<ysionneau>
has anyone used Achronix FPGAs? They say their fabric can run @ 1.5 GHz :o
<ysionneau>
oh RLDRAM3, I didn't know about that, SRAM like random access, with low latency and DDR3-like bandwidth
<cr1901_modern>
It would've been nice if SRAM densities were so severely limited... last I checked, it takes nearly $300 to get 16MB of SRAM. That's like early 90's DRAM prices o.0;
<cr1901_modern>
(to be fair, that price includes cost of labor to make boards as well- the raw cost is probably half of that)
<ysionneau>
it seems you can have 1.15 Gb of LRDRAM3 from Micron, but dunno about the price
<ysionneau>
RLDRAM3*
<cr1901_modern>
weren't so* limited XD
<cr1901_modern>
Ugh, BGA... out of the scope of my soldering skills
cr1901_modern has quit [Ping timeout: 244 seconds]
fengling has joined #m-labs
cr1901_modern has joined #m-labs
<sb0>
cr1901_modern, yeah, low volume = you pay through the nose
<cr1901_modern>
I blame the Pi for making me think otherwise, tbh
<sb0>
hahaha, totally true
<sb0>
also the pi isn't low-volume, it's a smartphone soc
<cr1901_modern>
i.e. it's possible to have all this memory and I/O for cheap. Sure, if you already have the capital and business deals with the company.
<ysionneau>
=)
* cr1901_modern
is getting sick of his modem going out to lunch
<cr1901_modern>
That article hits home a bit. I like older boards/electronics precisely because they're more open and easier to service.
<sb0>
it gets really cynical when people call the pi some great innovation platform. not only is it a blackbox, but also this sort of dirt-cheap device steals the oxygen of any computer project that doesn't ride on smartphone socs or similar...
<cr1901_modern>
I used to think that, I don't think I agree anymore. It shouldn't take 700MHz to run a Unix either
<cr1901_modern>
I can run it just fine at 50MHz- probably lower if I had the time to set up a board.
<sb0>
bah. cpu resource usage waste has its advantage: it's easier to write straightforward and easy to maintain (but inefficient) programs using high level/interpreted languages and tons of libs
<cr1901_modern>
Well yes, I'm not saying everyone should go back to assembly :P
<cr1901_modern>
I'm probably just bitter b/c I'm not qualified to design 100MHz or above boards lol
<cr1901_modern>
Hmmm, the person who added the minispartan platform used LVCMOS33 for all the IOpins, but the UCF says LVTTL. Is there any difference?
<cr1901_modern>
(Well, functional difference)
antgreen has quit [Ping timeout: 256 seconds]
<sb0>
yes, and it won't get better even if you were. GPUs do 7Gbps/pin single-ended with GDDR memory, whereas SERDES in modern FPGAs are 1Gbps or so
<sb0>
(the transceivers are much faster, but they are differential, few, and cannot be used with SDRAM)
<sb0>
and expensive, too.
<cr1901_modern>
Great for companies with a lot of existing capital, not so much for hobbyists :/
<ysionneau>
Achronix says they have 28 Gbps SerDes, don't know if that's marketing BS or not
<sb0>
afaik achronix fpgas are very expensive, and difficult to source
<ysionneau>
ah ok :/
<sb0>
plus I'm willing to bet their design software is 10 times worse than ISE
<cr1901_modern>
sb0: When adding a new platform to migen (as I'm currently doing), I am simply just reimplementing the UCF, correct?
<sb0>
basically yes
<cr1901_modern>
i.e. as long as the pins are correct, I don't need to use the same names as the "official" UCF
<sb0>
no
<ysionneau>
they are using Synopsys
<sb0>
synopsys doesn't do the place-and-route
<sb0>
and there's a technology-dependent part that needs to be written specifically for synopsys. if they contracted synopsys for developing it, I guess it should be halfway decent, but who knows
<ysionneau>
yes they use their own P&R tool
<cr1901_modern>
Actually, it seems that I have to use a preset set of names? I remember getting an error during a test run of MiSoC (MiniSoc subtarget for Papilio Pro) b/c "eth_clocks" doesn't exist.
<sb0>
cr1901_modern, consistency is always good anyway
<cr1901_modern>
I understand. The board I'm adding doesn't have anything not already implemented by another board anyway.
<cr1901_modern>
I won't actually be able to test for another month, btw. I just suspect that my real life will become busy between then and now, and am doing the work I can ahead of time.