MTCA.4 PCIexpress Uplinks - optical and copper
Up to 16 PCIexpress lanes inside and outside of a MTCA.4 system are demanded in application with high data throughput, low latency and huge computing requirements.
This presentation presents how to upgrade existing MTCA.4 systems
inside the system with
- 16 PCIexpress lanes to the local root complex as rear module behind the MCH
at the front-side with
- 16 lanes of optical PCIexpress uplinks (128 Gb/s)
- two times 8 lanes of optical PCIexpress uplinks (each 64 Gb/s)
- daisy chain with 8 lanes of optical PCIexpress uplinks multiple MTCA.4 systems (64 Gb/s)
at the backside-side with
- 16 lanes of electrical PCIexpress uplinks (128 Gb/s)
On new MicroTCA.4 system with new backplanes also a 16 PCIexpress port can be routed to one AMC slot.
Existing standard MTCA.4 system have only 4 PCIexpress lanes per AMC slot, which can become a bottleneck if only one CPU has to process data coming from up to 11 AMCs populated with high performance FPGAs DMAing data directly into the memory of the CPU.
If PCIexpress clustering with up to 6 PCIexpress cluster cannot be used due to the impossibility to distribute data to different CPUs, a bigger data path to the main CPU is needed.
If the AMC CPUs are not powerful enough and/or an GbE and 10 GbE interface creates too much latency, a high speed optical or copper PCIexpress path to a powerful external computer is needed.
In this presentation available solutions to upgrade existing MTCA.4 system with up to 16 PCIexpress lanes to the local CPU and/or to external computers via high performance, low latency PCIexpress uplinks are shown.