US 
Industrial GPU Computers Command Edge AI Inference Systems for Automated Food Factory
A USA muffin manufacturer needed every product on its lines baked to the same standard, around the clock and across batches of different recipes. Its answer was an AI vision inspection system built in two tiers: an industrial GPU computer that trains the models, and compact, fanless machine vision PCs on the production line that run them.
The machine vision PCs check each muffin as it leaves the oven, classify it as good or not good, flag rejects for the inspectors and adjust the oven temperature when the bake drifts. The training server keeps improving the models as new images arrive from the line. It is the kind of system we build for food and beverage manufacturers.
Why manual inspection stopped scaling
Perfect muffins depend on a constant oven temperature, 24 hours a day and 7 days a week, which is hard to hold when the plant bakes different products in batches. The line relied on visual inspection: staff checked the bake and adjusted the ovens from their own skill and experience.
That judgment is good but not identical from person to person or shift to shift, and as production volumes grew the plant could no longer depend on it. The business also wanted to bring new ranges, such as healthier products, to market quickly, which means a quality check that can learn a new product rather than retrain a team.
What the line needed from the hardware
The inspection systems had to fit into a working bakery without slowing it down.
-
01
Fanless
Flour and dust are everywhere on a bakery line, so the computers on the line could have no fans or vents to clog.
-
02
Compact
A small footprint, so each system could be built into existing equipment rather than redesigning the line.
-
03
Camera inputs
Several camera interfaces to capture video and images from cameras installed along the line.
-
04
No downtime
Inspection had to run in real time with the line at full capacity; nothing could hold production up.
-
05
Central reporting
Results, reject images and log data sent to the plant's management system for review.
-
06
Retraining on site
Models improved on the real production line, with new data feeding back into training.
From camera to oven control
Training stays on the server and fast inference runs on the line, so each job gets the hardware that suits it.
-
Capture
Training serverThe training server collects video and image datasets from the production lines and pre-processes images before labeling.
-
Train
RTX GPUNeural network models are trained on the server’s RTX GPUs, with the CPU managing the wider system.
-
Deploy
Model managementTrained models are pushed to the machine vision PCs on the line through the model management software.
-
Inspect
Jetson Xavier NXEach machine vision PC inspects every muffin in real time and classifies it as OK or NG.
-
Act
Oven and inspectorsNG products are flagged on a display for the inspectors, oven temperatures are adjusted where needed, and reject images and logs go to the management system.
-
Retrain
ContinuousAs new data arrives, the server retrains the models to raise accuracy on the real line.
Training server and machine vision PCs
| Specification | Vecow RCX-2750R PEG (training) | Vecow EAC-2100 (inspection) |
|---|---|---|
| Role | Collects data, trains and retrains models | Runs the models on the line |
| Compute | Intel Core i7-11700 or Xeon W-1350 (11th Gen Rocket Lake) | NVIDIA Jetson Xavier NX 16 GB, 21 TOPS |
| GPU | NVIDIA RTX 30 series (RTX 3090 as deployed), dual GPU support | Integrated Jetson GPU |
| Memory | 4× DDR4-3200 SO-DIMM, up to 128 GB | 16 GB on module |
| Networking | 1× 2.5GbE, 1× GbE | 4× GbE (2× PoE+) |
| Cameras | Via network | 4× GMSL camera inputs |
| Storage | 4× 2.5″ bays + 2× M.2, RAID 0/1/5/10 | 16 GB eMMC, M.2 SSD, microSD |
| Power | 9–55 V DC | 9–50 V DC |
| Temperature | −25 to 60 °C | −20 to 70 °C, fanless |
Splitting the work this way keeps heavy training off the line and fast inference on it; our glossary of AI GPU computing terms explains how training, inference and sensor fusion differ.
Products in this build
The Vecow training server and machine vision PC the bakery deployed.
Intel Xeon W-1350 or Core i7-11700 Industrial GPU Computer with 4 Front-Access SSD Trays
- CPU
- Intel Core 11th Gen (Rocket Lake)
- GPU
- NVIDIA GeForce RTX 3080 Ti
- AI
- 273–320 TOPS
NVIDIA Jetson Xavier NX Edge AI Computer with GMSL Cameras and PoE+
- AI
- 21 TOPS
- SOM
- NVIDIA Jetson Xavier NX 16 GB
- Cameras
- 4× GMSL
What changed on the line
The plant now runs a consistent, objective quality check on every product.
-
Outcome 01
Every muffin inspected
Each product is checked objectively by the AI, removing the shift-to-shift variation of manual checks.
-
Outcome 02
Ovens adjusted automatically
When the bake drifts, oven temperatures are corrected without waiting for a person to spot it.
-
Outcome 03
Traceable rejects
NG images and log data are recorded and sent to the management system for evaluation.
-
Outcome 04
Ready for new ranges
New products, such as healthier ranges, can be added by training new models rather than retraining staff.
The same build today
Jetson Xavier NX has since been succeeded by Jetson Orin NX, with up to 100 TOPS. For a new line we would specify the Vecow EAC-6100 where the cameras are GMSL, or the EAC-6200 with four PoE ports for network cameras, both fanless and rated −20 to 70 °C. For training, the ECX-3400 PEG brings 12th or 13th Gen Intel Core, a PCIe Gen4 x16 slot for the GPU and RAID storage. Where a line is hosed down between batches, sealing and chassis material join the list, and our washdown design guide explains how IP69K sealing and stainless steel grades keep food-line computers hygienic.
Current equivalents
Today's Vecow systems for the same two jobs.
NVIDIA Jetson Orin NX 16GB Robotics Computer with 4x GMSL Cameras
- AI
- 100 TOPS
- SOM
- NVIDIA Jetson Orin NX 16 GB
- Cameras
- 4× GMSL
NVIDIA Jetson Orin NX 16GB Machine Vision PC with 4x PoE+ and CAN FD
- AI
- 100 TOPS
- SOM
- NVIDIA Jetson Orin NX 16 GB
- Ethernet
- 6× GbE (4× PoE)
Intel 12th/13th Gen Core Industrial PC with GPU Support, Dual PCIe and RAID 5
- CPU
- Intel Core 13th Gen (Raptor Lake)
- Ethernet
- 4× 2.5GbE, 2× GbE (4× PoE)
- Temp
- −20 to +45 °C
Machine vision FAQs
-
What Is Machine Vision?
Machine vision is the use of cameras, lighting and a computer to inspect, measure, identify or guide objects automatically. In industry it checks products for defects, reads codes, verifies assembly and guides robots, faster and more consistently than a person can, and today it increasingly runs AI models on edge computers beside the line.
-
AI Inference vs Training: What Is the Difference?
Training is how an AI model learns: it is fed large sets of labeled data and adjusted until it predicts accurately, which needs powerful GPUs. Inference is using the trained model on new data to make a decision, which is lighter per image and usually runs in real time on an edge computer beside the camera or machine.
-
What Is GMSL2?
GMSL2 (Gigabit Multimedia Serial Link 2) is a high-speed serial link that carries uncompressed camera video, power and control data over one coaxial cable of up to around 15 meters. Developed for automotive cameras, it is now widely used on robots, autonomous machines and drones with NVIDIA Jetson computers.
-
What Replaced the NVIDIA Jetson Xavier?
The NVIDIA Jetson Orin family replaced the Xavier generation. The Xavier NX maps to the Orin NX and the AGX Xavier to the AGX Orin, with a large jump in AI performance per watt and the same JetPack software stack. With Xavier now end-of-life and its last-time-buy closed, Orin is the migration path.
-
Jetson Orin Nano vs Orin NX: What Is the Difference?
The Jetson Orin Nano delivers up to 67 TOPS in 7 to 25 W with 4 to 8 GB of memory, suited to entry-level, low-power vision. The Orin NX steps up to 157 TOPS in 10 to 40 W with 8 to 16 GB, adds deep learning accelerators and a hardware video encoder, and handles multi-camera, heavier inference.
-
What Are the Advantages of Fanless Computers?
Fanless computers cool themselves passively, through a finned heatsink chassis and heat pipes, instead of a fan. With no moving parts and no airflow drawing in dirt, they are more reliable in dusty or vibrating places, silent, low maintenance and easy to seal, which is why they are the standard choice for industrial and embedded use.
-
What does IP69K mean?
IP69K means an enclosure is dust-tight and protected against high-pressure, high-temperature water jets per DIN 40050-9 — 80°C water at 8—10 MPa from close range. It is the food and beverage industry standard for sanitary wash-down on both sides of the Atlantic — and it does NOT imply submersion protection.
Planning AI vision inspection on a food line?
Arrange a call to discuss your vision automation requirements.