3 ms·
At some point soon, Fab lines are going to become long term investments, instead of being depreciated in 5 years. Wafers won't get any larger, transistors won't
by mikewarot 4y ago
At some point soon, Fab lines are going to become long term investments, instead of being depreciated in 5 years. Wafers won't get any larger, transistors won't get any smaller, and it'll be on our hands as programmers to tweak every last ounce of performance out of things again.
I know the "demo" scene folks are keeping the craft alive, and it gives me hope for the future.
Of course, an array of non-cached, non-optimized but very simple and low power processors in a grid can do a lot more work than a single instruction-reordering monster trying to make a single thread go ever faster.
- throwawaylinux 4y agoThey are already. Commercial foundries operate lines that use decades old technology. GF has a .6 micron line that uses 200mm wafers for example. > Of course, an array of non-cached, non-optimized but very simple and low power processors in a grid can do a lot more work than a single instruction-reordering monster trying to make a single thread go ever faster. True if you look at theoretical best cases. Not always true on real problems, and it's not always the fault of the programmer if the problem can not be written to suit the flock of chickens processor. Limiting factors are latencies and communication overhead between various dependencies, which is the case whether you expose that to the programmer with your grid, or try to deal with in in the core with multiple execution units and reordering.
- pjmlp 4y agoIt is going to be a sad day in Electronland.