Cascade 100P
PCI Express AI Accelerator Card
High-density MemryX AI acceleration for servers and compute platforms with high-volume processing requirements. Now accepting purchase orders.
At a glance
Chips
16× MX3 AI Accelerator Chips
AI performance
Weight storage
Power
form factor
Best for
High-volume servers and PC’s
The Cascade 100P brings high-density MemryX AI acceleration to servers and compute platforms with high demand.
Powered by 16 MX3 accelerators, the Cascade 100P delivers substantial AI inference capacity in a standard PCIe card, offloading AI workloads from the host processor. Designed for higher-throughput and multi-stream applications, it provides a straightforward way to scale dedicated AI processing within existing systems.
PRODUCT PROOF
Purpose-built AI acceleration.
The key hardware characteristics of MX3 Edge AI Accelerator at a glance.
100
TFLOPS
AI performance
16x
MX3
AI accelerator chips
168 MB
On-chip
model weight storage
What you get
Dedicated AI inference that works alongside your host.
Designed for flexible integration
Bring dedicated MemryX AI acceleration into your system with an efficient design that balances performance, power, and thermal requirements.
Dedicated AI offload
Offload neural-network inference to MX3 while preserving host compute for the rest of your application. Unified software experience
Unified software experience
Use the same MemryX software workflow—from model compilation through runtime deployment—across single- and multi-chip configurations.
Real-time batch-1 inference
MX3’s pipelined dataflow architecture delivers low-latency inference without relying on large batch sizes.
Inside the architecture
Why MX3 works differently
At-memory compute and a pipelined dataflow architecture deliver efficient, consistent AI inference without requiring developers to redesign their applications around the accelerator.
Accuracy
No retraining required
BFloat16 activations preserve model accuracy through direct recompilation, without requiring retraining or hand-tuning.
Deterministic
Consistent batch-1 latency
A pipelined dataflow architecture keeps multiple inputs in flight at once, maintaining high hardware utilization without batching for asynchronous, low-latency inference.
Integration
Designed to work with your host
MemryX operates as dedicated AI acceleration, offloading neural-network inference from the host processor.
Architecture
At-memory dataflow
The architecture keeps model data close to compute, enabling high-throughput AI processing by minimizing costly data movement.
Flexibility
Build beyond a single model
Run models individually or combine them through concurrent, pipelined, or conditional execution so developers can build around the application — not the accelerator.
Deployment
Built to scale
Scale from compact deployments to multi-chip configurations with a common software foundation as performance requirements grow.
Specifications
Cascade 100P at full detail
Product
Part Number
Accelerators
AI performance
100 TFLOPS
Weight storage
Form factor
dimensions
input voltage
Power
5 W – 50 W
Interface
PCI Express Gen 3 and above
Frameworks
ONNX, TensorFlow, Keras, TFLite
Certification
CE/FC Class A
Operating temperature
-40ºC to 85ºC Industrial
Operating Humdity
10% ~80% RH non-condensing
target systems
Applications
Built for real-world AI
Video Management Systems
Industrial & Robotics
Transportation
Smart Devices
Start building
Everything you need to get started.
Install the MemryX SDK, compile a model, and start running accelerated inference with open-source tools and resources.
Documentation
Get Started
Install the SDK, run your first inference on the Pi, then compile your own model.
Support
Community Forum
Ask questions, share builds and get answers from the MemryX AI community.
Scale AI Inference inside your server.
Talk to MemryX about Cascade 100P deployment, system capability, and purchasing.
