⌘K

ACEMAGIC F9A (Ryzen AI Max+ PRO 495)

ACEMAGIC · Mini PC · Released 2026Verified 2026-07-21
192 GB
LPDDR5x-8533
NPU TOPS
GB/s Bandwidth
— W
TDP
192 GBMemory (LPDDR5x-8533)
— GB/sMemory Bandwidth
NPU TOPS
GPU Compute Units
CPU Cores
— WTDP

MSRP: $5,999

Sourced MSRP · Verified 2026-07-21

No current affiliate offer available. Check retailer directly.

AI Workload Fit

Local LLM Inference
Good — 70B class models
✓ Capable
Image Generation
SDXL, Flux
✓ Capable
NPU/AI Acceleration
No NPU — 0 TOPS
✗ None
Power Efficiency
Good — ?W TDP
⚠ Moderate

Model Fit Analysis

Usable AI memory: 142.5 GB (OS reserve: 2GB, GPU allocation: 75%) Conservative

Maximum model size at Q4: 70B class models (278B params, ~556GB at FP16)

Method: amd_igpu_bios_limited. Estimates account for OS reserve, GPU memory allocation fraction, and memory architecture. Actual results vary by workload, BIOS settings, and model architecture.
Llama 3.1 8B ~5.52GB Q4
✓ Fits
Llama 3.1 70B ~37.14GB Q4
✓ Fits
Llama 3.2 1B ~2GB Q4
✓ Fits
Llama 3.2 3B ~3.01GB Q4
✓ Fits
Qwen 2.5 7B ~5.01GB Q4
✓ Fits
Qwen 2.5 14B ~8.53GB Q4
✓ Fits
Qwen 2.5 32B ~18.06GB Q4
✓ Fits
Qwen 2.5 72B ~38.14GB Q4
✓ Fits
Mistral 7B ~5.01GB Q4
✓ Fits
Mixtral 8x7B ~25.44GB Q4
✓ Fits
DeepSeek R1 7B ~5.01GB Q4
✓ Fits
DeepSeek R1 32B ~18.06GB Q4
✓ Fits
DeepSeek R1 70B ~37.14GB Q4
✓ Fits
Phi-3 Medium 14B ~8.53GB Q4
✓ Fits
Gemma 2 9B ~6.02GB Q4
✓ Fits
Gemma 2 27B ~15.05GB Q4
✓ Fits
Stable Diffusion XL ~4.81GB Q4
✓ Fits
Flux.1 Dev ~7.52GB Q4
✓ Fits
Flux.1 Schnell ~7.52GB Q4
✓ Fits

Full Specifications

SpecificationValue
Pcie Interface PCIe 4.0 x4 OCuLink https://acemagic.eu/blogs/nachrichten/new-acemagic-f9a-ryzen-ai-max-pro-495-mini-workstation
Vram Gb 192 https://acemagic.eu/blogs/nachrichten/new-acemagic-f9a-ryzen-ai-max-pro-495-mini-workstation
Memory Type LPDDR5x-8533 https://acemagic.com/products/f9a-ai-workstation
Memory Gb 192 https://acemagic.com/products/f9a-ai-workstation

Compare with Other Mini PCs

Popular Comparisons

Specifications sourced from manufacturer datasheets. MSRP verified 2026-07-21. Model-fit uses conservative estimates accounting for OS reserve and GPU allocation — actual performance varies. Methodology