10-28-2016, 07:39 AM
Well, you would want to start a business, right?
I've digged a bit into this area. Running a general purpose processor and a linux operating system can get you the same speed as a Raspberry zero at 10$ cost. It's low cost, but your old cpu would do it faster than 10 sticks. Apart from that, you got 70 usb ports? Usb also has an exact limit somewhere at 140 where no more devices can be attached cause the host controller can't control them all. But that's a different story.
You would want to either go for ASIC's or FPGA's or similar stuff. Ultimately, ASIC's would be the best way, but that means the highest effort and the highest risk cause of relatively high cost unless you get really big sales numbers.
OK, FPGA's then. They are basically ASIC's but programmable, which makes it easier for semiconductor companies to produce and sell them (they wouldn't need to produce a line for one customer thus dumping the price). You can get relatively potent FPGA's for like 10$ per chip or dev kits for like 20-30$ as a USB pen with some ports on it.
Telling you that this 10$ chip runs with 500MHz and 128kbit of total memory might underwhelm you, considering the 1GHz processor of the pi zero and the 500mb memory, but these FPGA's can be highly optimized and put in some really small casings.
The main problem is: How the fuck am I going to run a compiler written in c++ on native hardware. Basically dropping the OS or reducing it to a minimum.
After all dropping the OS and rewriting the pi zero's firmware to run a compiler directly and solely might actually be the best go, as it involves the least amount of work. Gcc's pretty sophisticated after all. After all I don't like the idea of using a pi 0 because it's relatively big in size. It contains a lot more than just the needed stuff and makes so much overhead.
Last but not least I want to name the idea that just came to my mind. An FPGA alone won't solve the problem. Targeting around 50$ would give you enough money to think about a "quad core" array of 4 FPGA chips (two on the up and two on the down of the stick) as well as 2 ram chips (again one up and one down). But I have no idea if it's ever possible to adjust gcc or rewrite a valid c++ compiler for that thing. I mean, if you even could use standard gcc code as it's released, it wouldn't help you as you would need to optimize it in order to get it faster. After all, FPGA's aren't just make-it-faster devices. You just drop the OS and build the stuff that's crucially needed yourself.
For God's sake, please leave the cmake on your computer.
I've digged a bit into this area. Running a general purpose processor and a linux operating system can get you the same speed as a Raspberry zero at 10$ cost. It's low cost, but your old cpu would do it faster than 10 sticks. Apart from that, you got 70 usb ports? Usb also has an exact limit somewhere at 140 where no more devices can be attached cause the host controller can't control them all. But that's a different story.
You would want to either go for ASIC's or FPGA's or similar stuff. Ultimately, ASIC's would be the best way, but that means the highest effort and the highest risk cause of relatively high cost unless you get really big sales numbers.
OK, FPGA's then. They are basically ASIC's but programmable, which makes it easier for semiconductor companies to produce and sell them (they wouldn't need to produce a line for one customer thus dumping the price). You can get relatively potent FPGA's for like 10$ per chip or dev kits for like 20-30$ as a USB pen with some ports on it.
Telling you that this 10$ chip runs with 500MHz and 128kbit of total memory might underwhelm you, considering the 1GHz processor of the pi zero and the 500mb memory, but these FPGA's can be highly optimized and put in some really small casings.
The main problem is: How the fuck am I going to run a compiler written in c++ on native hardware. Basically dropping the OS or reducing it to a minimum.
After all dropping the OS and rewriting the pi zero's firmware to run a compiler directly and solely might actually be the best go, as it involves the least amount of work. Gcc's pretty sophisticated after all. After all I don't like the idea of using a pi 0 because it's relatively big in size. It contains a lot more than just the needed stuff and makes so much overhead.
Last but not least I want to name the idea that just came to my mind. An FPGA alone won't solve the problem. Targeting around 50$ would give you enough money to think about a "quad core" array of 4 FPGA chips (two on the up and two on the down of the stick) as well as 2 ram chips (again one up and one down). But I have no idea if it's ever possible to adjust gcc or rewrite a valid c++ compiler for that thing. I mean, if you even could use standard gcc code as it's released, it wouldn't help you as you would need to optimize it in order to get it faster. After all, FPGA's aren't just make-it-faster devices. You just drop the OS and build the stuff that's crucially needed yourself.
For God's sake, please leave the cmake on your computer.