Introduction
FPGAs (Field Programmable Gate Arrays) are crucial components in advanced high-performance and compute-intensive systems used for AI, networking, data centers, aerospace, defense, and more. Choosing the right FPGA often depends on factors like logic density, performance, power efficiency, features, and cost.
This article compares two popular FPGA families from Achronix Semiconductor – the Speedster7t and the newer Speedster22i HD. We will explore the architecture, capabilities, and applications of these FPGAs to understand how they differ and which works best for your requirements.
Overview of Achronix Speedster FPGAs
Achronix Semiconductor is a pioneering company focused on high-performance FPGAs for advanced computing markets. Their Speedster FPGA family is optimized for data acceleration applications with very high throughput and bandwidth needs.
Some key aspects of Achronix FPGAs:
- Utilize a 2D network-on-chip (NoC) architecture
- Have high-density and high-speed SerDes interfaces
- Include hard PCIe blocks, DDR memory controllers
- Feature high-bandwidth GDDR6 memory interfaces
- Offer dedicated AI engines and ML processors
- Provide leading-edge process nodes down to 7nm
The Speedster7t and Speedster22i HD represent two generations of Achronix FPGAs aimed at high-end applications.
Achronix Speedster7t FPGA
The Speedster7t family was the first generation of FPGAs built on the innovative 2D network-on-chip fabric by Achronix. Key highlights:
Process Node: 16nm FinFET
Logic Cells: Up to 1.5 million cells
Memory: Up to 68Mb embedded RAM, 2GB external GDDR6
Interfaces: 36 – 400GbE ports, 16 – 32GB/s SerDes lanes
AI Engines: 4 – 16 INT8 AIE processors
Performance: Up to 33 TMACs AI performance
Power: Typically <75W envelope
As a 16nm FinFET product, the Speedster7t series focuses on providing a balanced set of features including high density, SerDes interfaces, and built-in AI acceleration at an optimal power envelope.
Achronix Speedster22i HD FPGA
The Speedster22i HD is the latest generation from Achronix built using the advanced 7nm process node. It pushes the performance and density envelope further. Key features:
Process: 7nm FinFET
Logic Cells: Up to 4.5 million cells
Memory: Up to 576Mb RAM, 8GB GDDR6
Interfaces: 112 400GbE ports, 64 32Gb/s SerDes lanes
AI Engines: Up to 128 INT8 AIE cores
Performance: Up to 1 PetaMACs AI performance
Power: Up to 300W envelope
The Speedster22i HDT leverages the 7nm node to significantly boost logic capacity, bandwidth, AI performance and DSP capabilities for cutting-edge applications. The higher power envelope allows leveraging the density and speeds.
Head-to-Head Comparison
Here is a direct side-by-side comparison of the major specifications and capabilities of the two FPGAs:
| Parameter | Speedster7t FPGA | Speedster22i HDT FPGA |
|---|---|---|
| Process | 16nm FinFET | 7nm FinFET |
| Logic Cells | Up to 1.5 M | Up to 4.5 M |
| Embedded RAM | Up to 68 Mb | Up to 576 Mb |
| External Memory | 2GB GDDR6 | 8GB GDDR6 |
| DSP Slices | 25,200 | 132,000 |
| AI Engines | 4-16 INT8 cores | Up to 128 INT8 cores |
| AI Performance | 33 TMACs | 1 PetaMACs |
| SerDes | 16-36 @ 32Gb/s | 64 @ 32Gb/s |
| Ethernet Ports | 36x 400GbE | 112x 400GbE |
| Transceivers | 48 – 400GbE | 112 – 400GbE |
| Typical Power | < 75W | Up to 300W |
Key Differences
Based on the above comparison, we can summarize the main differences between the two FPGAs:
- Density – The Speedster22i HDT offers 3X higher logic capacity and 8X more embedded memory compared to the Speedster7t.
- Performance – The 7nm Speedster22i provides over 30X higher AI performance with up to 128 INT8 AIE cores vs just 16 cores in Speedster7t.
- Memory – Speedster22i has 8GB of cutting-edge GDDR6 memory compared to 2GB in previous generation.
- Interfaces – Speedster22i doubles the number of 32Gb/s SerDes lanes and triples the 400GbE transceiver count.
- Bandwidth – With over 3X more DSP slices and higher memory bandwidth, Speedster22i enables much higher overall system bandwidth.
- Power – The Speedster22i lacks power efficiency with up to 300W envelope, 4X more than the Speedster7t.
- Cost – Being a newer high-end 7nm product, the Speedster22i lineup carries a higher cost over the older 16nm Speedster7t.
Recommended Use Cases
Based on their capabilities, here are some ideal use cases for the FPGAs:
Achronix Speedster7t – Mainstream data center accelerators, network cards, industrial automation, defense systems, testing/prototyping of FPGA designs.
Achronix Speedster22i – Advanced AI acceleration, machine learning training, high-frequency trading, genome sequencing, aerospace computing, high-end networking, military systems.
The Speedster22i HDT is better suited for cutting-edge applications needing maximum performance and bandwidth in sectors like AI, cloud computing, networking, and high-end computing. The Speedster7t provides a more optimized solution for cost-sensitive or mid-range applications across defense, automation, 5G systems.
Conclusion
The Achronix Speedster22i HD FPGA offers significantly higher density, performance, bandwidth, and memory compared to the previous generation Speedster7t FPGAs. The 7nm process allows packing in more logic, memory, DSP blocks, and AI engines into the Speedster22i while improving energy efficiency. It represents the bleeding-edge of capabilities for accelerating challenging workloads like AI-inference.
On the other hand, the Speedster7t still powers a wide range of mainstream applications with its proven architecture. For systems with tighter power budgets or lower compute needs, the Speedster7t likely represents the cost-optimized option. Understanding these key differences allows selecting the right Achronix Speedster FPGA for your specific application and performance requirements.
Frequently Asked Questions
What process nodes are used to manufacture these FPGAs?
The Speedster7t uses a 16nm FinFET process while the Speedster22i leverages a more advanced 7nm FinFET node to provide a significant density and performance boost.
How do the embedded RAM and external memory resources compare?
The Speedster22i HD has almost 10X more embedded RAM at 576Mb vs 68Mb in Speedster7t. It also offers 8GB of cutting-edge GDDR6 external memory compared to just 2GB GDDR6 in the previous FPGA.
What machine learning capabilities are included in the FPGAs?
Both FPGAs contain dedicated AI Engines (AIE) for accelerating neural network inferencing workloads. The Speedster22i integrates up to 128 INT8 AIE cores delivering up to 1 PetaMACs of AI performance – over 30X more than the Speedster7t.
How many high-speed SerDes lanes are available in each of the FPGAs?
The Speedster7t offers between 16-36 lanes of 32Gbps SerDes while the Speedster22i doubles this number to 64 lanes of 32Gbps SerDes. This provides much higher aggregate bandwidth.
What is the maximum power consumption for the two FPGAs?
The Speedster7t typically consumes less than 75W making it suitable for mainstream and low power applications. In comparison, the Speedster22i can consume up to 300W to deliver maximum performance, so it is targeted at data center and HPC type workloads.
