A customer is planning to get real time data from the AD9081 (MXFE) board to a PC and viceversa, with signals with instantaneous bandwidth of 65MHZ, for two channels. This shall amount to about 600-800MB(yte)ps of thruput. The only medium shall eventually be PCI express. Are able to support this trasnfer, thru any Xilinx capture board, from the MXFE JESD interfaces, with proper fpga tested code modules? thanks
currently the Xilinx carriers which support the MxFE are the ZCU102 and VCU118, both support PCIe Gen 3x16 that gives 15.754 GB/s of throughput.
However there is no reference design available for such system from our group. All reference designs stream the channels from and to the local DDR and can export it through Ethernet.
Hi Laszlo, as Ethernet would not provide enough speed for real time; we could ask the customer to develop the bridge from the local DDR to PCI-E. Could we support the customer to do that, ie provide enough info to them, to take the data from the memory to PCI-E (assuming related drivers are available from Xilinx or else)? thanks
I'm the "customer" with the problem rised by @snapolit.
I think channel streaming through the DDR is ok if we can do it in real-time, but the gigabit ethernet doesn't support the throughput that we need.
Is there a driver/firmware to export the channel data from/to DDR memory through PCIe (or SFP+ port) instead Ethernet?
Otherwise, is there a carrier that supports the MxFE JESD and with a 10GbE interface?
we really don't have expertise in PCIe, in fact I was mistaken in my previous post regarding ZCU102. I think you can't connect the ZCU102 to the PC. The ZCU102 has root ports that can accommodate PCIe boards. Not the other way around.
We can only support bringing the samples to the application layer (in fabric of the FPGA) or to local memory.
The ZCU102 supports 10G Ethernet on SFP see
However I think that in Ethernet approach first you need to move the samples into the PS DDR and do the framing there. Software overhead might be too big. see some performance stats here.
In PCIe approach you could stream directly the samples with a DMA for PCIe to the PC memory (https://www.xilinx.com/products/intellectual-property/pcie-dma.html)