4 ms·

The long term issue is they are still two years away from a 10nm server chip while TSMC are already shipping 7nm parts (roughly equivalent to Intel 10nm). Furth

by AmVess 8y ago

The long term issue is they are still two years away from a 10nm server chip while TSMC are already shipping 7nm parts (roughly equivalent to Intel 10nm). Further, a return to 22nm chipset production indicates a large disruption in their product plans, as evidenced by Intel outsourcing production of some of their 14nm products to TSMC! Intel certainly gave away a two year process lead that is slipping to two years behind their competitors, which will certainly put an end to their lovely fat profit margins. Intel aren't doomed, but they gave away and will have given away all of their competitive advantage by sleeping at the wheel. Further, AMD have abandoned production of their high end server parts at Global Foundaries, so they will no longer be constrained in process capability or supply as they were in the past. It seems that no one in upper management have read the book, "Only the Paranoid Survive." Maybe it should have been required reading.

nickik 8y ago
Its also interesting that this happens right at a time where the switch from one node to another takes increasingly longer and viewer people adopt it. And anybody can now make chips in the most advanced nodes. There are startups now who are already working on TSMC7nm.
- kingosticks 8y ago
  Really!? What startup can afford those mask costs?
  - undersuit 8y ago
    Probably some Machine Learning startups.
  - nickik 8y ago
    Esperanto Technologies They are basically doing a AI chip with 8-16 general cores and 4000 floating point vector cores (probably with Tensor Type support on the Vectors). All in RISC-V.
vii 8y ago
Intel decided a couple of years ago to ramp down spending on new processes https://www.technologyreview.com/s/601102/intel-puts-the-brakes-on-moores-law/ https://www.technologyreview.com/s/601102/intel-puts-the-bra... Moore's law is not a law but a target improvement rate: companies aim to resource sufficient individually risky projects enough to improve the process to achieve Moore's law. Intel was far ahead of competitors on process and decided to ramp down their spending. This and the resurgence of the AMD platform are consequences. Was it the wrong strategy? Not clear at all - it would have been super expensive to keep going. They can choose to invest their huge cash pile and regain leadership

by AmVess 8y ago

nickik 8y ago

Its also interesting that this happens right at a time where the switch from one node to another takes increasingly longer and viewer people adopt it. And anybody can now make chips in the most advanced nodes. There are startups now who are already working on TSMC7nm.

kingosticks 8y ago
Really!? What startup can afford those mask costs?
- undersuit 8y ago
  Probably some Machine Learning startups.
- nickik 8y ago
  Esperanto Technologies They are basically doing a AI chip with 8-16 general cores and 4000 floating point vector cores (probably with Tensor Type support on the Vectors). All in RISC-V.

vii 8y ago

Intel decided a couple of years ago to ramp down spending on new processes https://www.technologyreview.com/s/601102/intel-puts-the-brakes-on-moores-law/ https://www.technologyreview.com/s/601102/intel-puts-the-bra... Moore's law is not a law but a target improvement rate: companies aim to resource sufficient individually risky projects enough to improve the process to achieve Moore's law. Intel was far ahead of competitors on process and decided to ramp down their spending. This and the resurgence of the AMD platform are consequences. Was it the wrong strategy? Not clear at all - it would have been super expensive to keep going. They can choose to invest their huge cash pile and regain leadership