08:35 <GitHub157> [artiq] gkasprow commented on issue #813: @sbourdeauducq ... https://github.com/m-labs/artiq/issues/813#issuecomment-329413563

10:47 <rjo> _florent_: it's probably best if you implement the hardware changes as well. and does greg have a bitstream that gives him the lowest possible jesd rate so he can play with the emphasis and look at it on the scope? or does that not make sense?

10:49 <rjo> sb0: do we have another recent full vivado license that i can have?

10:54 <sb0> rjo, for ultrascale?

10:54 <sb0> no

11:02 <rjo> sb0: are there different "full" licenses?

11:03 <sb0> we don't actually have any "full" license. it's device-locked to ku040

11:03 <sb0> and another one is locked to 7k325t

11:07 <rjo> ah. i thought the licenses that come with the eval kits were full ones.

11:47 <larsc> full license costs probably more than the eval kit

11:48 <sb0> yes

11:48 <sb0> for many eval kits that's the case

11:53 <GitHub181> [artiq] sbourdeauducq pushed 2 new commits to rtio-sed: https://github.com/m-labs/artiq/compare/1b61442bc339...490c9815a295

11:53 <GitHub181> artiq/rtio-sed 181cb42 Sebastien Bourdeauducq: rtio/sed: centralize all layouts in one file

11:53 <GitHub181> artiq/rtio-sed 490c981 Sebastien Bourdeauducq: rtio/sed: add TSC/gate (untested)

12:40 <rjo> well. the edu full license cost is 0 in batches of 25 licenses. ;)

12:51 <whitequark> i've recently heard that some software vendors don't check for a .edu TLD. they check for a .edu anywhere in the domain, or even anywhere in the email address... could be handy

12:55 <rjo> many edu's don't even have edu in their domain. and that works fine.

12:55 <rjo> i think for xilinx it's a manual check

13:31 <rjo> i think the general solution to xilinx licensing is to let the OS rename the interface as most do (eth0 -> enp1s0) and then have it take the loopback HOSTID=0000000000....

13:48 <larsc> they only check eth0?

13:49 <rjo> from what i can see yes.

13:49 <rjo> i had to create a tap just for that....

19:02 <d_n|a> How high do you expect the minimum latency for a 32 bit word through a pair of SFPs on Kasli to be, from user fabric to user fabric (i.e. including the tranceivers)?

19:06 <d_n|a> (I'm just pondering the feasibility of splitting something like nu-servo into two parts over a dedicated fibre, with only minimal framing/…. This is definitely not on our short-term agenda, but it would still make a lot of sense in terms of cabling/….)

20:30 <rjo> d_n|a: about 205 UI (https://www.xilinx.com/support/answers/58981.html) if you tweak everything, i.e. some 20.5-205 ns depending on speed. plus cable (<100ns). plus sfp (probably <10ns).

20:36 <rjo> ah. 6.6 Gb/s max for GTP. 31 ns min.

20:37 <rjo> but that's all best case and does not account for CDCs