Arduino MKR Vidor 4000 – Arduino, Cyclone 10 FPGA, MiniPCI Express, MicroHDMI | Hacker News Reader

Arduino MKR Vidor 4000 – Arduino, Cyclone 10 FPGA, MiniPCI Express, MicroHDMI

arduino.cc

The $75 price point is okay.

Not sure what this is trying to be other than an eval board for 'also rans' in the FPGA and Cortex-M space. One of the reasons I like the Lattice FPGA boards like the Icebreaker[1] is that there is a complete open source tool chain for them. Seems like adding a processor to something like that might be an interesting place.

[1] https://www.crowdsupply.com/1bitsquared/icebreaker-fpga

If you are alright with a softcore, I think a picorv32 will fit on the ice40.

Cyclone outclasses any ICE40 out there. If you need the logic capacity or peripherals your options become constrained. The sad thing is that we had cheap Spartan-3's and Spartan-6's with impressive resources but the vendors didn't want to keep making devices with that sort of bang for the buck.

The open tools also work on ECP5, which goes quite a bit bigger than the Cyclone 10 chip on the Arduino and is also pretty cost effective. In fact it's big enough that you can usually dispense with the separate microcontroller - I built a ulx3s which can boot Linux on the ECP5-85.

You can still buy Spartan 3 and 6 devices - last I checked they're even still being manufactured.

The tooling is not supported with bug fixes.

Another strike for Xilinx tooling then. Open source tooling can't come soon enough.

Same open source discussion again. I’ll try to explain short why it will never happen. Since FPGAs have applications outside hobbyist domain, things are getting complicated. There are lawyers involved, support contracts signed and vendors are liable for their silicon. Now imagine some open source toolchain is used and some problems occur. Who’s liable? No single vendor will invest a penny into something maintained by somebody else. No single manager on this world will allow me to use 3rd party tools without vendor’s support for €1M FPGA design. There is no business case for silicon vendor to open source anything. Maybe brave hackers will reverse engineer bitstream generation, but this knowledge will be used only among hackers and their friends. Personally I don’t see any advantage using these tools and writing this in my resume, because nobody will use them in industry anyway.

> Maybe brave hackers will reverse engineer bitstream generation

As far as I know, the bitstream for 15K/35K/50K Artix-7 is pretty much done, with Kintex-7 and Zynq devices being WIP [1]. I think there are gaps in support for the various hard blocks and PAR at higher layers of the toolchain, but the bitstream reverse engineering per se isn't a maybe-in-the-future thing.

[1] https://github.com/SymbiFlow/prjxray

You could surely have said the same about commercial software libraries 10 years ago.

> There is no business case for silicon vendor to open source anything

There is no good business case for the smaller vendors to develop and maintain a full set of very complex software either.

Xilinx gives away access to a multi-language license for Vivado, fully capable for all chips under hundreds of dollars, with a built in simulator (again, mixed language!) used by tens of thousands of actual hardware engineers around the world successfully, but they're the bad guys?

And their "unsupported" tool is also used successfully today by rafts of actual hardware engineers, and it's still better than the free alternatives by leaps and bounds.

There are no plucky garage shops or "wise crowds" making FPGA tools that are even in the same galaxy as ISE or Vivado.

> There are no plucky garage shops or "wise crowds" making FPGA tools that are even in the same galaxy as ISE or Vivado.

Symbiotic EDA comes to mind. However, I would argue they don't need to be great, just good.

I don't have a copy of Vivado to hand (I have an Altera dev board instead), but I can take PicoSoC with Yosys/nextpnr and have a bitstream in about a minute. If I try the same task in Quartus, that takes about five minutes. Producing a (relatively) inefficient bitstream actually speeds up the compile/run cycle in that scenario.

If the design meets timing, which is often decided by external factors, any extra time spend trying to make the design "better" is wasted time.

That thing came out over a year ago, and the page hasn't been updated in nearly as long. From what I've understood you don't get direct access to the FPGA, rather they've made a few basic peripherals like SPI, UART etc that you can enable.

So it is more like a configurable microcontroller, rather than a fully programmable FPGA, at least that's my understanding.

Which I guess makes sense, given that FPGA programming isn't exactly n00b friendly from what I've gathered.

> Which I guess makes sense, given that FPGA programming isn't exactly n00b friendly from what I've gathered.

It can be OK. The first class most EE students take at my undergrad uni had a significant FPGA component.

You can program the FPGA, just seems the docs aren't helpful.

https://forum.arduino.cc/index.php?topic=602341.0 (response #5 is useful)

Ah nice. I looked at this board a couple of months ago, and all indications I could find then was that the FPGA was not directly programmable. Good to see that's not the case.

Check this out also. http://smartaleks.io/?p=260

Seeing as there's a SystemVerilog tutorial on the Arduino website, I'm hopeful for more.

Here you go. I have followed this guide to program the fpga from the Intel Quartus IDE http://smartaleks.io/?p=260

One would hope it would be a bit more, at least given the price point.

Shameless self-plug for something similar in the market: WebFPGA (https://webfpga.io/kickstarter). This service actually exposes the full FPGA and allows you to synthesize Verilog remotely (or offline).

Love your project!

Now it has its own article on HN:

https://news.ycombinator.com/item?id=21057511

Remember to upvote!

(Oh, and "You're welcome!" :-) <g>)

Great! Thank you! I definitely see a lot of room for improvement in this field.

Commented and upvoted.

Question for experts in the field: are there any reasonable FPGA options capable of hosting a small (~2MiB) quantized deep learning model for which there's any sort of a viable, open toolchain?

I've looked into some of Lattice's stuff, and some of their chips seem pretty capable for this, but as with all the other FPGA manufacturers, the toolchain is crazy proprietary. I'm not a huge fan of sinking my time into tooling I have no control over.

Yes! Most Cyclone Vs or Zync/Artix chips can support quantized networks, and people have been making them easier to get started with, for example this toolchain[1]: https://github.com/danielholanda/LeFlow

Can take Tensorflow models and convert them to Verilog. Tensorflow already has lots of tools to train quantized models, so all you need to do is convert them to hardware.

If you had a big enough Lattice FPGA you the open source Yosys toolchain, or use one of the Vendor proprietary toolchains for Altera/Xilinx.

[1] Disclaimer - I went to school with the creator of LeFlow.

There is an open source syntheses and bitstream generation for some lattice devices.

It has inferior QoR compared to the closed toolchain.

I think in one of the CCCamp19 talks, the ECP5 routing with the OpenSource tooling blew the Lattice tooling out of the water.

FPGA manufacturers have pretty perverse incentives when it comes to filling your FPGA devices up with useless junk.

Link to talk? My Google Fu fails me.

The Lattice ECP5 is supported by open source tooling, does it support your usecase otherwise?

Is the VHDL or Verilog code for MicroHDMI and CSI-2 interfaces available somewhere?

Not sure if this board will be wild success. Diligent has better boards with Xilinx Artix-7 for this price. Arrow MAX1000 is even more affordable. Intel MAX10 is a great device! Learning FPGA means always learning quirks of vendor’s IDE.

Title is slightly misleading, it isss the mini-pci express connector but the FPGA itself can't talk pci-express as it lacks the necessary transceivers

Link to the actual board: https://store.arduino.cc/mkr-vidor-4000

€62.90

Can someone try Clifford Wolf's tiny risc-v on this fpga?

I'm mainly interested in size and speed for this specific chip and speed grade.

Reply on news.ycombinator.com