#milkymist IRC log for Tuesday, 2013-06-04

lekernelhi12:13
lekernelwhat do you think about a name change to "milectra"?12:13
lekernelsounds more generic than "milkymist" - milkdrop stuff is hardly what works best here. and less strange, too.12:14
juliusblekernel: for the entire project?15:14
lekernelyes15:16
juliusbsounds like the name of a company which installs electric fences around military installations :P15:17
FallenouI kind of like milkymist :p15:19
Action: lekernel trying tabula. 9GB download, flexlm DRM shitware, installer taking 33% CPU to download files from the internet, UI toolkit from 1991, installer crashing when mirror is unavailable. so far, sounds familiar.15:19
Fallenouahah15:20
juliusbI don't know what tabula is but I'll assume it's EDA from the description of the process you're going through15:20
lekernela new FPGA company that claims to achieve halfway decent (2GHz counters) performance15:21
lekerneltabula.com15:21
wpwrakmilkymist as a project has more or less ended already, so why rename it now ?15:21
juliusbohh yes, these guys15:21
wpwrakif you don't like the name anymore, give new things a new name ..15:21
Fallenou2 GHz, sounds attractive :)15:24
juliusbholy hell, that ABAX A1EC04 thing has nearly 400k LUTs, and 5.5MB of block ram?!?! How much does it cost?15:24
juliusbahhh, yes, $7,500, not quite enthusiast gear15:24
lekernelthey have smaller and cheaper ones it seems15:25
lekernelwpwrak, M1 is dormant, but the project (as a OSHW community) has not ended15:25
juliusbthe virtex7 2000T parts have about that much block ram but only 1/3 of the LUTs, but are an order of magnitude more expensive in development kit form15:26
lekernelthe 2000T has nearly 2 million LUTs15:27
lekernelno?15:27
lekernelbut half the speed of the tabula stuff15:28
juliusb305k slices and each slice has 4LUTs and 8FF15:28
juliusbit has nearly 2 million "logic cells"15:28
lekernelmh15:28
lekernelnew confusing terminology from xilinx marketing ...15:28
juliusboh they always do this15:28
juliusbbut yeah probably those tabula things are faster, but, how do the same designs fare on both? Tablua are new, and it's unlikely their backend tool is as well tested as the Xilinx stuff15:29
wpwraklekernel: i mean that you probably won't make another vj station. that's what's associated with the name "milkymist". of course, there's a diverse community around it, but if it's for things like mailing lists and such, just creating a new places would be no more intrusive than a rename15:30
wpwrakheh, see ? "new confusing terminology" :)15:30
lekerneljuliusb, given how buggy the xilinx stuff is, I can imagine that a skilled team of developers can do better15:31
lekernelbut I must say that so far I'm not impressed with the tabula software15:31
juliusbwhat are you trying to put through it?15:32
lekernelstill at the bloated install process now... then I have to wait another 24 hours for the license approval... meh15:32
lekernelbut there are already familiar red flags in the installer: flexlm, 9GB download, outdated UI toolkit, etc.15:33
lekerneland even the fact that there is an installer at all...15:34
juliusbah these FPGA guys love this shit15:39
wpwrakthe more shackles the better15:40
Action: Fallenou dreams of a 500 MHz lm32 core15:44
juliusbbuild an ASIC15:48
juliusband why are you dreaming of that lame core, mor1kx every day of the week man ;)15:49
lekernelhmm, last time I checked that 'lame core' was still several times the speed and several times less area than mor1kx - and it has a MMU15:50
juliusbmor1kx has MMU now too (stekern's great work)15:51
lekernelidk why everyone in the open FPGA community wants to have their own processor, this results in a huge graveyard of unfinished, buggy and unusable projects15:51
juliusbor1200 is not unfinished, buggy or unusuable, mor1kx is very much actively under development15:51
juliusbor1200 is far from optimal though, and is the point of the mor1kx15:51
lekernelM1 could not run on OR1200 due to low performance. that qualifies as 'unusable' for me.15:52
juliusbbut you're right, everyone does want their own ;) because we think we can do it better15:52
juliusblekernel: you're probably just using it wrong, it runs OK with caches15:52
lekernelthere were even latches inferred in the HDL when I tried it - I will admit that you have fixed that now, but still it was buggy15:52
juliusbanyway, that's enough stirring from me :)15:52
lekernelwell in the end there's only one argument you can use15:53
juliusbfor what?15:53
lekernelmibench (or equivalent) results15:53
lekernelaccompanied with a LUT count :-)15:53
juliusbwell, I think for FPGA stuff, synthesisable speed and coremark result (I presume mibench is similar?)15:54
juliusbFPGAs are big enough now not to worry too much about LUT count15:54
lekernelnot really, I'm toying with the idea of a entry-level video mixer based on 6slx9 or even 6slx4 atm15:54
juliusbI have heard the numbers of the lm32, it is impressive and I'm not sure why - could be architectural - or1k has some bad things in there15:55
juliusbwell you're a masochist then15:55
juliusbon the other hand, you will have quick compile times :)15:56
lekernelwhy? the current milkymist-ng design, including ethernet which I will remove, fits effortlessly in lx915:56
lekernelyes, that too15:56
juliusbseriously? which ethernet are yuo using? the Xilinx stuff or do you have your own?15:56
lekernelown. I'm not using any Xilinx stuff.15:56
juliusbohh i need to look at that15:56
juliusbthe OpenCores one is massive and i tried to fix it but failed a long time ago15:57
juliusba better one would be nice15:57
lekernelyes, I tried it15:57
lekernelwell mine is very minimal15:57
juliusbthat's fine15:57
juliusbi gotta, go, ttyl15:57
lekernelno DMA, just block RAM, and none of the stupid/legacy ethernet features implemented15:57
lekernelworks fine at slow speeds (due to CPU copying lots of memories, doing the CRCs, etc.) though15:58
Fallenoufarnell and digikey don't sell tabula devices :(17:39
--- Wed Jun 5 201300:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!