FPGAs are mostly used for small series and custom hardware. If you add the software price to the FPGA unit price, it becomes more expensive if you buy, lets say, 1000 FPGAs to ship in your product. Even though you might only license the software once.
It is a trade off, and I have no idea about the state and quality of Vivado. Back in the day I was tinkering with FPGAs, the Xilinx software stack was horrible.
Which is a pro-argument for the licensing change - maybe if they can make it a revenue stream, quality can improve. Back at Uni when we tinkered with FPGA in a lab course I was excited about the topic of reconfigurable computing, but I got mostly repelled by the bad state and quality of the dev tools and I (and some other students) did decide not to persue that field further. They lost some potential clients that year early on.
Ideally yes, SW should be free, but we don't live in an ideal world. This isn't Apple or Google who can give you SW for free since they take a 30% cut on everything on the Appstore besides the profit margins on the HW they sell you.
The typical customers of FPGAs are large HW companies with money to spend on SW, not tinkerers in their garage who might some day build a billion dollar company. And if you do become a billion dollar garage company, you will still buy their FPGAs because they're some of the best and the SW costs will be a rounding error plus a tax write-off. You think Anduril doesn't use Xilinx FPGAs because Palmer LUckey didn't get SW for free 15 years ago?
So there's zero incentive to give away costly to develop SW, for free.
> So there's zero incentive to give away SW, that's costly to develop, for free.
The cost of the software required to develop or operate the hardware should be included in the cost of the hardware. I say this as someone who's been into embedded development for 20 years. It's simply a part of the cost to make the hardware.
Support should be an additional paid service, that every single of my former employer would have paid. But the toolchain itself, as-is with no warranty, should be free.
> So there's zero incentive to give away SW, that's costly to develop, for free.
I don't think that's necessarily a hard rule, here. Nvidia won over the HPC segment by offering CUDA and PTX as a free value-add to their hardware, and ended up becoming a multi-trillion dollar company that ate AMD's datacenter market for lunch.
Giving away software to commoditize your compliment might be a good idea, for AMD.
>Nvidia won over the HPC segment by offering CUDA and PTX as a free value-add to their hardware
Yeah but people would buy Nvidia chips mainly for gaming, not (just) for CUDA, so basically the massive gaming clientele would finance expensive projects that don't yet make any money like CUDA. Meanwhile nobnody buys FPGAs for playing at home en-masse. There's no equivalent consumer market like gamers for the FPGA vendors, their sales are almost exclusively B-2-B.
FPGAs isn't something most people, even the hardcore tinkerers ned at home. Consumers into hardware tinkering are more than fine with what you can do with Raspberry PIs, ESP32s, STM32 boards.
Nvidia played dirty, sure, but they did give away expensive, professional B2B grade software for free. It bolstered their position as the dominant GPGPU vendor in a B2B context, obsoleted half-assed libraries like OpenCL, and killed competing businesses in the cradle. Gamers subsidized it, but not by much (especially nowadays).
IMO, this is a question about where AMD wants to be in the stack. They can sell hardware, they can license their IP, but they're going to be bent over a table selling software licenses that people don't want. A free HDL software suite can help you sell hardware and license IP, whereas an expensive one can cut you out of the FPGA/ASIC market entirely.
SW dev costs can be accounted for in the HW price tag, especially when you sell a lot of units, and assuming the SW is much better than "good enough". HW is the best licensing key; if you can do that, go for it.
>SW dev costs can be accounted for in the HW price tag
That would then make you chips more expensive than the competition and less competitive on the BOM side, and BOM costs are a different accounting budget than SW licenses.
People in the industry have already thought of this. HN likes to think they somehow found the magic solution nobody ever has thought before, when the truth is HN doesn't understand sales of the HW industry.
It depends on the HW:SW cost ratio and the volume. In favourable case the extra bit on the price tag can amount to a rounding error. Moreover, the price tag includes a profit margin, which is usually adjusted to match the competition already.
Besides, "ignorant" HN comments on e.g. FOSS software and funding problems have mentioned that the paperwork you have to do to buy software licences can be a PITA, so free software can be perceived as a plus when the R&D selects the parts.
People have already thought of that, sure. But has it been thought about and executed properly? Like the added value of open-sourcing the SW (made a lot easier when you don't need to copy-protect it). Or do they just go for the extra source of profit the SW licence sales provides, anyway?
In my Embedded work experience, the BOM costs matters more than SW licensing costs. Which matches the business model of Xilinx.
The company I worked for was selling million or tens of million of units per year. So in such cases it's easy to see why BOM costs trump the SW costs of a few licences, and why HW companies would like to keep chips costs down and charging you per SW license.
If you haven't shipped million of products in embedded and only worked in SW development, it's easy to get a warped picture of reality.
I haven't shipped millions of units but I do have occasional glimpses of the HW side as a firmware programmer. My impression is that decisions regarding costs are not always entirely rational, hence my questioning. Thanks for the answers.
Exactly, do they want people to buy their FPGAs or not? Charging per-seat licenses for developers and heavily restricting the free version mean people will buy from other vendors or just not use an FPGA at all.
It is absurd that in 2026 you have to pay for such tools. It feels like buying a propietary compiler in 80's or 90's.
No one wants that anymore.