The last time I used the Linux version of the Xilinx tools, they were pretty awful. If they've improved, then yeah, I can avoid Windows. One concern is that the board-locked license for the tools that comes with this board will be WIndows-only, and it'll take 6 months to get the Linux version. We'll have to make sure. I can also pop a big AMD GPU in that box for some OpenCL stuff. As for the FPGA size, we can play around with various interconnect options, although the FPGA maybe won't model them so well. Fair point about the smaller FPGA--perhaps we should also get a smaller one too.
On Sun, Apr 28, 2013 at 5:30 PM, hozer <[email protected]> wrote: > I cannot think of any reason to run windows for synthesis these days. > > Xilinx works quite well on amd64-linux, so why not just invest in a nice > case and an AMD FM2/FM3 (cpu + GPU) motherboard, and something like 32GB > of RAM? > > I'll also question the need for a 'big' FPGA. I've *almost* managed to fit > the Milkymist-NG into an Spartan6-LX25, and this has a cpu + video > processing > (the regular milkymist platform has an LX45 IIRC) > > I think you could do a LOT with just putting 4-16 shaders on this: > http://www.xess.com/prods/prod055.php > > The benefit of 'small' FPGAs is you will make things accessible to a HUGE > number of hobbyists, open-source hardware hackers, and students. > > I would think the results from 4, 8, 16 shaders would scale in a pretty > straightforward way to thousands of shaders without having to spend > megabucks > on a big FPGA board that only one person can really use at once. > > If you want big, I think you can scale much faster and farther with LX75T > FPGAs, the https://bitbucket.org/dahozer/infiniband-fpga code, and an > Infiniband switch. (but that requires some infrastructure work to finish > that code and make it work on current MGT implementations) > > On Thu, Apr 25, 2013 at 07:48:24PM -0400, Timothy Normand Miller wrote: > > Possibly in the summer, I'm going to invest in some really big FPGA to > help > > with this GPU work. Give or take, I'm looking to spend about USD 10000 > to > > 15000 for the FPGA board and an "adequate" Linux PC to plug it into as a > > host (assuming the FPGA board has PCIe). I might need to also splurge > for > > a Windows PC to run the FPGA synthesis tools. > > > > Note: I'd benefit from a reasonably fast Windows PC for synthesis, but > > otherwise, I don't need to spend a lot on fast CPUs, which is why the > Linux > > host can be fairly low-end. I have access to more than enough fast > compute > > power from existing facilities. > > > > Any suggestions? > > > > BTW, if we're comparing Virtex to Spartan, I need lots of SRAMs and > > multipliers, but I don't yet know the relative proportion of random > logic. > > IIRC, Virtex has more random logic, while Spartan has more RAMs and DSP > > blocks. So if we can narrow down the choices a bit, then I can pick the > > right one after I have enough logic to synthesize something. > > > > Right now, I have tentatively 5 masters students wanting to work on > > OpenShader over the summer. > > > > > > -- > > Timothy Normand Miller, PhD > > Assistant Professor of Computer Science, Binghamton University > > http://www.cs.binghamton.edu/~millerti/ > > Open Graphics Project > > > _______________________________________________ > > Open-graphics mailing list > > [email protected] > > http://lists.duskglow.com/mailman/listinfo/open-graphics > > List service provided by Duskglow Consulting, LLC (www.duskglow.com) > > -- Timothy Normand Miller, PhD Assistant Professor of Computer Science, Binghamton University http://www.cs.binghamton.edu/~millerti/ Open Graphics Project
_______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
