If you round to the nearest integer, Intel makes about 0% of the worlds CPUs. Thats right, Pentiums et al account for less than.5% of the CPU chip count. The rest are embedded CPUs. Even on a PC, which is the most Intel dominated platform, Intel is only 1 out of at least 5, usually more like 10. There are probably 4 cpus between your HD and your CD-ROM, plus all those keyboard controller, DMA engines, sequencers, etc.
Intel does, however, account for 90% of the revenue... bastards:)
At my previous job (large multinational, 400k+ employees), we were attempting to produce a DVD-ROM system-on-a-chip, integrating the CPU, decoder, servo controller, (some) memory, etc. Even using the a 0.18u process on our own fabs, it always came out cheaper to do 2+ chips.
You see, the cost of a chip this size (some of these integrated chips are huge) is largely driven by the yield, and yield goes down exponentially above a certain size. The bigger the chip, the more bad chips your're gonna make.
Our conclusion was that integration only makes sense if you HAVE to do it, for whatever reason (performance, pin out, ram size constraints), but it always turns out more expensive.
Another interesting Integration fact:
.5% of the CPU chip count. The rest are embedded CPUs. Even on a PC, which is the most Intel dominated platform, Intel is only 1 out of at least 5, usually more like 10. There are probably 4 cpus between your HD and your CD-ROM, plus all those keyboard controller, DMA engines, sequencers, etc.
:)
If you round to the nearest integer, Intel makes about 0% of the worlds CPUs. Thats right, Pentiums et al account for less than
Intel does, however, account for 90% of the revenue... bastards
At my previous job (large multinational, 400k+ employees), we were attempting to produce a DVD-ROM system-on-a-chip, integrating the CPU, decoder, servo controller, (some) memory, etc. Even using the a 0.18u process on our own fabs, it always came out cheaper to do 2+ chips.
You see, the cost of a chip this size (some of these integrated chips are huge) is largely driven by the yield, and yield goes down exponentially above a certain size. The bigger the chip, the more bad chips your're gonna make.
Our conclusion was that integration only makes sense if you HAVE to do it, for whatever reason (performance, pin out, ram size constraints), but it always turns out more expensive.