Your performance engineer wants to know if dropping Unit 2 to 430 MW for the next 6 hours saves more on fuel than it loses on grid penalties. Your dispatch desk wants to test switching to a 70/30 coal blend tomorrow. Your maintenance head wants to know if the boiler tube inspection can safely defer 14 days. None of these decisions can be tested on a live unit — the cost of being wrong is six figures. But on the What-If Scenario Engine, every one of those questions runs in about 5 seconds against a physics-plus-AI twin of your plant: boundary conditions in, physics simulation, AI surrogate model, ISO 20816-3 safety envelope check, impact projection out. Last month, one combined-cycle plant ran 1,248 scenarios — and saved $290,000. Register for the event to run your first live scenario at the booth.
The On-Prem Server Stack That Runs Your Scenario Engine
What-if simulation is GPU-heavy. Each scenario chains a CFD/Modelica physics solver to an AI surrogate model, runs Monte Carlo Tree Search across thousands of branch decisions, then evaluates the result against the ISO 20816-3 safety envelope — all in about 5 seconds. That is impossible on a cloud connection from your control room, and impossible on a generic IT server. The simulation runs on a DGX Station GB300 Ultra primary node, with an RTX PRO 6000 Blackwell secondary for live tag ingestion and a Jetson AGX Orin at the edge to keep the twin synced. Everything inside your perimeter, behind your firewall. Register for the event to inspect the full simulation stack.
EDGE · TWIN SYNC
NVIDIA Jetson AGX Orin Edge AI · DCS Bridge
GPU2048-core Ampere · 2× DLA accelerators
CPU12-core ARM Cortex-A78AE
RAM64GB unified LPDDR5
ProtocolOPC-UA · PI Web API · Modbus TCP
FormIndustrial enclosure · DIN-rail
Best for: Streams live boundary conditions to the twin every second so simulations always run on current plant state
CONTROL ROOM · INFERENCE
RTX PRO 6000 Blackwell Workstation Edition · Secondary Node
GPURTX PRO 6000 Blackwell · 96GB VRAM
CPUAMD Ryzen 7 9900X · 12-core
RAM128GB DDR5 6000MHz
Storage2TB NVMe M.2 SSD
PreloadedAI surrogate models · Omniverse runtime
Best for: Runs the AI surrogate model and drives the operator-facing scenario console in the control room
PRIMARY NODE · PHYSICS + MCTS
NVIDIA DGX Station GB300 Ultra · Simulation Engine
GPUGrace Blackwell GB300 Ultra superchip
RAM768GB unified memory
Network400GbE · multi-node simulation
Storage30TB NVMe + 100TB cold archive
PreloadedCFD-Modelica co-sim · MCTS engine
Best for: The brain of the scenario engine — runs CFD-Modelica physics co-simulation and Monte Carlo Tree Search across decision branches
100%
On-prem · behind your firewall · NERC CIP friendly
$0/mo
Perpetual license · no recurring fees ever
Air-Gap
Optional · zero internet egress if required
Source
Code & modification rights included
Inside the 5-Step Scenario Pipeline — How a Question Becomes an Answer in 5 Seconds
Every scenario — load dispatch, fuel mix, excess O2, maintenance deferral — flows through the same five-step pipeline. Plain-language version below. Register for the event to walk through this pipeline live with our engineers.
01
INPUT
Boundary Conditions
Live PI tags + the operator's question. Current load, fuel quality, ambient temperature, equipment status — captured as the starting state for the simulation.
Jetson AGX Orin · ~50ms
02
SIMULATE
Physics Co-Simulation
CFD models the combustion, heat transfer, and steam path. Modelica handles the thermodynamic cycle, controls, and mechanical response. The two solvers run in lock-step on the DGX.
DGX GB300 Ultra · ~2.5 sec
03
ACCELERATE
AI Surrogate Model
A neural network trained on millions of physics runs takes shortcuts the physics solver cannot — pruning the decision tree, suggesting candidate operating points, and running Monte Carlo Tree Search across branches.
RTX PRO 6000 Blackwell · ~1.5 sec
04
VALIDATE
Safety Envelope Check
Every candidate solution gets validated against ISO 20816-3 vibration limits, ASME pressure boundaries, NOx/SO₂ permit limits, and your unit's specific operating envelope. Unsafe scenarios are filtered out.
DGX GB300 Ultra · ~0.5 sec
05
OUTPUT
Impact Projection
A clean operator-facing answer: heat rate change, fuel cost delta, MWh impact, NOx delta, equipment stress score, confidence level. Plus a "do this" recommendation your operations head can sign off.
RTX PRO 6000 console · ~0.4 sec
~5 sec
From operator question to validated answer. 1,248 scenarios per month per plant at typical operator usage. The 5-second cycle is what makes scenario-driven dispatch a daily habit instead of a quarterly study.
Three Real Scenarios — Problem, Hardware Solution, Result
Three real scenarios pulled from a combined-cycle plant's last 30-day log: SC-1077, SC-1080, SC-1082. Each shows the operator question, how the hardware stack solves it, and the dollar impact. Register for the event to run your own scenario at the booth.
SC-1077
"Should we drop Unit 2 to 430 MW for the next 6 hours?"
THE PROBLEM
Day-ahead power prices crashed. Running Unit 2 at full 510 MW will lose money for 6 hours, but dropping too low will hit minimum stable load and risk a trip. Your dispatch desk needs to know the optimal load point — and they need to know it in the next 4 minutes before the market closes.
HOW THE HARDWARE SOLVES IT
Jetson AGX Orin
Captures live boundary conditions — current 510 MW load, ambient temp, fuel HHV, equipment health flags — and passes them to the engine.
DGX GB300 Engine
Runs the physics simulation across 400 candidate load points between 350 MW and 510 MW. Filters out anything that violates the safety envelope. Surfaces 430 MW as the optimal balance of fuel cost and grid penalty avoidance.
Operator Console
Dispatch desk sees: "Drop to 430 MW. Net savings: $34,000 over 6 hours. Heat rate penalty: 1.8%. Confidence: 92%." Decision made in 12 seconds.
THE RESULT
Unit dropped to 430 MW for 6 hours. Returned to full load when prices recovered. $34,000 saved on a single dispatch decision. Repeated 8 times that month.
SC-1080
"What happens if we run a 70/30 PRB/bituminous blend tomorrow?"
THE PROBLEM
Coal supplier offering a discount on PRB. Your performance engineer suspects a 70/30 PRB/bituminous blend will save fuel cost — but a higher-PRB blend changes combustion temperature, slagging risk, mill grindability, and NOx emissions. Test it on the live unit and you risk a furnace fouling event that takes 4 days to recover.
HOW THE HARDWARE SOLVES IT
Jetson AGX Orin
Snapshots current combustion conditions, mill loadings, and SCR performance to lock the simulation against today's plant state.
DGX GB300 Engine
CFD models the new blend's flame profile and slagging tendency at 24 hours. Modelica projects mill amps and SCR ammonia slip. Safety check confirms NOx stays under permit. Result: blend is safe but heat rate degrades 0.4%.
Operator Console
Performance engineer sees: "70/30 blend safe. Net savings: $18K/day after heat rate penalty. Slagging risk acceptable. Confidence: 89%." Procurement signs the contract.
THE RESULT
Blend run for 30 days. Actual savings landed at $17,400/day — within 4% of the simulation. $522,000 saved over the month. Zero slagging events.
SC-1082
"Can we safely defer the boiler tube inspection by 14 days?"
THE PROBLEM
A 48-hour boiler tube inspection is scheduled, but a heat wave is forecast and the grid is offering peak prices. Your maintenance head wants to defer the inspection 14 days. The question: is it safe? Defer too long and you risk a tube rupture mid-peak. Don't defer and you give up $400,000 in peak revenue.
HOW THE HARDWARE SOLVES IT
Jetson AGX Orin
Pulls the tube wall thickness history, current heat flux, and operating temperature trends — the exact state the engine needs to project the next 14 days.
DGX GB300 Engine
MCTS runs 1,000 scenarios across the 14-day defer period at peak load. Projects worst-case wall thinning. Result: 96% of scenarios end with adequate margin. Recommended action: defer 10 days, not 14, with daily UT spot checks.
Operator Console
Maintenance head sees: "Defer 10 days, capture $310K peak revenue, daily UT recommended. Confidence: 96%." Decision documented for audit.
THE RESULT
Inspection deferred 10 days. Peak revenue captured. UT confirmed tubes within spec. $310K captured. Zero safety risk taken. Maintenance still happened on time.
$290K/mo
Average monthly savings across all scenarios run on a typical combined-cycle plant — 1,248 scenarios per month, including dispatch, fuel mix, energy/APC, and maintenance deferral. The hardware pays for itself inside the first quarter.
Why This Matters — The Math on 1,248 Scenarios per Month
Without scenario simulation, every operational decision is either a gut-feel call or a 6-week engineering study. With it, decisions become a 5-second query — and operators ask 1,248 of them per month. The cumulative impact below is from real combined-cycle deployments.
1,248
Scenarios run per month at a typical combined-cycle plant — about 41 per day
$290K
Average monthly savings across dispatch, fuel mix, APC, and maintenance deferral scenarios
~5 sec
Per-scenario solve time. Fast enough that operators run scenarios as a daily habit, not a quarterly study
±4%
Typical accuracy of simulation predictions vs actual plant outcomes — operators trust the numbers
6-12wk
Pilot to fully running with your live PI tags wired to the twin. Server ships pre-installed
$0/mo
No subscription. Buy the stack once, own it forever. Source code included.
May 12 · 5:30 PM EST · Orlando · Hands-On
Bring Your Hardest Operator Question. Watch It Solve in 5 Seconds.
Walk into the booth, type your real plant question into the scenario console — load dispatch, fuel mix, excess O2, deferral — watch the engine run physics co-simulation, AI surrogate, safety check, and impact projection in front of you. Walk out with a quote and an order date. Pilot to fully running in 6 to 12 weeks. Buy it once, own it forever.
What You Get — Engine, Twin, Source Code, All In One Box
Pre-configured DGX Station GB300 Ultra arrives at your control room dock with the What-If Scenario Engine pre-installed, the CFD-Modelica co-simulation runtime, the AI surrogate models, and full source code. Plug it into your OT network, point it at your PI tags, run your first scenario.
Perpetual License
No monthly fees, no per-scenario charges, no per-user billing. Ever.
Data Sovereignty
PI tags, twin model, scenario logs — all behind your firewall. NERC CIP friendly.
Source Access
Source code and modification rights. Build unit-specific physics models in-house.
Physics + AI Native
CFD-Modelica co-sim, MCTS, AI surrogates — all bundled and pre-tuned.
Frequently Asked Questions
How fast can the Scenario Engine be running on our plant?
From signed order to live operator scenarios is typically 14 to 22 weeks. The DGX Station GB300 Ultra and RTX PRO 6000 Blackwell secondary node arrive in 4 to 6 weeks pre-configured. Once on-site, our team baselines the twin against your historical PI data, validates simulation accuracy by replaying past dispatch decisions, and goes live in 6 to 12 weeks. The engine ships pre-trained on combined-cycle and coal-fired physics — only unit-specific calibration is needed.
How accurate are the simulation predictions?
Typical accuracy is within ±4% of actual plant outcomes for dispatch and fuel mix scenarios, and within ±2% for short-horizon load scenarios. Accuracy is highest where the AI surrogate has been trained on your unit's own historical operating data — which happens automatically during the 6-12 week deployment. Every scenario carries a confidence score (typically 85-96%) so operators know how much weight to give the recommendation.
What scenarios can the engine handle out of the box?
Four scenario classes ship pre-built: load dispatch optimization, fuel mix and blending, energy/APC tuning (excess O2, attemperator flow, mill loading), and maintenance deferral. Each class has a built-in safety envelope check. Custom scenario types — startup sequence, runback recovery, load rejection — can be added during deployment by extending the source code, which is included with the perpetual license.
Will it work with our existing PI Historian and DCS?
Yes. The engine reads PI Historian directly via PI Web API and OPC-UA. We have integrated with PI servers paired with Emerson Ovation, GE Mark VI, ABB Symphony Plus, Siemens SPPA-T3000, and Yokogawa Centum DCS systems. The engine does not write back to your DCS — recommendations appear on the operator console, and your team retains full control of what gets executed.
What does "you own it" really mean?
You pay one price up front for the hardware, software, source code, and modification rights. There are no monthly fees, no per-scenario charges, no per-user billing, no annual escalator. Optional future costs are entirely your choice: support contracts for model retraining, custom scenario development, and hardware refresh whenever you decide. You can run the system forever without paying us anything more.
May 12 · 5:30 PM EST · Hands-On Hardware
Lock Your Spot. Run Your First Scenario at the Booth.
Walk in. Type a real operator question. Watch the DGX engine return a validated answer in 5 seconds. Ask any question to the engineers who built it. Leave with a quote and an order date. Pilot to fully running in 6 to 12 weeks. Buy it once, own it forever — no monthly fees, ever.
