Stratix 10 NX FPGA Module for AI and Vector Processing
The Stratix 10 NX module marks a major step forward in AI acceleration and vector processing for embedded and high-bandwidth systems. Gidel brings this architecture to ready-to-use module and PCIe board platforms. Developers can therefore implement high-throughput AI and vector-processing workloads in compact FPGA systems.
Proc10N and Proc1C10N: Stratix 10 NX Platforms
Gidel has released its Proc10N FPGA Module and Proc1C10N PCIe Board, the first commercial platforms built around the Altera Stratix 10 NX.
These compact, high-performance solutions target compute-intensive applications that require high data rates and low latency. Stratix 10 NX integrates AI Tensor Blocks for deep-learning acceleration. It also combines them with high-bandwidth HBM2 memory to provide high compute density and memory throughput.
| Area | Proc10N | Proc1C10N |
|---|---|---|
| Platform Type | Compact FPGA module | PCIe FPGA accelerator board |
| FPGA Architecture | Altera Stratix 10 NX | Altera Stratix 10 NX |
| AI Acceleration | Stratix 10 NX AI Tensor Blocks | Stratix 10 NX AI Tensor Blocks |
| High-Bandwidth Memory | Integrated HBM2 architecture | Integrated HBM2 architecture |
| System Integration | Integrates with a Gidel carrier board or customer-designed carrier | Installs directly into a PCIe host system |
| Custom Hardware Integration | Well suited to compact embedded and custom carrier-board designs | Well suited to PCIe host-based acceleration systems |
| Evaluation and Development | Supports evaluation, FPGA development, and transition to custom module-based systems | Provides a ready-to-use PCIe platform for evaluation and application development |
| Typical System Fit | Compact embedded, customized FPGA, and high-bandwidth module-based systems | PCIe acceleration, AI, vector processing, and high-bandwidth host systems |
Gidel, a Titanium member of the Intel Partner Alliance, was among the first companies to bring this FPGA architecture to a commercially available module and PCIe platform.
AI, Vector Processing, Security, and High-Bandwidth Applications
The combination of AI Tensor Blocks and HBM2 memory enables Stratix 10 NX to address real-time, high-throughput workloads. Proc10N and Proc1C10N can support applications such as AI inference, video analytics, cybersecurity, 5G processing, radar, and other bandwidth-intensive algorithms.
The compact platform architecture also supports embedded, low-power, and weight-constrained systems where high compute density and high-bandwidth data movement are important design requirements.
High-Bandwidth Performance of the Proc10N Module
The Proc10N module supports 1,600 Gb/s of customizable I/O, up to 143 INT8 TOPS through the Stratix 10 NX AI Tensor Block architecture, and 400 GB/s of HBM2 throughput.
This processing capability supports convolution layers, FFTs, encoders, linear algebra, filtering, and other highly parallel workloads. Despite its performance, the module remains compact at only 97.4 × 101 mm. This supports integration into space-constrained systems.
Intel reported up to a 9.5× acceleration factor for LSTM batch processing compared with an NVIDIA V100 GPU in its Stratix 10 NX benchmark testing.
FPGA Development with ProcVision and Proc Dev Kit
Gidel’s modular development tools help developers build FPGA-based systems while reducing integration effort, project risk, and development complexity. The Proc10N module can be integrated with Gidel carrier boards or with a user-designed carrier through the Proc Dev Kit.
The ProcVision Suite supports FPGA development and system integration through a unified environment for acquisition, buffering, FPGA flows, debugging, simulation, and host communication. Developers can therefore focus more effort on application algorithms rather than low-level infrastructure.
The Proc Dev Kit supports hardware configuration and real-time system testing. It generates Application Support Packages (ASP) matched to each system configuration. These packages help accelerate FPGA-to-host integration and application development.
ProcVision also supports distributed multi-unit acquisition, enabling synchronized capture across multiple cameras or sensors. This capability can support robotics, autonomous systems, 360° imaging, high-volume inspection, and other distributed acquisition architectures.
The Mini FPGA Modules portfolio also includes Proc10N, supporting compact, high-performance FPGA system integration.
Proc10N Availability and Custom FPGA Designs
Proc10N and Proc1C10N can also serve as evaluation and development platforms for engineers assessing Stratix 10 NX technology before moving to custom FPGA system designs. Gidel and its partners support developers evaluating Stratix 10 NX platforms. They also assist with new FPGA-based systems that require high-bandwidth I/O, HBM2 memory, AI acceleration, or vector processing.
Developers can contact Gidel for assistance with platform selection, carrier-board integration, FPGA development, and custom system design.
