If you tie yourself to one particular board that you don't have control over, you'll end up very sad. I think it would be better to find several different platforms and try to develop a graphics solution that's somewhat generic. Each is likely to need a different memory controller, and each may need a different PCIe controller. But if we provide a layer that gives us a uniform memory and host interface, then we can make a video rasterizer and maybe even some rendering logic that works on all of them.
On Wed, Oct 31, 2012 at 1:50 PM, Patrick McNamara <[email protected]> wrote: > No argument that it's expensive and there is probably a better card out > there somewhere. This was just the first one I came across that met the > basic needs. As far as what kind of FPGA is needed, that would depend > entirely one what you were trying to do. Assuming you have a hw block for > PCIe, you can probably do a basic framebuffer in most any FPGA. If you want > to go further, my gut feeling is that you get into chips that aren't > supported in the free Xilinx tools pretty quick. > > Patrick M > > > > ________________________________ > From: ""Ing. Daniel Rozsnyó"" <[email protected]> > To: Patrick McNamara <[email protected]> > Cc: "[email protected]" <[email protected]> > Sent: Wednesday, October 31, 2012 9:56 AM > > Subject: Re: [Open-graphics] The next OGA card? > > That is quite an expensive piece of hardware given that: > - it has an old generation device > - it has nonstandard extension plugs > > The nice thing is the support for DIMMs, however is that a soft-core or hw > core, like the one in Spartan6? (there are up to 4x 12.8 Gb/s DDR3 > interfaces). > There are far more accessible boards which already got some digital video > output and have support in the free version of ISE (The SP605). > > Is there really a need for Virtex class device? > > /* From my viewpoint, the slowness of OG development was also caused by the > fact, that the boards were not for everyone. If there would be hardware, > maybe smaller, but more exposed, the individuals in the community could > "play" more - resulting in a specialized cores / modules of smaller scale, > which can be embed into a bigger and more complex card, by the ones, who > have access to ISE licenses and expensive hardware. */ > > Another playground will be a Zynq device - there is an FPGA and a CPU, but > no graphics card - development kits will be available soon. > > Daniel > > > On 10/31/2012 03:37 PM, Patrick McNamara wrote: > > In the commercial board space, something like this would be a good fit: > http://twinind.com/catalog_detail.php?id=346 > > Up to 2GB of DDR2 and PCIe native support in the Virtex. Good off board > connections. If you are looking for a basic framebuffer, the all you need > is a daughter card with DVI/HDMI/VGA/etc components. If you have to do a 3D > engine, the add an FPGA for the engine to the daughter card. Let the > Virtex5 do the memory interface/framebuffer and let the daughter board FPGA > do the computational engine. > > The problem, of course, is that it isn't going to be cheap by hobbyist > standards. But then people need to realize that you are getting into the > realm of FPGAs that cost hundreds of dollars just for the chip itself in > single and low number quantities. > > ________________________________ > From: Timothy Normand Miller <[email protected]> > To: Ing. Daniel Rozsnyó <[email protected]> > Cc: [email protected] > Sent: Wednesday, October 31, 2012 8:48 AM > Subject: Re: [Open-graphics] The next OGA card? > > OGD1 is open source, and the mods to add a PCIe to PCI bridge for it > wouldn't be too tough (for someone who is good at PCB design). If you > can do the board, the FPGA logic is there. OGD1 already works and can > be plugged into a PC (with a PCI slot) and booted as console. > > The biggest challenge is the expense of having it produced. 25 OGD1 > boards cost somewhere around $13000 to make. > > Or we could start from scratch. Andre Pouliot and I had some > (slightly divergent) ideas for an OGD2, which would be based on a > newer FPGA and be highly modular. > > Another idea would be to see if there is _already_ an FPGA board out > there with a PCIe on it. Actually, I'm sure there is. Xilinx and > others make project boards for their FPGAs. I don't know if you'll > find one with video on it, but if you did, we could do a graphics core > pretty easily. If not, we might be able to design a DVI daughter > board. > > Another weird idea (particularly because of the signal integrity > challenges) would be to see if one of these project boards brings > SERDES outputs to some pins with reasonably good quality; we could > then design our own TMDS/DVI circuit to go in the FPGA. :) If we > could pull that off, it would be by far the least expensive OGD2 > option, because it would require almost zero hardware design on our > part. We'd need to do FPGA logic, drivers, and adapt I/Os to a DVI > connector. Analog video might be out of the question. > > So, here are the features we need in an off-the-shelf FPGA project board: > > - PCIe connector > - Memory > - DVI and/or VGA/DAC is bonus > > Can you find something like that? > > > Actually, if we have lots of I/Os to spare, we might be able to do > analog video using a resistor network and a low-pass filter. Someone > with a better EE background could answer that. > > > On Fri, Oct 26, 2012 at 5:24 AM, "Ing. Daniel Rozsnyó" > <[email protected]> wrote: >> Hello all, >> currently I am having quite a trouble to get my system working - the >> board >> is Intel DN2800MT, with an n2800 dualcore Atom (Cedar View) - which got >> that >> crappy PowerVR SGX 545 graphics. >> >> Currently, there is no driver for it for 64bit windows and the driver for >> Linux is a binary only thing which links to an old xorg and old kernel >> (3.0 >> or 3.1). >> >> There is same problem with the older generation of Atoms (Poulsbo) as >> they >> have also this 3rd party graphics core and comes with no support. >> >> I was thinking - would it be possible to put together a simple graphics >> card based on Opengraphics, with the following parameters: >> >> - PCIe 1x >> - single DP/HDMI/LVDS output >> - form factor to match a full mini pcie card (50mm) >> >> If I picked up correctly, there is a drive for embedded targets.. so this >> could be interesting. >> >> From the hardware point, it should be enough to use a single FPGA with a >> single (or dual) ddr3 memories (that makes 12.8 or 25.6 gbit/s) which is >> quite enough for filling it over pcie (2.5Gb) and reading to DP. >> >> For the start, a simple 2D framebuffer with region copy acceleration >> would >> be sufficient.. >> >> What does the masses say to that? >> >> >> I can manage to make the hardware, is there somebody who can make the >> fpga >> core and somebody to write drivers for it? >> >> Daniel >> >> >> >> _______________________________________________ >> 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 > http://www.cse.ohio-state.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) > > > > > > > > _______________________________________________ > 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 http://www.cse.ohio-state.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)
