Inside the lab

Make the experiment useful to the next person.

Fix the conditions, measure what matters, and publish the evidence. Explore UT’s reported migration story or prepare a reproducible test.

Founder-reported deployment

Understand Tech · Cloud to local

From a cloud bill
to an office rack.

UT’s founder reported moving its AI application workload from two AWS GPU instances to two DGX Spark systems.

$6,600Reported monthly cloud compute

2 SparksReported replacement setup

88%Claimed year-one cost reduction

Reference product image, not the deployed rack

Why this matters

A useful reason to investigate.

The founder’s supplied LinkedIn post describes local LLMs, retrieval, agents and coding tools. It claims roughly six-week payback and savings that could be reinvested in the company.

Treat the reported savings as a starting point for investigation. Publish workload volumes, application quality, operating costs and the assumptions behind the comparison.

Source: screenshots supplied by UT’s founder. Costs and performance have not been independently audited.

Make it your comparison

Bring the assumptions into view.

Equivalent quality, workload volume and latency
Hardware, energy, software and staff costs
Backup, downtime and recovery needs
Cloud instance and API alternatives

Build my cost scenario

What did deployment teach you?

Open the notebook for operating lessons, failures, fixes and evaluation checklists.

Open the field notebook

Manufacturer demonstration

See the platform in action

Loading experiments…

Inside the lab

Make the experiment useful to the next person.

Fix the conditions, measure what matters, and publish the evidence. Explore UT’s reported migration story or prepare a reproducible test.

Founder-reported deployment

Understand Tech · Cloud to local

From a cloud bill
to an office rack.

UT’s founder reported moving its AI application workload from two AWS GPU instances to two DGX Spark systems.

$6,600Reported monthly cloud compute

2 SparksReported replacement setup

88%Claimed year-one cost reduction

Reference product image, not the deployed rack

Why this matters

A useful reason to investigate.

The founder’s supplied LinkedIn post describes local LLMs, retrieval, agents and coding tools. It claims roughly six-week payback and savings that could be reinvested in the company.

Treat the reported savings as a starting point for investigation. Publish workload volumes, application quality, operating costs and the assumptions behind the comparison.

Source: screenshots supplied by UT’s founder. Costs and performance have not been independently audited.

Make it your comparison

Bring the assumptions into view.

Equivalent quality, workload volume and latency
Hardware, energy, software and staff costs
Backup, downtime and recovery needs
Cloud instance and API alternatives

Build my cost scenario

What did deployment teach you?

Open the notebook for operating lessons, failures, fixes and evaluation checklists.

Open the field notebook

Manufacturer demonstration

Make the experiment useful to the next person.

From a cloud bill
to an office rack.

A useful reason to investigate.

Bring the assumptions into view.

What did deployment teach you?

See the platform in action

A first look at DGX Spark

Make the experiment useful to the next person.

From a cloud bill
to an office rack.

A useful reason to investigate.

Bring the assumptions into view.

What did deployment teach you?

See the platform in action

A first look at DGX Spark

Make the experiment useful to the next person.

From a cloud billto an office rack.

A useful reason to investigate.

Bring the assumptions into view.

What did deployment teach you?

See the platform in action

A first look at DGX Spark

Make the experiment useful to the next person.

From a cloud billto an office rack.

A useful reason to investigate.

Bring the assumptions into view.

What did deployment teach you?

See the platform in action

A first look at DGX Spark

From a cloud bill
to an office rack.

From a cloud bill
to an office rack.