What we tested
Salt Data tested the Dell Pro Max with GB10 for local inference and a document assistant. Its 128 GB of unified memory gives models and supporting services a shared capacity budget.
Dell Pro Max with GB10: What 128 GB Buys for Local AI
A summary of our August 4, 2026 review, updated August 10: measured results and deployment considerations for companies evaluating local AI.
The practical results
- The synthetic HR document assistant answered all 12 questions with the expected facts and citations, at 1.516 seconds median latency.
- A separate Qwen service completed 956 requests without an HTTP failure.
- Larger models left less memory available for the rest of the application.
Planning a business deployment
The verdict is a conditional buy for centrally managed local inference. Size the complete service, including context, concurrency, retrieval, and monitoring. Budget for integration, access controls, updates, and validation with your own documents. The full review covers setup, power, thermals, and SSD access.
SALT DATA BLOG ยท AUGUST 4, 2026
Read the full GB10 review
Test configurations, measurements, and the complete verdict.
Read on the blogConsidering local AI for your company?
We can help define the workload, select the hardware, and integrate an AI service into your IT environment.