<gruetzkopf>
i heard rumors that that meeting would happen, but i know nothing about the how, where, why
bibor has quit [Client Quit]
<edmoore>
just check that lattice dipolmats are not leaving from the back door with lots of diplomatic bags
<q3k>
i can confirm that edmund and clifford left the meeting unscathed
<daveshah>
We think....
<q3k>
no people with bonesaws were seen walking into the building
<gruetzkopf>
did anyone xray them and check if their brains now contain ECP-4?
bibor has joined ##openfpga
Miyu has joined ##openfpga
<whitequark>
ECP-4 more like SCP-4
<q3k>
whitequark:
<whitequark>
hahaha
<gruetzkopf>
12 rusty keys and a door?
<whitequark>
that sounds like an old naval song
<gruetzkopf>
SCP-004
bibor has quit [Client Quit]
rohitksingh has joined ##openfpga
bibor has joined ##openfpga
pie__ has joined ##openfpga
pie___ has quit [Remote host closed the connection]
bibor has quit [Client Quit]
<swetland>
If I had to make a bet on who would publicly if not embrace, but at least acknowledge and work with, open source tooling first, I'd lean toward Lattice -- stuff already working with their chips so they can see demonstrable interest/value, and the FPGA world is basically Xilinx and Altera(Intel now) and everyone else... folks down in the "other" category would seem more likely to take a risk to stand out
bibor has joined ##openfpga
<swetland>
after them, maybe Intel, because Altera-as-part-of-Intel, I suspect, may be less focused on revenue-from-tools.
<daveshah>
Even Xilinx make ~no revenue from tools
<tnt>
Yeah, they're all about chips. All design house I worked with always got the xilinx tools for free from their fae.
<daveshah>
I suspect they charge for Vivado for that very reason - they can give it away for "free" to make big customers feel special
<_whitenotifier>
[whitequark/Glasgow] whitequark pushed 3 commits to master [+0/-0/±19] https://git.io/fp55h
<_whitenotifier>
[whitequark/Glasgow] whitequark b94da9b - gateware.boneless: use explicit ALU datapath (-16 CARRY, -11 LUT).
<marcan>
have you considered, like... manually implementing parts of it as 4-LUTs to have a baseline human optimization? :P
<whitequark>
yes. it is not necessary
<whitequark>
the ALU is actually close to optimal, if not optimal
<cr1901_modern>
or better yet, extract all the combinational circuits, run a QM solver for a few days :P
<whitequark>
±1 LUT
azonenberg_work has quit [Ping timeout: 246 seconds]
rohitksingh has quit [Ping timeout: 240 seconds]
wpwrak has joined ##openfpga
pie___ has joined ##openfpga
pie__ has quit [Remote host closed the connection]
m4ssi has joined ##openfpga
azonenberg_work has joined ##openfpga
<azonenberg_work>
daveshah: re support costs
<azonenberg_work>
i think xilinx already has nothing but forum support for webpack / non-industrial users
<daveshah>
I know Lattice's support, at least officially, is predicated on whether your email address is commercial or not
<qu1j0t3>
that's... odd?
azonenberg_work has quit [Ping timeout: 245 seconds]
m4ssi has quit [Remote host closed the connection]
rohitksingh has joined ##openfpga
azonenberg_work has joined ##openfpga
catplant has quit [Ping timeout: 252 seconds]
catplant has joined ##openfpga
jcarpenter2 has quit [Ping timeout: 252 seconds]
jcarpenter2 has joined ##openfpga
pie__ has joined ##openfpga
pie___ has quit [Remote host closed the connection]
catplant has quit [Ping timeout: 246 seconds]
catplant has joined ##openfpga
<bubble_buster>
I thought the main issue was proprietary bitstream, not tool license revenue
catplant has quit [Read error: Connection reset by peer]
<bubble_buster>
that ship has at least partially sailed and sunk for lattice right?
<bubble_buster>
may as well pivot instead of continuing to lose to the 2 giants at their own game
catplant has joined ##openfpga
<hl>
you'd think, but FPGA companies seem incapable of understanding this
<hl>
who's to say lattice won't continue keeping their head in the sand?
<hl>
the case to get a competitive advantage by opening tools was apparently once made to an executive of Xilinx and they just don't get it, apparently
Sellerie has quit [Quit: leaving]
<bubble_buster>
I'm tempted to make my own open fpga but I figure by the time I get something going, someone like lattice could flip the switch to being open
<Finde>
there are two open FPGAs under DARPA's POSH program
<Finde>
we'll see where they go...
<bubble_buster>
nice link, thanks
<mumptai>
anyone seen a usefull open hardware zynq design?
<ZipCPU>
Mumptai: Nothing public, only an open source AXI-lite slave and set of formal properties
<mumptai>
i meant schematic and layout
<mumptai>
(preferably kicad)
* ZipCPU
waits for another to answer
<mumptai>
but than, is it worth the effort to design this?
<mumptai>
how many amateurs would be "dedicated" (read crazy) enough to tempt this. in the end the board would have at least 2 or 3 BGAs and half a dozen power converters
<swetland>
https://www.crowdsupply.com/krtkl/snickerdoodle is a zynq platform, they don't mention a specific license but claim that schematics, bom, gerbers, etc will be published in answer to an "open source" question
<mumptai>
gerbers are kinda useless, i dont' want to copy a thing 1:1, i want to modify it
<sorear>
how different are the zynqs and non-zynqs from a board designer POV?
<mumptai>
power sequencing is compable to big fpgas, booting is more complex but that is handled in the toolchain, its basically the need to setup lpddr2 or ddr3 correctly
<mumptai>
so "a bit" of tricky signal and power integrity stuff
<mumptai>
its not tooooo complicated, just annoying and expensive to fix
<mumptai>
gerbers might be nice to check how others handled that
<swetland>
from the fpga side the zynqs are basically an artex or kintex -7 series with a PS7 block (the A9 complex, ddr controller, peripherals, etc exposed via a set of AXI ports)
<mumptai>
i used it before
<mumptai>
its kinda complex, but okay, its not that much more pain complared to the FPGA + MCU experience
<swetland>
Mostly just that it's an A-series core
<swetland>
if you're working bare metal they're a bit more involved than M-series to bring up
<swetland>
if you're using linux, well then you're dealing with linux
Bob_Dole has quit [Ping timeout: 250 seconds]
<swetland>
but nothing particularly magical, just a lot of little details
<hl>
5
<hl>
5
<hl>
...whoops
catplant has quit [Ping timeout: 268 seconds]
Sellerie has joined ##openfpga
Sellerie is now known as Sellerie
catplant has joined ##openfpga
rohitksingh has quit [Ping timeout: 246 seconds]
rohitksingh has joined ##openfpga
<mumptai>
swetland, most cortex-m are a walk in the park in comparison
<swetland>
yup
<mumptai>
arm7tdmi was a pita, cortex-m is way better
<swetland>
I actually am not a fan of the somewhat overly clever thread/interrupt mode stuff on -M as it's still a bit of a headache to mesh up with traditional task switching. but certainly no worse than 7TDMI
<mumptai>
cortex-a with zynq has quiet some challenges, but ain't that bad once the initialization is done (also true for loading the PL via fsbl)
<swetland>
the debug situation is waaaay nicer with SWD and the v5 debug interfaces
<swetland>
as opposed to pushing instructions into the pipeline with jtag
catplant has quit [Ping timeout: 244 seconds]
<TD-Linux>
yeah one way to approach the cortex-m is just to set it up for 1 priority and no interrupt preemption
<TD-Linux>
the intersection between "my application needs to have preemptible interrupts" and "my application needs to be correct to be safe" worries me
catplant has joined ##openfpga
<mumptai>
hmmm, should i put a zynq minimal design in my hobby task queue?
<mumptai>
like a 7010 + 16bit ddr3 and something to boot from
<swetland>
it'd be neat to see something like that out there
<swetland>
I'd lean toward dropping an HDMI output and an ethernet PHY + MagJack on such a critter
<swetland>
as both are pretty trivial to drive and extend the reach of the base platform a bunch
<adamgreig>
trenz make a bunch of similar sorts of things
<adamgreig>
seems like it'd be fun to diy though
<mumptai>
i don't see any reason to "copy" typical SOMs or eval baords
<swetland>
yeah, there's a number of SOMs out there, though they tend to be rather pricey
catplant has quit [Read error: Connection reset by peer]
<mumptai>
this is not goona be cheap either
<TD-Linux>
zynq has a native rgmii block iirc
<mumptai>
very likely more expensive for small batches
<mumptai>
the only thing in favor is the fact that the rather expensive board-to-board connectors are eleminated
<swetland>
ouch. smallest zynq in qty1 from digikey is $46
<mumptai>
TD-Linux, correct
<mumptai>
TD-Linux, two actually
<swetland>
yeah there's some interesting choices with zynq as far as hooking stuff up to the SoC complex pins or FPGA IOs
<daveshah>
swetland: no significant discount for qty1000 from Arrow UK either
<mumptai>
cheapest zynq board ist also around 100€ and the SOMs are 180€
<mumptai>
and more
catplant has joined ##openfpga
<swetland>
I dunno if I've ever seen price breaks for xilinx or altera fpgas from any distributor
<swetland>
if you're not big enough to buy direct, you suffer, appears to be the state of things
<daveshah>
This was a custom quote
<adamgreig>
from what i hear the distributor prices are all very inflated though
<adamgreig>
buying direct is a big win
<daveshah>
Distributor stock of FPGAs is often awful too
<daveshah>
Quite common to only see 10s in stock
<mumptai>
but whats the threshold?
<swetland>
often if you're a hobbyist or small shop, getting time of day from direct sales is rough going
<daveshah>
I guess 10k-1M depending on the part
<daveshah>
Something like ice40 you'd have to buy near enough 1M to get attention from Lattice
<swetland>
I've only ever dealt with things as a hobbyist or as a Google employee. widely differing experiences, obviously ^^
<mumptai>
yeah name dropping helps
<swetland>
for some of the larger devices, 50-100k is "quantity" (which surprised me, I was expecting 1-5M)
<mumptai>
but this open thing won't do the trick
<cr1901_modern>
I was able to buy direct from a fabless company for a period of time after a really botched attempt to buy from a distributor. Don't think I can take advantage of it today tho.
catplant has quit [Ping timeout: 246 seconds]
<cr1901_modern>
it was like "we're sorry you had such a bad time, email us next time and we'll get you the parts you need"
jevinskie has joined ##openfpga
<daveshah>
Kynix have some good prices on Zynqs. No idea how legit they are, but I bought about 3 Artix-7s (wanted a package/grade that was OOS everywhere) from them once and it was fine
<mumptai>
yes, distributors can be annoying from time to time
catplant has joined ##openfpga
<mumptai>
XC7Z010-1CLG400I 1+: $23.62976
<swetland>
that strikes me as a pretty non-terrible price for single digit units
<daveshah>
Probably production surplus
<daveshah>
I'd be fine with Kynix for a hobby project
<daveshah>
As I say my one experience with them was fine
<daveshah>
But I wouldn't use them for a space shuttle or anything
<mumptai>
same part is $72,17 at digikey
<mumptai>
maybe student cubeSat?
<daveshah>
Rumoured volume pricing for Zynqs is around $10-15
<daveshah>
But I suspect that would be 100ks direct from Xilinx before you got that
Bob_Dole has joined ##openfpga
<mumptai>
the prices also vary at kynix, might be surplus or some production left-overs
<mumptai>
the same part in "c" temperature range is 8 usd more expensive
<mumptai>
i'm not sure if there is a valid reason for not using a SOM, besides the need for special formfactors. especially for hobby stuff and open-hardware
<mumptai>
any small batch run is easily as expensive as a SOM based solution
<mumptai>
I kinda don't care about that fact. But I assume that this is probably the reason why there is no such open-hardware design available
<mumptai>
also dealing with schematics to linux, and everything in between, might be unappealing to many
catplant has quit [Quit: ttyl]
rohitksingh has quit [Remote host closed the connection]
<mumptai>
Anyone with recommendations about routing ddr3?
<mumptai>
or just a case of follow the mfg's recommendations?
<mumptai>
so basically: Anyone with meta-recommendations about routing ddr3?
<daveshah>
I've always just followed random appnotes
<daveshah>
The only time ddr3 hasn't worked for me is when I speced a 240k Zq resistor on the BOM instead of 240r
<daveshah>
Luckily that one was easy enough to fix, after a day of debugging though
<mumptai>
i guess that is the nature of the game
<daveshah>
I'd recommend a 6 layer board, makes things easier and not too big a price gap these days
jevinski_ has joined ##openfpga
<mumptai>
that would have been my 1st choice
marcan_ has joined ##openfpga
edmoore_ has joined ##openfpga
ovf has quit [Disconnected by services]
<mumptai>
and a layerstack like: signal, power plane, signal, signal, power plane, signal ?
<daveshah>
Yeah
ovf has joined ##openfpga
<mumptai>
i actually can't think of a diffrent one that appears sensible
marcan has quit [Ping timeout: 250 seconds]
edmoore has quit [Ping timeout: 250 seconds]
jevinskie has quit [Ping timeout: 250 seconds]
lrvick has quit [Ping timeout: 250 seconds]
miek has quit [Ping timeout: 250 seconds]
edmoore_ is now known as edmoore
marcan_ is now known as marcan
lrvick1 has joined ##openfpga
miek has joined ##openfpga
<sorear>
huh. First I've seen the "240r" notation
<daveshah>
Quite common on schematics etc
<daveshah>
If you cba to find the ohm symbol or type ohm
<mumptai>
some also replace the dot with it: 2.2 Ohm is written as 2R2
Flea86 has joined ##openfpga
<daveshah>
Yeah, that's from the days of printed schematics where a dot could be hard to see
<daveshah>
or confused for dirt, or lost in photocopying etc
<mumptai>
and copier either added or removed them
Zorix has quit [Ping timeout: 268 seconds]
<sorear>
these days, copiers just replace entire digits at random /s
<qu1j0t3>
you'll also see 2K2, 1M5 etc
azonenberg_work has quit [Ping timeout: 250 seconds]