I've had some fun learning how to implement various bits and pieces of networking on FPGAs as a hobby for a while, and while boards like this that focus on gigabit network are fine, the fact is that there are a lot of FPGA boards with gigabit and 100Mbps interfaces. What there are not enough of are low cost boards that can do 2.5Gbps, 5Gbps and 10Gbps. Lattice has some very affordable FPGAs with 5Gbps SERDES, and their newer 10Gbps capable chips remain extremely affordable.
One of the things I would absolutely love to have are a couple of FPGAs boards in SFP and QSFP form factors. Why might you ask? Because it would be seriously useful to have a PPPoE / L2TP data plane to plug into the port of a 100Gbps capable switch for use in the network edge. Modern ethernet switches have plenty of Layer 3 networking capabilities, but most switch vendors fail to expose any functionality for these protocols even though the underlying ASICs often enough have the capability to handle them. Sure, you'll never see these protocols in a cloud data center, but plenty of incumbent telecoms make use of them in their FTTP networks due to the legacy of xDSL deployments and the need to support wholesale access to those networks. Sadly, developing such a board is beyond my hobbyist electronics capabilities, but I'd have no problem bashing a bunch of Verilog / VHDL into shape to make it work in fairly short order... I just hope it uses an FPGA like the Polarfire for which the SERDES are about 100x easier to use than the gawd awful Xilinx 7 series (KC705, I'm glaring at you for eating weeks of my hobbyist life to that bring up).
You can absolutely do this using something like a CertusProNX from Lattice (I much prefer it to the Polarfire), mounting an SFP/SFP+ cage onto the FMC connector of the board, and wire their transceiver lanes to the FPGA SERDES pins. HiTech Global has a 4-port SFP/SFP+ FMC module. I believe there are also QSFP mezzanine cards but I haven't looked much into that. ISI or Trenz probably make something.
I already have a 4 x SFP+ FMC board attached to my KC705, and that is what I have used for a bunch of development. However, that is the opposite of what I want to do. I want the FPGA to be inside of the SFP+ / QSFP module to be able to plug it into a switch without external hardware.
Microchip has an app note on putting a Polarfire into an SFP+ module complete with a board layout, but nobody is building them as near as I can tell. Lattice has a similar app note, but, again, they only sell the much larger eval boards.
Thank you for all the great comments and feedback so far. Note that Betsy uses an Altera Cyclone 10 LP (not GX). This is a low cost, general purpose FPGA. The Ethernet PHY interface is RGMII, which utilizes 5 bits of parallel DDR + clock instead of SERDES.
As many of you probably know, SERDES and specialized PCS/CDR blocks will get you well past 1 GigE, but 1 GigE for RGMII is challenging with 125 MHz single-ended traces.
The project compiles super fast with the Quartus tools and Signal Tap enabled with several active configurations. Quartus bundles the Questa simulator, so there is a great environment for simulation.
Regarding Certus-NX, this indeed would also be a great choice. Lattice does a very nice job exposing their I/O primitives, and I believe the RGMII DDR could be instantiated directly in the I/O cells for both input and output (this could definitely be accomplished with the earlier ECP5). We actively design with Certus-NX, and a future Betsy revision using it is very possible.
1) As others mentioned, two GbE interfaces seems really limited for a 2025 project. Modern FPGAs can support 100GbE and up - I don't necessarily expect that on a hobbyist-level project, of course, but 1GbE is well behind the curve.
2) There don't appear to be any hardware design files (e.g. schematics, PCB layouts) in the Git repository. In fact, the only mention of the current FPGA is a single text file stating that "Cyclone 10 GX port in progress"...
3) There's basically zero open source support for Intel/Altera FPGAs. Yes, you can open-source your HDL, but the vendor tools are all closed-source and there's no alternatives.
Agreed. They are using an Cyclone 10 GX - that's over $250 for a part from 2017...
If the restriction is wanting low price to suite the FPGAs open-source low-budget market, they'd be better off using a Lattice Certus-NX or something. 5Gbps SERDES on that for ~$40, or better yet a CertusPro-NX with 10Gbps SERDES for ~$70. Altera and Xilinx are just throwing away the sub-100K-LUT market to Lattice at this point, yet people are still building systems out using these expensive, antiquated parts. That's shelf pricing too - go through a distributor and it'd be 50% of that price!
I've had some fun learning how to implement various bits and pieces of networking on FPGAs as a hobby for a while, and while boards like this that focus on gigabit network are fine, the fact is that there are a lot of FPGA boards with gigabit and 100Mbps interfaces. What there are not enough of are low cost boards that can do 2.5Gbps, 5Gbps and 10Gbps. Lattice has some very affordable FPGAs with 5Gbps SERDES, and their newer 10Gbps capable chips remain extremely affordable.
One of the things I would absolutely love to have are a couple of FPGAs boards in SFP and QSFP form factors. Why might you ask? Because it would be seriously useful to have a PPPoE / L2TP data plane to plug into the port of a 100Gbps capable switch for use in the network edge. Modern ethernet switches have plenty of Layer 3 networking capabilities, but most switch vendors fail to expose any functionality for these protocols even though the underlying ASICs often enough have the capability to handle them. Sure, you'll never see these protocols in a cloud data center, but plenty of incumbent telecoms make use of them in their FTTP networks due to the legacy of xDSL deployments and the need to support wholesale access to those networks. Sadly, developing such a board is beyond my hobbyist electronics capabilities, but I'd have no problem bashing a bunch of Verilog / VHDL into shape to make it work in fairly short order... I just hope it uses an FPGA like the Polarfire for which the SERDES are about 100x easier to use than the gawd awful Xilinx 7 series (KC705, I'm glaring at you for eating weeks of my hobbyist life to that bring up).
You can absolutely do this using something like a CertusProNX from Lattice (I much prefer it to the Polarfire), mounting an SFP/SFP+ cage onto the FMC connector of the board, and wire their transceiver lanes to the FPGA SERDES pins. HiTech Global has a 4-port SFP/SFP+ FMC module. I believe there are also QSFP mezzanine cards but I haven't looked much into that. ISI or Trenz probably make something.
I already have a 4 x SFP+ FMC board attached to my KC705, and that is what I have used for a bunch of development. However, that is the opposite of what I want to do. I want the FPGA to be inside of the SFP+ / QSFP module to be able to plug it into a switch without external hardware.
Microchip has an app note on putting a Polarfire into an SFP+ module complete with a board layout, but nobody is building them as near as I can tell. Lattice has a similar app note, but, again, they only sell the much larger eval boards.
Thank you for all the great comments and feedback so far. Note that Betsy uses an Altera Cyclone 10 LP (not GX). This is a low cost, general purpose FPGA. The Ethernet PHY interface is RGMII, which utilizes 5 bits of parallel DDR + clock instead of SERDES.
As many of you probably know, SERDES and specialized PCS/CDR blocks will get you well past 1 GigE, but 1 GigE for RGMII is challenging with 125 MHz single-ended traces.
The project compiles super fast with the Quartus tools and Signal Tap enabled with several active configurations. Quartus bundles the Questa simulator, so there is a great environment for simulation.
Regarding Certus-NX, this indeed would also be a great choice. Lattice does a very nice job exposing their I/O primitives, and I believe the RGMII DDR could be instantiated directly in the I/O cells for both input and output (this could definitely be accomplished with the earlier ECP5). We actively design with Certus-NX, and a future Betsy revision using it is very possible.
NetFPGA has been around since 2007, with NetFPGA-1G. They have some very fancy offerings these days, well past the 4 x 1Gbe they started with.
https://netfpga.org/ https://en.wikipedia.org/wiki/NetFPGA
Maybe I'm just jaded and demanding, but:
1) As others mentioned, two GbE interfaces seems really limited for a 2025 project. Modern FPGAs can support 100GbE and up - I don't necessarily expect that on a hobbyist-level project, of course, but 1GbE is well behind the curve.
2) There don't appear to be any hardware design files (e.g. schematics, PCB layouts) in the Git repository. In fact, the only mention of the current FPGA is a single text file stating that "Cyclone 10 GX port in progress"...
3) There's basically zero open source support for Intel/Altera FPGAs. Yes, you can open-source your HDL, but the vendor tools are all closed-source and there's no alternatives.
Agreed. They are using an Cyclone 10 GX - that's over $250 for a part from 2017...
If the restriction is wanting low price to suite the FPGAs open-source low-budget market, they'd be better off using a Lattice Certus-NX or something. 5Gbps SERDES on that for ~$40, or better yet a CertusPro-NX with 10Gbps SERDES for ~$70. Altera and Xilinx are just throwing away the sub-100K-LUT market to Lattice at this point, yet people are still building systems out using these expensive, antiquated parts. That's shelf pricing too - go through a distributor and it'd be 50% of that price!