ailia LLM
LLM Prefill: ~3.9x Faster!*
Speeds up reading long text and shortens the time until the AI starts responding to image-based prompts.
*Gemma 4 E2B / prefill of 2,048 tokens / NPU vs. CPU
View Detailed ResultsQCS6490 × ailia SDK × Advantech
AI that runs inside the device, without relying on the cloud.
Easily add AI capabilities to industrial equipment.
* Compared with the CPU on actual QCS6490 hardware. LLM: prefill of 2,048 input tokens with Gemma 4 E2B. Speech recognition: Whisper Large-v3-Turbo. Overall processing is not uniformly accelerated by these factors.
* The on-screen UI is for illustration purposes only.
* QCS6490 refers to Qualcomm Technologies, Inc.'s IoT processor "Dragonwing™ QCS6490."
The NPU takes on heavy AI workloads, speeding up speech recognition and LLM input processing.
AI processing doesn't tie up the CPU, making it easy to combine with the device's core tasks.
AI inference runs on the device itself, so it also suits use cases without a network connection.
* Power consumption and continuous-operation conditions vary depending on the device configuration.
ailia LLM
Speeds up reading long text and shortens the time until the AI starts responding to image-based prompts.
*Gemma 4 E2B / prefill of 2,048 tokens / NPU vs. CPU
View Detailed Resultsailia AI Speech
Transcribes about 40 seconds of audio in about 7.7 seconds, for smoother voice control.
*Whisper Large-v3-Turbo / actual QCS6490 hardware / NPU vs. CPU
View Detailed Results* Both results are from published benchmarks under specific conditions. The speedup differs for other processes, such as LLM text generation.
Converts speech to text so the AI can understand instructions, helping with equipment status checks and data entry.
Ask questions about captured images or have the AI describe what is happening.
Build mechanisms that pull up equipment information and business data in response to questions to the AI.
Use AI models optimized for QCS6490 for text generation, image understanding, speech recognition, and more.
ailia supports implementation tailored to your hardware.
Supported technologies: LLM, VLM, speech recognition (Whisper), audio understanding, Tool Use, MTP
* Available features depend on the supported models and the execution environment. External operations via Tool Use are executed by the application.
Contact us about integrating AI into Advantech devices or using ailia SDK.
Qualcomm QCS6490 Development Kit
Learn More (opens in a new tab)About ailia SDK
Contact UsDeveloper guides and benchmarks