1 Rookie
•
1 Message
3
261
May 15th, 2025 14:01
AirgapAI + Dell Pro AI Studio - 100% Local AI for 1/10th the Cost and 78X Accuracy - Fully Integrated!
We are very excited about Dell Pro AI Studio (DPAIS) and the value it brings to our company and our customers.
Our solution - AirgapAI - is a 100% Local platform of AI applications designed for the AI PC.
Because of our unique pricing model where AirgapAI is sold as a one-time perpetual license with an MSRP of $96 (Dell.com sells it for less!) - the need for a solution that can run fully isolated with no external dependencies is critical. We can keep the cost to the customer low because all of the AI happens locally on their own hardware.
Additionally the performance speed and efficiency from DPAIS makes it possible to bring our datacenter-based patented Blockify Data Ingestion technology that improves the accuracy of an LLM using RAG by 78X, to the local device!
DPAIS makes it easy for AirgapAI to leverage the Dell AI PC's NPU with the enterprise model management and deep hardware integrations to ensure easy compatibility on both Intel and Qualcomm chipsets.
Here's a link to a technical demo of the solution running on the NPU: http://video.iternal.ai/ffdb744ab103ed46f646757188fa2d42
Here's a link to a more detailed technical demo of the product: https://video.iternal.ai/b19be2e977cb2ca5f74a7806d7f08928
We are very excited about the performance and power that AirgapAI brings to our application!
Here are the results we are seeing on Qualcomm:
Hardware | Model | Test Type | Test Case | Status | Load Time (ms) | Time to Start Response (ms) | Prefill Characters/Second | Input Tokens | Input Tokens/Second | Generation Time (ms) | Decode Characters/Second | Output Tokens | Tokens/Second | |
os: Microsoft Windows 11 Pro | DPAIS - phi3.5 | contextSize | summarize - 2048 (25%) | Completed | N/A | 0.926 | 2121.179625 | 495 | 534.5572354 | 18.373 | 45.74061973 | 200 | 11.46328882 | |
cpu: Snapdragon® X Elite - X1E80100 - Qualcomm® Oryon™ X Elite - X1E80100 - Qualcomm® Oryon™ | DPAIS - phi3.5 | contextSize | summarize - 2048 (50%) | Completed | N/A | 1.47 | 2722.845935 | 1007 | 685.0340136 | 22.205 | 47.94219926 | 249 | 12.00868097 | |
npu: No | DPAIS - phi3.5 | contextSize | summarize - 2048 (75%) | Completed | N/A | 2.224 | 2723.034161 | 1519 | 683.0035971 | 30.104 | 46.05716253 | 321 | 11.51362984 | |
memory: 31.57GB RAM | DPAIS - phi3.5 | contextSize | summarize - 4096 (50%) | Completed | N/A | 2.926 | 2768.706324 | 2031 | 694.1216678 | 60.546 | 39.03395729 | 563 | 9.770912877 | |
GPU - 1: Qualcomm(R) Adreno(TM) X1-85 GPU | DPAIS - phi3.5 | contextSize | summarize - 4096 (75%) | Completed | N/A | 4.202 | 2902.689347 | 3055 | 727.0347454 | 43.077 | 45.17448516 | 439 | 11.2926045 | |
VRAM - 1: 0 MB | DPAIS - phi3.5 | contextSize | summarize - 8192 (50%) | Completed | N/A | 5.683 | 2866.530784 | 4079 | 717.754707 | 22.921 | 41.89174292 | 181 | 10.50005801 | |
DPAIS - phi3.5 | contextSize | createStory - 4096 (50%) | Completed | N/A | 0.194 | 565.7568238 | 29 | 149.4845361 | 101.362 | 36.45468321 | 922 | 9.113553693 |
Dell-StephenO
Moderator
•
153 Posts
1
May 20th, 2025 15:00
Thanks for posting your success John! For anyone visiting DTW today they can find John demoing this great partnership at the Intel booth today at 11:30AM Event time.
The other story I would like to share is just how quickly John was able to get this done. He emailed me around 4:50PM that he had started to deploy our private preview build. Around 4:55 he emailed back that DPAIS was up and running. Finally two or three minutes later he had already switched the configuration of his app to start using the local Phi model loaded by DPAIS.
That is the textbook turnkey experience we're looking to enable with DPAIS and were so excited that all our planning and development got us where we expected to be.
Thanks again John and enjoy your time at DTW 2025!
raed_hijer
1 Rookie
•
1 Message
0
May 21st, 2025 06:14
@Dell-StephenO / John, this is a great story to showcase the value of how quickly DPAIS can integrate into AI apps. Is it possible to show a video of the integration process? as you said, it's a mere 5 minutes!
Dell-StephenO
Moderator
•
153 Posts
2
May 30th, 2025 19:57
@raed_hijer Hey there! As we digest from DTW 2025 we will be getting some videos and demonstrations online for everyone to see.
For now, use cases will vary but if a customer is currently leveraging a cloud based AI resource and that resource is leveraging the OpenAI API stack it could be as simple as changing some code like below
From:
client = OpenAI(
base_url="https://your-custom-endpoint",
api_key="your-api-key",
)
To:
client = OpenAI(
base_url="http://localhost:8553/v1/openai",
api_key="dpais"
)
Some other examples of quick getting started examples are available at the following KB:
How to Get Started with Dell Pro AI Studio Using Code Samples
Finally some quick installation guides that will help you download and install DPAIS on supported Intel or Qualcomm devices.
Intel x64 Chipsets
(edited)