2 ms·
Simple. It takes time to get state of the art, bleeding edge CPU nodes to the point where they can have a consistently high yield of large-die-area chips, such
by gjsman-1000 4y ago
Simple. It takes time to get state of the art, bleeding edge CPU nodes to the point where they can have a consistently high yield of large-die-area chips, such as desktop processors. Yields on a new production line are often poor due to the many defects that would take out far too many large chips, but allows decent yields of smaller chips. Thus, the typical node starts manufacturing on small chips and works its way upward to larger chips as yields improve.
- freemint 4y agoThat or you produce defect resistant large chips like FPGAs which can literally route around any defect.
- gjsman-1000 4y agoBut FPGAs, despite being way faster than software, are only ~1/5th to ~1/10th as fast as dedicated silicon, at best, on a small scale. An Apple M1 in FPGA would run like it was manufactured 2 decades ago, if not worse. Plus, FPGAs get absolutely huge with terrible yields for complex work. RED Cameras have a massive, multi-thousand-dollar FPGA (crossing over $10K as a part), and it is only powerful enough to process their custom video codec and nothing else. Still cheaper than designing a custom chip for that considering how niche RED Cameras are, but absolutely stupid for anything broader. An FPGA that could run an M1 equivalent would be larger than the laptop, cost over $100K, and be slow as a turtle racing across Oregon.
- freemint 4y agoThat doesn't change the fact FPGAs have stupid high yields.