Electronics > Projects, Designs, and Technical Stuff
Modular FPGA-Only computer
nctnico:
If you just want raw computing power then nothing beats a GPU when it comes to power consumption and cost. Look into CUDA.
james_s:
--- Quote from: mr_darker on August 31, 2019, 02:21:43 am ---Also where do people find tech-oriented friends lol.
--- End quote ---
I've met most of mine at work, and a few on online forums like this. If you can succeed at a project like this, surely you can find a job in tech if you don't already have one, at the right company you will meet lots of like minded people. If you are in a tech-heavy region you might look for hiring events if only to get to know people. I've been to a couple of those and it's a bit like looking for a date in a pub except instead of chatting up cute girls I was approaching nerdy looking guys who looked like they might be interesting to talk to.
NiHaoMike:
--- Quote from: nctnico on August 31, 2019, 10:41:01 pm ---If you just want raw computing power then nothing beats a GPU when it comes to power consumption and cost. Look into CUDA.
--- End quote ---
That's if the task is well suited to run on GPU. Some crypto algorithms, most notably those used in some cryptocurrencies, are specifically designed to not run well on GPU and a FPGA or even a cluster of cheap smartphones get better performance per watt and/or better performance per dollar. ASICs do the best, but the development costs mean they only get made if there's a big financial incentive to do so.
--- Quote from: james_s on August 31, 2019, 11:59:35 pm ---If you are in a tech-heavy region you might look for hiring events if only to get to know people. I've been to a couple of those and it's a bit like looking for a date in a pub except instead of chatting up cute girls I was approaching nerdy looking guys who looked like they might be interesting to talk to.
--- End quote ---
If only there were more girls in tech so that it would be easier to find one willing to "join forces" and help with taking on more advanced projects that are a bit complex for one individual.
Kilrah:
--- Quote from: daqq on August 31, 2019, 08:32:24 pm ---That's how it works as I understand it.
You define a region and then compile modules that 'fit' into the region.
--- End quote ---
Yes, but usually you do that in your own design and define the regions for the couple of little things you intend to change - here you need to figure out a good way to define multiple regions of various sizes that cover the vast majority of the fabric in a way that you hope will be useful to people. Something like 2 quarters, 2 8ths, 2 16ths etc, and provide interconnects so that people with a large design can split it to fit in multiple of your predefined blocks and still connect them together...
laugensalm:
--- Quote from: mr_darker on August 31, 2019, 02:11:42 am ---The idea would be that instead of allocating time on a processor, the OS allocates circuitry/hardware. Yes this would requre programs to have an fpga version of their own, which is where the hard processor comes in for compatibility with programs not made for foga architecture. Im sure when out of hardware resources youd be forced to add more hardware, close other programs, or do something like share time on the hardware (or resort to aprocessor). Anyways if it can be done itllbe done byporting linux over to the fpga only computer.
..
--- End quote ---
Well, there are many who are on the same page as you. Some use C++ classes to define and build the hardware, some use classic XML descriptions, some use Python or pure HDL. Why don't you check this thread:
https://www.eevblog.com/forum/fpga/2019-the-best-way-to-learn-cpu-on-fpga/
Linux might be a little too heavy weight though. Why would you want to do that? FPGAs perform well at real time and very flat hardware abstractions. That's not really what Linux performs well at. You'll also run into many, many issues with DDRx RAM, Caching, and the whole lot. You might want to keep it simple and stupid.
Navigation
[0] Message Index
[*] Previous page
Go to full version