5 ms·

Lattice appears to be OK with this and that's great - they're the only FPGA company at this point that seems to understand that open source could actually help

Lattice appears to be OK with this and that's great - they're the only FPGA company at this point that seems to understand that open source could actually help them with tooling and adoption. Xilinx and Altera are hostile towards open source so I'm rooting for Lattice.

DoingIsLearning 6y ago
The best business move they could make against the competition giants, is to ask people like Dave Shah and Clifford Wolfe to just organize a single team under a foundation umbrella and then make a large annual donation, plus technical support/documentation where needed. If Lattice disrupts the market as de facto enablers of an LLVM of sorts for FPGA's they will make a lot of clients think twice for their next projects. It doesn't matter how much IP secret sauce Xilinx or Altera have in their fabric, every single team out there hates the locked in toolchain and is slowed down by it in a non-negligible way. In my opinion this is the type of bold move that could flip the market share upside down in favour of Lattice.
- UncleOxidant 6y ago
  The FPGA companies need help with software because quite frankly, their software sucks - so damn buggy. And yet, here they could get some significant help for not much $$. But Xilinx and Altera prefer having customers locked in. I've long thought that the best way for the smaller players to gain marketshare is for them to actively court open source - not only to allow it, but to encourage and fund it.
- sam_bristow 6y ago
  FYI it's Claire Wolfe now.
  - DoingIsLearning 6y ago
    Thank you for the correction. The time window I was allowed to edit the comment has passed, perhaps Dang can fix the original comment but otherwise noted.
- tn1 6y ago
  > It doesn't matter how much IP secret sauce Xilinx or Altera have in their fabric, every single team out there hates the locked in toolchain and is slowed down by it in a non-negligible way. Unfortunately this isn't really a compelling reason to switch for most shops. The choice of FPGA is entirely driven by the features of the parts that the vendor offers. Salespeople talk to the relevant engineers and a part + pricing is negotiated accordingly. Once you start pushing the limits of what your part can do, your design necessarily becomes tied to the details of the chip. The secret sauce is the reason why a specific part is bought! At least Vivado (on Linux) is actually quite painless once you get the hang of it. If you're an important enough customer, you can complain to your FAE and they'll get you access to unreleased versions with a specific problem fixed.
  - DoingIsLearning 6y ago
    I have to agree with the comment from bsder, unless you are doing heavy-duty video, RADAR, or RF comms the large majority of chips are never fully utilized, these business are really quite dependent on sales pitches/push. For the large majority of clients they just need to implement a small/medium time critical task. > At least Vivado (on Linux) is actually quite painless once you get the hang of it. If you're an important enough customer, you can complain to your FAE and they'll get you access to unreleased versions with a specific problem fixed. I disagree, I haven't touched Vivado in about 1 year but I doubt it revolutionized itself in 2020.X . I am going to politely ignore the shear volume of files that get auto-generated, duplicated, cached etc. and make it near impossible to version control. I also won't enumerate the GUI bugs that are only resolved with TCL scripting. However, as of 2019, Vivado was still unable to support VHDL-2008 standard simulations, same thing for packaging IP in VHDL-2008, that is literally 11 years behind in VHDL standard support. In any other industry people would be flipping the table. The problem is they don't make money out of the software they make money selling chips, and that is the wrong incentive to improve the toolchain. I would much rather they charge for the toolchain but actually support it with adequate resources and a sane development roadmap, OR they just accept the lack of budget, capacity, profit in software and open-source it.
oytis 6y ago
But also Lattice doesn't produce FPGAs nearly as powerful as Xilinx Ultrascale. So it's not like people are free to choose whatever. Plus, as the number of gates and other elements gets larger, sophisticated software with a lot of domain knowledge becomes more important - probably that's why Xilinx and Altera are not rushing with open source support.
- UncleOxidant 6y ago
  True, but I'm about to start an FPGA contract where a Lattice FPGA would be a good fit. Doesn't need a huge amount of gates and performance would be just fine. There are a lot of designs that can fit in that design envelope.
  - bcrl 6y ago
    Lattice seems to do quite well in the small and affordable market segment. There are lots of interesting products using them at astoundingly cheap price points. https://hackaday.com/2020/01/24/new-part-day-led-driver-is-fpga-dev-board-in-disguise/ https://hackaday.com/2020/01/24/new-part-day-led-driver-is-f... is probably the most affordable FPGA board that has been fully reverse engineered with 2 x gigabit ethernet ports and SDRAM.
- bsder 6y ago
  The mega-FPGA's are irrelevant to the consumer and small business segment. Sure, the guys doing software defined radio need beefy FPGAs (but, even they're starting to use much lower class ones). However, over the past 20 years, the things I have seen small businesses use FPGAs for rarely stresses the FPGA fabric even slightly. Only a single company managed to actually fill an FPGA that we had to bump to the next class of the same family. Mostly the usages are just: 1) extra pins to control/sequence "stuff" or 2) one single latency sensitive thing or 3) ADC communication. I'm struggling to think of any company I did work for that was outside those applications--"safety lockout" would probably count. Back in the days of 8-bit micros (yeah ... I'm old), you would probably offload the PID control loops so they didn't clog your micro. Nowadays, microcontrollers are so powerful and FPGAs are so annoying to program that you're better off shunting as much as you can into the micro and minimizing the FPGA. > Plus, as the number of gates and other elements gets larger, sophisticated software with a lot of domain knowledge becomes more important I'm going to call this out. I simply don't believe it. The publications of Xilinx and Altera (when there are any) simply look like the same algorithms from 20+ years ago but with Moore's Law applied to them. FPGA tools still suck at anything that looks like a "datapath".
lnsru 6y ago
From user perspective: the tools are shitty, version control isn’t existing. But I don’t have time to self host any open source tools. Nor my colleagues do. We must write code and design electronics around FPGAs. My manager will never pay for time spend with open source products. Big Corp I work for has a dedicated team with direct contact to Xilinx support if something goes badly. Next thing is legacy designs of the Big Corp. If it was Virtex-6, the next one will be Virtex-7. If it was Spartan-6, next one will be Spartan-7. Even if Altera/Intel or Lattice or some cool startup could offer similar chips for 60% the Xilinx’s price we would stay with Xilinx because of the codebase. And don’t forget supply chain, as Big Corp buys tons of chips from single vendor the price is unbeatable. Changing vendor and starting with clean account doesn’t make sense financially. And I am sure, that this single Big Corp makes more money for Xilinx than all the hobbyist worldwide. Honestly I can’t see a reason why Xilinx should open source anything.
- UncleOxidant 6y ago
  Sure, that's the vendor lock-in that Xilinx and Altera both bank on - A sort of Stockholm syndrome. But as newer players enter the FPGA field (a lot of the basic FPGA patents have expired over the last five years or so) open source is going to be the best bet for them to both get decent tools and win some market share. It's not like Xilinx/Altera will lose any significant market share anytime soon, but I can see a slow erosion on the lower end over the next 10 years or so. It'll probably look like the FPGA pie getting bigger with newer players (like Lattice) expanding into the smaller design niche. BigCorp has apparently been doing FPGA designs for many years now, but there are a lot of smaller companies that are just starting to consider jumping into doing FPGA designs.
  - lnsru 6y ago
    I am not sure if there are any good areas for products with FPGAs for new players. I am working on one, but the single reason for that is that I am not good with Raspberry Pi. If I had cofounder with embedded Linux skills I would immediately ditch my FPGA design. 25$ price tag of Raspberry Pi compute module 4 is very attractive. I’ll end with ~100$ FPGA board making whole product more expensive.

DoingIsLearning 6y ago

The best business move they could make against the competition giants, is to ask people like Dave Shah and Clifford Wolfe to just organize a single team under a foundation umbrella and then make a large annual donation, plus technical support/documentation where needed. If Lattice disrupts the market as de facto enablers of an LLVM of sorts for FPGA's they will make a lot of clients think twice for their next projects. It doesn't matter how much IP secret sauce Xilinx or Altera have in their fabric, every single team out there hates the locked in toolchain and is slowed down by it in a non-negligible way. In my opinion this is the type of bold move that could flip the market share upside down in favour of Lattice.

UncleOxidant 6y ago
The FPGA companies need help with software because quite frankly, their software sucks - so damn buggy. And yet, here they could get some significant help for not much $$. But Xilinx and Altera prefer having customers locked in. I've long thought that the best way for the smaller players to gain marketshare is for them to actively court open source - not only to allow it, but to encourage and fund it.
sam_bristow 6y ago
FYI it's Claire Wolfe now.
- DoingIsLearning 6y ago
  Thank you for the correction. The time window I was allowed to edit the comment has passed, perhaps Dang can fix the original comment but otherwise noted.
tn1 6y ago
> It doesn't matter how much IP secret sauce Xilinx or Altera have in their fabric, every single team out there hates the locked in toolchain and is slowed down by it in a non-negligible way. Unfortunately this isn't really a compelling reason to switch for most shops. The choice of FPGA is entirely driven by the features of the parts that the vendor offers. Salespeople talk to the relevant engineers and a part + pricing is negotiated accordingly. Once you start pushing the limits of what your part can do, your design necessarily becomes tied to the details of the chip. The secret sauce is the reason why a specific part is bought! At least Vivado (on Linux) is actually quite painless once you get the hang of it. If you're an important enough customer, you can complain to your FAE and they'll get you access to unreleased versions with a specific problem fixed.
- DoingIsLearning 6y ago
  I have to agree with the comment from bsder, unless you are doing heavy-duty video, RADAR, or RF comms the large majority of chips are never fully utilized, these business are really quite dependent on sales pitches/push. For the large majority of clients they just need to implement a small/medium time critical task. > At least Vivado (on Linux) is actually quite painless once you get the hang of it. If you're an important enough customer, you can complain to your FAE and they'll get you access to unreleased versions with a specific problem fixed. I disagree, I haven't touched Vivado in about 1 year but I doubt it revolutionized itself in 2020.X . I am going to politely ignore the shear volume of files that get auto-generated, duplicated, cached etc. and make it near impossible to version control. I also won't enumerate the GUI bugs that are only resolved with TCL scripting. However, as of 2019, Vivado was still unable to support VHDL-2008 standard simulations, same thing for packaging IP in VHDL-2008, that is literally 11 years behind in VHDL standard support. In any other industry people would be flipping the table. The problem is they don't make money out of the software they make money selling chips, and that is the wrong incentive to improve the toolchain. I would much rather they charge for the toolchain but actually support it with adequate resources and a sane development roadmap, OR they just accept the lack of budget, capacity, profit in software and open-source it.

oytis 6y ago

But also Lattice doesn't produce FPGAs nearly as powerful as Xilinx Ultrascale. So it's not like people are free to choose whatever. Plus, as the number of gates and other elements gets larger, sophisticated software with a lot of domain knowledge becomes more important - probably that's why Xilinx and Altera are not rushing with open source support.

UncleOxidant 6y ago
True, but I'm about to start an FPGA contract where a Lattice FPGA would be a good fit. Doesn't need a huge amount of gates and performance would be just fine. There are a lot of designs that can fit in that design envelope.
- bcrl 6y ago
  Lattice seems to do quite well in the small and affordable market segment. There are lots of interesting products using them at astoundingly cheap price points. https://hackaday.com/2020/01/24/new-part-day-led-driver-is-fpga-dev-board-in-disguise/ https://hackaday.com/2020/01/24/new-part-day-led-driver-is-f... is probably the most affordable FPGA board that has been fully reverse engineered with 2 x gigabit ethernet ports and SDRAM.
bsder 6y ago
The mega-FPGA's are irrelevant to the consumer and small business segment. Sure, the guys doing software defined radio need beefy FPGAs (but, even they're starting to use much lower class ones). However, over the past 20 years, the things I have seen small businesses use FPGAs for rarely stresses the FPGA fabric even slightly. Only a single company managed to actually fill an FPGA that we had to bump to the next class of the same family. Mostly the usages are just: 1) extra pins to control/sequence "stuff" or 2) one single latency sensitive thing or 3) ADC communication. I'm struggling to think of any company I did work for that was outside those applications--"safety lockout" would probably count. Back in the days of 8-bit micros (yeah ... I'm old), you would probably offload the PID control loops so they didn't clog your micro. Nowadays, microcontrollers are so powerful and FPGAs are so annoying to program that you're better off shunting as much as you can into the micro and minimizing the FPGA. > Plus, as the number of gates and other elements gets larger, sophisticated software with a lot of domain knowledge becomes more important I'm going to call this out. I simply don't believe it. The publications of Xilinx and Altera (when there are any) simply look like the same algorithms from 20+ years ago but with Moore's Law applied to them. FPGA tools still suck at anything that looks like a "datapath".

lnsru 6y ago

From user perspective: the tools are shitty, version control isn’t existing. But I don’t have time to self host any open source tools. Nor my colleagues do. We must write code and design electronics around FPGAs. My manager will never pay for time spend with open source products. Big Corp I work for has a dedicated team with direct contact to Xilinx support if something goes badly. Next thing is legacy designs of the Big Corp. If it was Virtex-6, the next one will be Virtex-7. If it was Spartan-6, next one will be Spartan-7. Even if Altera/Intel or Lattice or some cool startup could offer similar chips for 60% the Xilinx’s price we would stay with Xilinx because of the codebase. And don’t forget supply chain, as Big Corp buys tons of chips from single vendor the price is unbeatable. Changing vendor and starting with clean account doesn’t make sense financially. And I am sure, that this single Big Corp makes more money for Xilinx than all the hobbyist worldwide. Honestly I can’t see a reason why Xilinx should open source anything.

UncleOxidant 6y ago
Sure, that's the vendor lock-in that Xilinx and Altera both bank on - A sort of Stockholm syndrome. But as newer players enter the FPGA field (a lot of the basic FPGA patents have expired over the last five years or so) open source is going to be the best bet for them to both get decent tools and win some market share. It's not like Xilinx/Altera will lose any significant market share anytime soon, but I can see a slow erosion on the lower end over the next 10 years or so. It'll probably look like the FPGA pie getting bigger with newer players (like Lattice) expanding into the smaller design niche. BigCorp has apparently been doing FPGA designs for many years now, but there are a lot of smaller companies that are just starting to consider jumping into doing FPGA designs.
- lnsru 6y ago
  I am not sure if there are any good areas for products with FPGAs for new players. I am working on one, but the single reason for that is that I am not good with Raspberry Pi. If I had cofounder with embedded Linux skills I would immediately ditch my FPGA design. 25$ price tag of Raspberry Pi compute module 4 is very attractive. I’ll end with ~100$ FPGA board making whole product more expensive.