You don't want random nut to fly to Mars just to go bat shit insane and wreck perfectly good experiment.

Why not? The movie rights alone would pay for the mission.

Which is clearly not the case based on experiences with the mars rovers and similar devices.

Ridiculous. Think of one of the most interesting discoveries made by the Phoenix lander -- the frozen condensate that formed on one of the landing struts. A human would have noticed that immediately and been able to analyze it in detail. Conversely, a robotic probe can do only what it's programmed to do. All we can do is stroke our beards and say "Hmm, wonder what that is?"

When you're not only expecting the unexpected, but hoping for it, you want human boots on the ground. One human mission is easily worth twenty robotic missions.

Hell, NASA should consider offering one-way trips. They'd have enough volunteers to crash their Web server. Most people aren't doing anything that important or interesting with the rest of their lives, are they? Send one old guy with a shovel, a microscope, and a carbon-monoxide canister, and we'll learn more than we would from the next hundred years' worth of robots.

Of course, the problem is that I don't know of anyone who can comfortably solder BGA chips.

People have been doing it with toaster ovens (http://www.instructables.com/id/Toaster-Oven-Reflow-Soldering-BGA/ and http://www.sparkfun.com/commerce/tutorial_info.php?tutorials_id=60&page=). I haven't tried that yet. I have soldered QFNs by laying down some solder paste and blasting the other side of the board with a heat gun, but it's a high-pucker-factor operation. I plan to try the toaster oven hack the next time I need to do that.

True, but I was not able to get the FX2's pipe to operate reliably on every edge of IFCLK. This might have to do with the fact that my data transfer FPGA logic is clocked off IFCLK as well, but might be a limitation of the FX2 or board design.

Clocking the FPGA from the 48 MHz IFCLK is the right thing to do IMHO. That's how my setup's working. Email me offline if you want a code dump; no guarantees that it's optimal or even "correct", but I've run it for several days at a time without any problems.

You aren't trying to use asynchronous writes, are you? Those can't be pushed beyond about 6 MB/sec from what I could tell.

Also, what USB driver are you using? CyUSB? If so, make sure you use overlapped reads or you'll miss tons of data.

I agree that Gigabit (or 10G) is the way to go, but do you know of any convenient boards (like Nexys) for that ? Thank you !

Unfortunately, no... it seems there are numerous boards with 10/100 Ethernet interfaces but I haven't seen any GigE or better, except for the expensive Virtex-based ones like the NetFPGA board. Digilent had something called the FX12 for awhile that had a Webpack-supported Virtex chip and a GigE port, but it's been discontinued.

I really want to see some level of text based gaming come back.

Text gaming didn't leave, it just went indie. Some of the best works since the Infocom days have appeared in the annual rec.arts.int-fiction competition, the 15th of which is in progress now.

Someone below mentioned Photopia, and that's a good place to start (it took first place in the IF competition nine or ten years ago).

I don't think the 8-bit path between the CY68013 and FPGA is a bottleneck, really. With an IFCLK of 48 MHz, you'd have to be able to shove 48 MBytes/sec through the endpoint to be bus-width limited, and that just isn't going to happen over USB 2.0. You'll have to spin on FIFO-full either way.

In practice about 32 MBytes/sec is about the best you can count on achieving, although I understand that the GNU Radio guys are pushing 40 MB/sec with some 8051 bus-mastering hacks.

Seems that most of the speed-constrained designs are moving to Gigabit Ethernet these days, avoiding the problem entirely.

Opal Kelly modules are clean, well-documented designs, but because of their relatively high cost they're better thought of as embeddable modules rather than as learning platforms. For education and basic home tinkering, you want a cheap well-documented board with lots of peripherals to play with.

I think the best overall learning platform right now is the Nexys2. You can teach yourself everything from USB interfacing to VHDL/Verilog design for $129, and it's got a lot of switches, lights, ports, and outboard RAM to play with.

There are also a couple of reasonably priced books written especially for novices learning logic design with the Digilent platforms. $200 will get you both the basic- and advanced-level books from LBE and a Nexys2 to do the exercises. Very hard to beat that combination IMHO. (You can use the Nexys2 with the first book, even though it was written for the lower-end Basys board.)

For a newbie, CtoVerilog.com is the mother of all bad ideas. Verilog is not isomorphic to C. Being able to represent a few trivial C loops as Verilog code will not help you learn what's going on under the hood, and it will not help you get the fundamental aspects of your design up and running.

Also, much of what's done in HDL is gluing other chips together. If you try to use CtoVerilog to make a high-speed USB chip talk to an ADC, the results will be amusing at best. If you find yourself wanting to use C in an HDL design, consider either falling back to a conventional microcontroller or using a software core to run it.