4 ms·

I know that. I've said something similar (only regarding operating systems, and approaches to security) elsewhere today: https://news.ycombinator.com/item?id=

I know that. I've said something similar (only regarding operating systems, and approaches to security) elsewhere today: https://news.ycombinator.com/item?id=32468423 https://news.ycombinator.com/item?id=32468423 Changing "too much" doesn't seem to be an option, so we're stuck with the stuff invented many many decades ago, and likely never will be able to overcome those local "optima" of the past—just because the market won't accept any disruptive changes any more by now. And sure: It's always the software… (As said also in the long reply to the sibling comment).

cmrdporcupine 4y ago
The good news is that FPGAs have democratized a bit of the HW stuff for us software folks. It's fairly trivial to get started with an FPGA dev board and building out our own custom CPU architectures, though of course it's likely going to be "just for fun." So even if a new and novel concept can't take over the world, we can build our own lightsabers, even just for hobby stuff.
- still_grokking 4y ago
  Yes and no. At least to validate my ideas FPGAs would not really be sufficient. Core to my idea are three things (on the HW-level): Runtime reconfigurability, dense and fast reconfigurable on chip networking, and asynchronous clockless logic. (Besides "smart memory", but that's something we can actually have). Nothing of that FPGAs really offer. (There is some kind of support for reconfigurability on the bigger Xilinx chips, but that does not really smell like what I'm looking for; also it's proprietary and very limited). FPGAs can of course simulate those named things. But without exposing their advantages. Actually such a simulation would be very slow compared to when doing things "the usual way". This is especially sad as FPGAs as such have already a lot of the other desired properties: They're reconfigurable hardware at all! So one would need first to build some new / modified kind of "FPGA" (especially adapted to async logic and runtime reconfigurability) only to try out my ideas in a halfway realistic setting. On the other hand, doing some ISA research and experiments doesn't need any hardware in the first place. This could be done fully in some software simulated environment. Only that this way it's impossible to know whether there would be any real merit in all this stuff. Simply, because I would expect a SW simulation to be quite inefficient and slow (even on a FPGA). So this wouldn't be anything that could convince anybody, frankly. Well, maybe I run at some point into some crazy person that has a few dozens $ mils. spare to throw at a "high-fantasy" project like that, to validate the basic assumptions. IF it would work it could actually change for the better everything we take for granted today about how computers work. Imho it's time. The von Neumann bottleneck is killing any progress! We should get rid of it. This needs some radical¹ rethinking of our machines (and especially their software). ¹ from lat. "radix", eng. "the root"

cmrdporcupine 4y ago

The good news is that FPGAs have democratized a bit of the HW stuff for us software folks. It's fairly trivial to get started with an FPGA dev board and building out our own custom CPU architectures, though of course it's likely going to be "just for fun." So even if a new and novel concept can't take over the world, we can build our own lightsabers, even just for hobby stuff.

still_grokking 4y ago
Yes and no. At least to validate my ideas FPGAs would not really be sufficient. Core to my idea are three things (on the HW-level): Runtime reconfigurability, dense and fast reconfigurable on chip networking, and asynchronous clockless logic. (Besides "smart memory", but that's something we can actually have). Nothing of that FPGAs really offer. (There is some kind of support for reconfigurability on the bigger Xilinx chips, but that does not really smell like what I'm looking for; also it's proprietary and very limited). FPGAs can of course simulate those named things. But without exposing their advantages. Actually such a simulation would be very slow compared to when doing things "the usual way". This is especially sad as FPGAs as such have already a lot of the other desired properties: They're reconfigurable hardware at all! So one would need first to build some new / modified kind of "FPGA" (especially adapted to async logic and runtime reconfigurability) only to try out my ideas in a halfway realistic setting. On the other hand, doing some ISA research and experiments doesn't need any hardware in the first place. This could be done fully in some software simulated environment. Only that this way it's impossible to know whether there would be any real merit in all this stuff. Simply, because I would expect a SW simulation to be quite inefficient and slow (even on a FPGA). So this wouldn't be anything that could convince anybody, frankly. Well, maybe I run at some point into some crazy person that has a few dozens $ mils. spare to throw at a "high-fantasy" project like that, to validate the basic assumptions. IF it would work it could actually change for the better everything we take for granted today about how computers work. Imho it's time. The von Neumann bottleneck is killing any progress! We should get rid of it. This needs some radical¹ rethinking of our machines (and especially their software). ¹ from lat. "radix", eng. "the root"