> > Why do you think that CPUs have negative value? You don't have to use > them > > at all. > > I already explained that the MPSoC has to be dealt with and cannot be > completely ignored. If we have two SDRAM systems, maybe we can to a > reasonable > extent, but then it does complicate the board. > I clearly see your point. The only reason to use here the SoC is its price and design reuse. I can prepare bootloader configuration that will keep the CPUs on hold and dedicate all the SDRAM controller to your needs. In this way you won't need to waste any logic resources and forget about the CPUs.
> > > Anyway, to build MCH, we need first tongue. We can either build it or buy > > it. Buying is the simplest option at the moment and we get gigabit > switch > > and management. > > > On the MCH Tongue 3, there is single port Ethernet connection that can be > > used for slow control and remote upgrade. > > We do not need Ethernet (nor IPMI in fact) inside the crates. The only > part of > the system that has Ethernet is between the control PC and the root master > (Metlino); this can go over one of its regular SFPs. > this can also go over embedded MAC, CPUs and SFPs as well. It's easy to use this MAC in bare-metal way. You will save logic resources. This MAC is working well, I managed to get 900Mbit/s under TCP/IP. And you get the IEEE1588 for free. And you have 4 such MACs. I don't know how many logic resources do you plan to use in Metilno, but at some point some non time-critical stuff can be offloaded by the CPUs. > > > 6. MCH Tongue 2 clock distributor(Metlino). We can use NAT clock module > with > > two low jitter crossbar switches. > > Why do we need a fancy clock module? Data transfer certainly does not > require > one. Are you reconsidering your position that the AMC backplane should not > be > used for DAC clocks? > Well, during last discussion there is an idea to use these clocks for less demanding applications. Moreover such clocks + MLVDS lines make sub-ns synchronisation of all AMC cards trivial. > > Also, (5 data/clock signals)*(10 AMCs)*(2 for differential) = 100 pins, > plus > power supply. Do we really need all those "tongues"? > This is good question:) Tongue 3/4 has pins for communication - x4 fat pipes between AMCs and MCH. Tongue 2 has only clock pins - 3 sets of independent clocks to all AMCs Tongue 1 has supplies, I2Cs and x1 Ethernet ports to all AMCs. How do you count these 5 data/clock signals ? > Sébastien > >
_______________________________________________ ARTIQ mailing list https://ssl.serverraum.org/lists/listinfo/artiq
