ai.canshu.com
中文
VRAM Calculator Advisor & TCO API Pricing Compare Models Leaderboard GPUs

Search models, GPUs, and tools

↑↓ Navigate · Enter Open · ★ Watchlist prioritized ai.canshu.com
Q4_K_M: Smooth Fit 71.92 GB / 99.8 GB Usable

Llama-4-Scout-109B-A17B (MoE) on AMD Ryzen AI Max+ 395 Mini PC (128GB Unified): VRAM Requirement & Speed Estimate

Yes. At Q4_K_M 4-bit quantization with a 16K context window, Llama-4-Scout-109B-A17B (MoE) requires 71.92 GB of VRAM (66.08 GB weights + 3.22 GB KV cache + 2.62 GB overhead), fitting within AI Max+ 395 128G's 99.8 GB usable memory at ~14.2 tokens/s.

Concurrency (Batch):
MoE 109B / A17B·48L (d=5120)·40Q/8KV·Ctx 1M·Text+Image
Runnable Models →
128 GB Unified·256 GB/s·Safe Usable 99.8 GB
Smooth Fit71.92 / 99.8 GB~14.2 tok/s
3. Weight Quantization
4. Per-Request Context16K (16,384 tok)
Verified: 2026-11-20·Roofline bandwidth/compute estimate (Batch=1)
Smooth Fit
71.92/ 99.8 GB usable

Physical 128 GB · Load 72%

Theoretical Decode Speed
~14.2tokens/s

MoE: reads only 10.31 GB active weights/step

Max Context at Batch=1
154K/ 1M

Enable Q8/Q4 KV quantization to scale batch concurrency

Weights: 66.08 GBKV Cache (1×16K): 3.22 GBOverhead: 2.62 GB
Reserves ~22% unified memory for OS
🚶 Readable Pace~35.2s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run llama4:scout

AI Max+ 395 128G Fits Llama-4-Scout-109B-A17B (MoE) · Cost & TCO

Approx. $1,999 (~$16/GB VRAM) · 120W TDP

📊 All Quantization Levels for Llama-4-Scout-109B-A17B (MoE) on AI Max+ 395 128G

Quantization Total VRAM (16K) Verdict Est. Speed Max Context
FP16 / BF16 226.88 GB Out of VRAM (OOM) — —
FP8 / Q8_0 122.65 GB Out of VRAM (OOM) — —
GGUF Q6_K 95.69 GB Tight Fit (Watch Context) ~10 tok/s 36K
GGUF Q5_K_M 83.6 GB Smooth Fit ~12.3 tok/s 96K
Q4_K_M / AWQ 71.92 GB Smooth Fit ~14.2 tok/s 154K
GGUF IQ4_XS 63.59 GB Smooth Fit ~15.9 tok/s 196K
GGUF Q3_K_M 58.86 GB Smooth Fit ~17 tok/s 219K
GGUF Q2_K 48.3 GB Smooth Fit ~20.3 tok/s 272K

💡 Recommended Alternative GPUs & Macs for Llama-4-Scout-109B-A17B (MoE)

Frequently Asked Questions

Can AMD Ryzen AI Max+ 395 Mini PC (128GB Unified) run Llama-4-Scout-109B-A17B (MoE) locally?

Yes. At Q4_K_M 4-bit quantization with a 16K context window, Llama-4-Scout-109B-A17B (MoE) requires 71.92 GB of VRAM (66.08 GB weights + 3.22 GB KV cache + 2.62 GB overhead), fitting within AI Max+ 395 128G's 99.8 GB usable memory at ~14.2 tokens/s.

What is the best quantization for Llama-4-Scout-109B-A17B (MoE) on AI Max+ 395 128G?

The highest-quality quantization that fits cleanly on AI Max+ 395 128G is GGUF Q6_K (Ultra High Quality), using 95.69 GB VRAM at ~10 tokens/s with up to 36K context.

How much VRAM does Llama-4-Scout-109B-A17B (MoE) need in full FP16 precision?

In unquantized FP16/BF16 precision (16K context), Llama-4-Scout-109B-A17B (MoE) requires 226.88 GB VRAM (218 GB for weights alone). Because it uses a Mixture-of-Experts (MoE) architecture, all 109B parameters must sit in memory, while only 17B active parameters are read per token.