Innodisk · NVIDIA · 2U short-depth server
APEX-S100
Fine-tuning, training and RAG on your own data in a 2U short-depth rack server with up to two H200 NVL; it also serves LLMs to a whole team on premise.
Answers Data stays in-house
Product Details
Edge AI Applications
From Development to the Field: APEX for AI Applications
Machine VisionVideo AnalyticsLLM and RAG ApplicationsGenerative AIModel Training and Fine-TuningRobotics and AMR
Your Model, Your Hardware
Companies that build AI software, from visual inspection and video analytics to LLM assistants and robotics, need hardware at two stages: during development, to train, fine-tune and test models on their own data; and in the field, to run the finished application at the customer's machine, vehicle or server room. The two stages call for different systems, and choosing them together avoids porting later.
Innodisk's APEX series covers both stages with eight systems. Björn Tech helps you choose the right APEX for each stage, complete it with cameras and remote management, and integrate it into enclosures and cabinets for the field.
APEX at a Glance
| Who it is for | Software companies and integrators building AI applications: vision, video analytics, LLM assistants |
|---|---|
| Stages | Development, training and fine-tuning; on-premise LLM and RAG; inference at customer sites |
| Range | Eight APEX systems, from a 36 TOPS box PC to a 2U server with two NVIDIA H200 NVL |
| Accelerators | NVIDIA RTX and H200 NVL, Qualcomm Cloud AI 100 Ultra and Dragonwing, Intel Core Ultra with NPU |
| Accelerator memory | Up to 128 GB on one card (Cloud AI 100 Ultra) and 141 GB HBM3e per GPU (H200 NVL) |
| Operating temperature | From 0 ~ 30 °C in the server room to -40 ~ 70 °C in vehicles and outdoor cabinets |
| Software | Innodisk AI SDK, iVIT vision AI toolkit and iCAP remote management |
| Scope of supply | APEX system, cameras, remote management and integration into enclosures and cabinets from one supplier |
What AI Software Teams Need
-
Data stays in-house
Training data and customer documents often cannot leave the company. Models have to be trained, fine-tuned and served on premise, without a cloud account.
-
Large models on one system
LLMs and VLMs need large accelerator memory; the model has to fit on the system that runs it.
Our answer APEX-X100APEX-X100-Q
-
Predictable cost
Cloud GPU hours grow with every experiment. Own hardware turns development and inference into a fixed, plannable investment, in the lab and at every customer site.
-
The same platform from lab to field
Choosing the field platform during development avoids porting later: NVIDIA systems from the desk to the machine, Qualcomm and Intel options where power and temperature demand it.
-
Harsh field conditions
Outside the office, systems face heat, cold, vibration and vehicle power; a desktop PC does not last there.
-
Reliable camera input
A vision application is only as good as its images; cameras should be validated on the platform the model runs on.
Our answer MIPI CSI-2 Camera ModulesMIPI over Type-C Cameras
-
Managing installed devices
Once the application runs at dozens of customer sites, models have to be updated and frozen devices recovered without a site visit.
Our answer InnoAgent EZ2N-0XL1
One Family from Desk to Field
In development, the model and the data decide: fine-tuning and RAG on large data sets need a server such as APEX-S100, large language models need the accelerator memory of APEX-X100 or APEX-X100-Q, and prototypes run cost-efficiently on APEX-X200.
In the field, the site decides: temperature, power supply, size and the number of cameras. APEX-P100 and APEX-P200 bring NVIDIA GPUs to the machine, APEX-A100 runs from vehicle power at -40 ~ 70 °C, and APEX-E100 handles low-power inference with up to 36 TOPS.
Every APEX is built with Innodisk's own SSDs and DRAM and supported by the Innodisk AI SDK; iVIT trains and runs vision models without code, and iCAP deploys models and updates to installed devices remotely.
- 01 Development, Training and On-Premise AI Office, lab or server room: train and fine-tune models on your own data, build and test LLM applications and run them on premise, without sending data to the cloud. Innodisk
- 02 Deployment in the Field At the machine, in the vehicle or in an outdoor cabinet: compact systems run the trained model where the data is created. Innodisk
- 03 Cameras and Remote Management The eyes of vision applications and control of the devices once they are installed. Innodisk
The Right APEX for Each Stage
Each card shows why the system fits this stage and which needs it answers; the product page has the full specifications.
Development, Training and On-Premise AI
Office, lab or server room: train and fine-tune models on your own data, build and test LLM applications and run them on premise, without sending data to the cloud.
Innodisk · NVIDIA · workstation-class AI system
APEX-X100
Workstation-class system for building LLM and VLM applications at the desk: 96 GB GDDR7 on the RTX PRO 6000 Blackwell Max-Q, and 10 GbE and PoE for test cameras.
Answers Large models on one system
Product Details
Innodisk · Qualcomm · Cloud AI 100 Ultra
APEX-X100-Q
870 TOPS and 128 GB of accelerator memory on one card: large language models and generative AI run on premise, for testing or serving at the customer.
Answers Large models on one system
Product Details
Innodisk · NVIDIA · GeForce RTX 5080
APEX-X200
Cost-efficient prototyping of generative AI and SLM applications with a GeForce RTX 5080, on the same NVIDIA software stack as the field systems.
Answers Data stays in-housePredictable cost
Product DetailsDeployment in the Field
At the machine, in the vehicle or in an outdoor cabinet: compact systems run the trained model where the data is created.
Innodisk · NVIDIA · RTX 5000 Ada MXM
APEX-P100
RTX 5000 Ada with 5 × 2.5 GbE and 5 × DP++ for multi-camera inspection and imaging applications at the machine, rated 0 ~ 50 °C.
Answers The same platform from lab to field
Product Details
Innodisk · NVIDIA · compact RTX 2000 Ada
APEX-P200
The most compact APEX with an NVIDIA GPU: runs models developed on NVIDIA systems in a 182 × 158.6 × 90 mm chassis at -20 ~ 60 °C.
Answers The same platform from lab to fieldHarsh field conditions
Product Details
Innodisk · Qualcomm · Dragonwing IQ-9075
APEX-A100
100 TOPS for robots, AGVs and vehicles: 9–36 V DC input, -40 ~ 70 °C, four MIPI CSI-2 camera inputs and chipset support through 2038.
Answers Harsh field conditions
Product Details
Innodisk · Intel · Core Ultra with NPU
APEX-E100
Low-power inference with up to 36 TOPS for detection and recognition at many sites; MIPI over Type-C cameras connect directly.
Answers Predictable costHarsh field conditions
Product DetailsCameras and Remote Management
The eyes of vision applications and control of the devices once they are installed.
Innodisk · MIPI CSI-2 · 2 MP ~ 13 MP
MIPI CSI-2 Camera Modules
Low-latency MIPI CSI-2 cameras from 2 MP to 13 MP, validated on NVIDIA Jetson Orin and Intel platforms, with image-quality tuning for your application.
Answers Reliable camera input
Product Details
Innodisk · MIPI over Type-C
MIPI over Type-C Cameras
MIPI image quality over a Type-C cable, supported directly by APEX-E100: a quick start for vision prototypes and field installations.
Answers Reliable camera input
Product Details
Innodisk · Out-of-band management
InnoAgent EZ2N-0XL1
Out-of-band management with its own network link: power-cycles and recovers devices at customer sites even when the operating system has frozen; with iCAP, models and updates reach the whole fleet.
Answers Managing installed devices
Product DetailsCompany-Internal AI with Aetina
A 2U server, a GPU-expansion workstation and an agentic AI system that run in-house without sending data to the cloud.
Aetina · SuperEdge 2U · AI inference / training server
AEX-2UA2
A short-depth 2U server with dual double-width GPUs for generative AI, LLMs and deep learning; with NVIDIA H200 NVL, RTX PRO 6000 Blackwell Server Edition or L40S it runs models in-house.
Product DetailsAetina · MegaEdge · GPU-expansion AI workstation
AIP-FR68S
A GPU-scalable edge AI workstation: PCIe Gen5 x16 for the RTX PRO 6000 Blackwell Workstation Edition and an integrated swappable 2000W power supply.
Product DetailsAetina · MegaEdge · Enterprise 2U edge agentic AI system
AIP-RQ87
A ready-to-deploy 2U edge agentic AI system for AI agents and private LLMs, without the complexity of traditional AI infrastructure.
Product DetailsMakers of the Recommended AI Systems
Industrial Storage, Memory and Edge AI
Innodisk
Industrial SSDDRAM ModulesCamera ModulesNetwork CardsEdge AI Systems
With Innodisk, we offer industrial-grade flash storage, DRAM modules, embedded camera modules, network and I/O expansion cards and edge AI systems. Founded in Taiwan in 2005, Innodisk designs its products for extreme temperatures, shock and vibration, unstable power and long product life cycles; we integrate them into the enclosures and thermal solutions of our other brands.
Edge AI Systems and GPU Accelerators
Aetina
Edge AI SystemsMXM GPU ModulesPCIe GPU CardsAI ServersRemote Management
With Aetina we offer Jetson-based edge AI systems, in-vehicle AI computers, GPU-expansion workstations, short-depth 2U AI servers and MXM / PCIe GPU cards. Founded in Taiwan in 2012 and part of the Innodisk Group since 2013, Aetina develops field-ready edge AI solutions; as its authorised representative in Türkiye we plan these systems together with our enclosure and thermal solutions.
Why Björn Tech?
Right-Sized Hardware
We match the APEX model to your model size, latency target and budget, so you neither overpay in the lab nor run short in the field.
From Prototype to Rollout
Fixed bills of materials and long-term supply let you install the configuration you validated at every customer site.
Ready for the Field
We integrate APEX systems with cameras, storage and POLYRACK enclosures and cabinets, and plan the cooling together with our thermal management partners.
Talk to our engineering team about your project.
Contact Us