4 ms·

I'm very excited about India's ambitions. It would not only be meaningful for India, but for many people around the world. We are lacking open chips, and projec

by madez 7y ago

I'm very excited about India's ambitions. It would not only be meaningful for India, but for many people around the world. We are lacking open chips, and project Shakti is on a trajectory to solve a part of that. I wonder when GCC learns to use the source code of the hardware along with the source code of the software to optimize the binary with full knowledge of the system.

analognoise 7y ago
We will never have competitive chips that are open down to the transistor and process level. Down to the RTL, sure. But a truly open competitive chip, never ever. You might get one on some ancient process from 20 years ago, which will be handily outclassed in every single metric except openness, which people make noise about but don't actually pay for (so no business case).
- happycube 7y ago
  A ~20 year old process... that'd be 250nm or 180nm, which is Just Fine for many microcontrollers, if not the new 600mhz-1ghz Cortex M7's...
  - analognoise 7y ago
    If the 180nm design is good enough, and migrated it to a new process, you'd get more chips per wafer. Hence never competitive. "It's open!" doesn't make a business decide to buy a million and put them in a product - cost, features, availability, support do. Nobody cares about "open silicon" - it's only spouted by people who don't understand actual electronics, or if they do, they certainly don't understand the economics. It will never happen. If it did, it wouldn't be useful. It makes zero sense.
    - ghettoimp 7y ago
      From a corporate perspective, I think you're absolutely right, this kind of thing will never be profitable. From a state perspective, being reliant on other countries for your electronic may seem like such a threat to your ability to remain sovereign that it's worth developing your own, even if you are generations behind the latest and greatest.
      - analognoise 7y ago
        True! But do you think they'd open up that process? Hell no!
saagarjha 7y ago
> I wonder when GCC learns to use the source code of the hardware along with the source code of the software to optimize the binary with full knowledge of the system. Wouldn’t this just mean not having to rely as much on Agner Fog?
hackworks 7y ago
GCC uses intrinsic functions when available during compilation. It does optimize for a specific architecture. Is there something beyond this you are referring to?

by madez 7y ago

analognoise 7y ago

We will never have competitive chips that are open down to the transistor and process level. Down to the RTL, sure. But a truly open competitive chip, never ever. You might get one on some ancient process from 20 years ago, which will be handily outclassed in every single metric except openness, which people make noise about but don't actually pay for (so no business case).

happycube 7y ago
A ~20 year old process... that'd be 250nm or 180nm, which is Just Fine for many microcontrollers, if not the new 600mhz-1ghz Cortex M7's...
- analognoise 7y ago
  If the 180nm design is good enough, and migrated it to a new process, you'd get more chips per wafer. Hence never competitive. "It's open!" doesn't make a business decide to buy a million and put them in a product - cost, features, availability, support do. Nobody cares about "open silicon" - it's only spouted by people who don't understand actual electronics, or if they do, they certainly don't understand the economics. It will never happen. If it did, it wouldn't be useful. It makes zero sense.
  - ghettoimp 7y ago
    From a corporate perspective, I think you're absolutely right, this kind of thing will never be profitable. From a state perspective, being reliant on other countries for your electronic may seem like such a threat to your ability to remain sovereign that it's worth developing your own, even if you are generations behind the latest and greatest.
    - analognoise 7y ago
      True! But do you think they'd open up that process? Hell no!

saagarjha 7y ago

> I wonder when GCC learns to use the source code of the hardware along with the source code of the software to optimize the binary with full knowledge of the system. Wouldn’t this just mean not having to rely as much on Agner Fog?

hackworks 7y ago

GCC uses intrinsic functions when available during compilation. It does optimize for a specific architecture. Is there something beyond this you are referring to?