3 ms·

> This is wishful thinking. ... Most people in business simply won't mix x86 and ARM binaries on Windows so they're going to be one field or the other. Those sa

> This is wishful thinking. ... Most people in business simply won't mix x86 and ARM binaries on Windows so they're going to be one field or the other. Those same people want to run those Windows tools at home too. ARM Macs will be a niche product for those looking to spend excessively. I think this ignores the historical reasons for x86 success. Intel made their bones by crushing Unix boxes and mainframes with high volume x86 parts. PC CPUs were the highest volume microprocessor for decades, and they leveraged that into making high performance, yet low cost server and workstations parts to buff up their margins. Well, the tables have turned - ARM is far, far higher volume than x86 due to its ubiquitous use in smart phones. And ARM is going to do to Intel what Intel did to IBM, Sun, SGI, HP, DEC, etc. The problem is the same as it always has been: NRE (fixed costs) for new chips climbs ever higher and needs to be offset by even more volume to be acceptable to consumers. ARM won that game in 2007. It's just taken this long before market forces and product maturity have forced Intel's hand. To be fair, Intel may have staved off the inevitable for a few years without their well publicized 10nm production stall.

bgorman 6y ago
> The problem is the same as it always has been: NRE (fixed costs) for new chips climbs ever higher and needs to be offset by even more volume to be acceptable to consumers. While this is certainly true for bleeding edge nodes, hasn't the NRE become significantly lower for mature process chips? I would guess that EDA tools and a greater availability of IP blocks would drive down the real dollar cost of designing your own ASIC on a node 1-2 generations back.
- nordsieck 6y ago
  > While this is certainly true for bleeding edge nodes, hasn't the NRE become significantly lower for mature process chips? I would guess that EDA tools and a greater availability of IP blocks would drive down the real dollar cost of designing your own ASIC on a node 1-2 generations back. Honestly, I'm not familiar enough to have a nuanced opinion, but from what I understand, that's probably true. However, in the battle for cloud VPS and PC CPU sockets, using a 1-2 generation old process doesn't appear to be a winning strategy.
pritovido 6y ago
>ARM is far, far higher volume than x86 due to its ubiquitous use in smart phones Smart phones, tablets, TVs, projectors, routers and also refrigerators, 3D printers and ovens. Their volume is 10x to 20x times bigger than PC and laptops.

bgorman 6y ago

> The problem is the same as it always has been: NRE (fixed costs) for new chips climbs ever higher and needs to be offset by even more volume to be acceptable to consumers. While this is certainly true for bleeding edge nodes, hasn't the NRE become significantly lower for mature process chips? I would guess that EDA tools and a greater availability of IP blocks would drive down the real dollar cost of designing your own ASIC on a node 1-2 generations back.

nordsieck 6y ago
> While this is certainly true for bleeding edge nodes, hasn't the NRE become significantly lower for mature process chips? I would guess that EDA tools and a greater availability of IP blocks would drive down the real dollar cost of designing your own ASIC on a node 1-2 generations back. Honestly, I'm not familiar enough to have a nuanced opinion, but from what I understand, that's probably true. However, in the battle for cloud VPS and PC CPU sockets, using a 1-2 generation old process doesn't appear to be a winning strategy.

pritovido 6y ago

>ARM is far, far higher volume than x86 due to its ubiquitous use in smart phones Smart phones, tablets, TVs, projectors, routers and also refrigerators, 3D printers and ovens. Their volume is 10x to 20x times bigger than PC and laptops.