F2 PCIe Peer-to-Peer (FPGA FPGA Connectivity)

Question

Do F2 FPGA instances support PCIe P2P connective between FPGAs, similar to F1? The goal is to implement a design that spans across two FPGAs with the lowest latency, which is why PCIe P2P is more appealing than DPDK.

The only information I found was this github issue, but is the response referring to connectivity between the FPGA and the ENA for networking or any P2P connectivity, i.e. between FPGAs?

Thanks a lot!

The only information I found was [this github issue](https://github.com/aws/aws-fpga/issues/725), but is the response referring to connectivity between the FPGA and the ENA for networking or any P2P connectivity, i.e. between FPGAs?

Thanks a lot!

AWS-joeluc · Accepted Answer

Hello!

Thank you for posting your question. My response on the [GitHub issue](https://github.com/aws/aws-fpga/issues/725) you referenced applies to PCIe traffic between the ENA PCIe endpoint and the FPGA PCIe endpoint. I believe this is separate from your question on peer to peer between FPGA endpoints.

F2 does not support peer to peer PCIe traffic directly between FPGAs, which is why we recommend DPDK. To help us better understand your needs, what are you trying to achieve with your application? Is P2P functionality necessary for your use case?

This context would be very helpful for the team. Thanks!

re:Post Agent · Answer

Based on the available information, there is no explicit confirmation that F2 instances support PCIe Peer-to-Peer connectivity between FPGAs in the same way that F1 instances do.

The F2 instances are the second-generation FPGA-powered EC2 instances that offer significant improvements over F1, including more processing power and memory bandwidth. They feature AMD Virtex UltraScale+ VU47P FPGAs with 2.9 million logic cells and 9024 DSP slices, and come in configurations with 1, 2, or 8 FPGAs per instance.

What we do know is that F2 instances support virtual ethernet features and DPDK networking for efficient packet processing. The documentation mentions that F2 instances can achieve high throughput (9 GB/s was measured in testing) using DPDK Software Patch Panel (SPP) running in user space, which bypasses kernel interrupts for maximum performance.

For your use case requiring the lowest latency connectivity between FPGAs, the sources don't specifically address whether PCIe P2P is supported on F2 instances. The information about DPDK suggests that this is a recommended approach for inter-FPGA communication on F2, but there's no clear statement about PCIe P2P support or how it compares to F1's capabilities in this regard.

If your design requires spanning across multiple FPGAs with the lowest possible latency, you might need to contact AWS support directly for definitive information about PCIe P2P support on F2 instances.  
**Sources**  
[Agile satellite communication ground systems with Amazon EC2 F2 FPGA solutions | AWS Public Sector Blog](https://aws.amazon.com/blogs/publicsector/agile-satellite-communication-ground-systems-with-amazon-ec2-f2-fpga-solutions/)  
[Cloud Compute Instances – Amazon EC2 Instance Types – AWS](https://aws.amazon.com/ec2/instance-types/)  
[Now Available – Second-Generation FPGA-Powered Amazon EC2 instances (F2) | AWS News Blog](https://aws.amazon.com/blogs/aws/now-available-second-generation-fpga-powered-amazon-ec2-instances-f2/)  
[F2](https://aws.amazon.com/ec2/instance-types/f2/)  
[How to migrate a VeriFire Emulator design from F1 to to F2 Instances | AWS HPC Blog](https://aws.amazon.com/blogs/hpc/how-to-migrate-a-verifire-emulator-design-from-f1-to-to-f2-instances/)

F2 PCIe Peer-to-Peer (FPGA <-> FPGA Connectivity)

Add your answer

Relevant content