs from 24 GB (Q2_K · ~46 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Karnak 40B v1.0
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
vllm serve Applied-Innovation-Center/Karnak-40B-v1.0 --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX PRO 4500 (32 GB) (6 GB headroom) · Q4_K_M
- Generation speed
- ~123 tok/s
- generation speed
- Cost per 1M output tokens
- $0.81
- per 1M output tokens
How Much VRAM Does Karnak 40B v1.0 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 17.7 GB | 19.7 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 64K context · ~155 tok/s | |
| 20.3 GB | 22.3 GB | 3-bit medium — 3-bit medium quantization | Fits: 32K context · ~145 tok/s | |
| 24.8 GB | 26.9 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Does not fit: needs 25.8 GB | |
| 29.4 GB | 31.4 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Does not fit: needs 30.4 GB | |
| 34.0 GB | 36 GB | 6-bit, very good — 6-bit quantization, very good quality | Does not fit: needs 34.9 GB | |
| 41.1 GB | 43.1 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 42 GB | |
| 81.8 GB | 83.8 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 82.7 GB |
Selected Q4_K_M
4-bit medium quantization25.8GBat 16K context
- Weights
- 24.4 GB
- KV cache
- 1.1 GB
- Overhead
- 0.3 GB
Doesn't fit RTX 4090 (reference) · needs 1.8 GB more, Q3_K_M fits
Or .
Which Hardware Can Run Karnak 40B v1.0?
Q4_K_M · 24.8 GBKarnak 40B v1.0 (Q4_K_M) requires 24.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 33+ GB is recommended. Using the full 262K context window can add up to 17.0 GB, bringing total usage to 41.9 GB. 1 GPU can run it, including NVIDIA GeForce RTX 5090. 98 devices with unified memory can run Karnak 40B v1.0, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M5 Max (36 GB).
20 GB
Q4_K_M · best fit Q2_K (18.7 GB)
1 GPU
RX 7900 XT~59Simulate output on RX 7900 XT32–40 GB
fits · 6.2 GB free · up to 64K ctx
1 GPU · 3 Mac classes · 2 device families
RTX 5090~171Simulate output on RTX 5090M5 Max36 GB~47Simulate output on Mac Studio M5 Max (36 GB)M532 GB~26Simulate output on MacBook Air 13" M5 (32 GB)Orin~42Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~27Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
2| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Phi 3.5 MoE InstructMicrosoft | — | 23.9 GB | ~141 | same size |
| Falcon 40B InstructTII UAE | — | 23.8 GB | ~33 | same size |
Specifications
- Publisher
- Applied-Innovation-Center
- Parameters
- 40.7B
- Active parameters
- 4.4B per token
- Architecture
- Qwen3MoeForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 192,728
- Release Date
- 2026-02-06
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Karnak 40B v1.0 need?
Karnak 40B v1.0 requires 24.8 GB of VRAM at Q4_K_M, or 81.8 GB at BF16. Full 262K context adds up to 17.0 GB (41.9 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 40.7B × 4.8 bits ÷ 8 = 24.4 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 17.5 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Karnak 40B v1.0?
Yes, at Q3_K_M (20.3 GB) or lower. Higher quantizations like Q4_K_M (24.8 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Karnak 40B v1.0?
For Karnak 40B v1.0, Q4_K_M (24.8 GB) offers the best balance of quality and VRAM usage. Q5_K_M (29.4 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 17.7 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Karnak 40B v1.0 on a Mac?
Yes — Mac Mini M5 Pro (24 GB) and 106 other Macs can run Karnak 40B v1.0, the smallest at Q2_K. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 24.8 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio M1 Max (32 GB).
Can I run Karnak 40B v1.0 locally?
Yes — Karnak 40B v1.0 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Karnak 40B v1.0?
At Q4_K_M, Karnak 40B v1.0 can reach ~75 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 2.7 GB active ÷ (8000 × 0.65) = 0.52 ms, plus 64 layers × 0.055 ms = 3.52 ms, so 1000 ÷ 4.04 ms = ~248 tok/s
Estimated speed at Q4_K_M (24.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Karnak 40B v1.0?
At Q4_K_M, the download is about 24.40 GB. The full-precision BF16 version is 81.34 GB. The smallest option (Q2_K) is 17.28 GB.
Which GPUs can run Karnak 40B v1.0?
1 consumer GPU can run Karnak 40B v1.0 at Q4_K_M. Top options include NVIDIA GeForce RTX 5090.
Which devices can run Karnak 40B v1.0?
98 devices with unified memory can run Karnak 40B v1.0 at Q4_K_M (24.8 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.