3 ms·

Qualcomm already tried a custom ARM chip and failed at it. That was Krait. The "secret sauce" behind the M1 is that it's actually more expensive than a similarl

Qualcomm already tried a custom ARM chip and failed at it. That was Krait. The "secret sauce" behind the M1 is that it's actually more expensive than a similarly powerful Intel chip, and far, far more expensive than a typical ARM chip. The BOM cost for ARM Cortex is pennies, while M1 is easily 100x that. Apple is in the luxury business, so they can afford to spend a lot and take margins. Android phones are not luxury devices and spending as little as possible gets you poor performance and unimpressive results. ARM's greatest marketing trick was convincing people their architecture was so much better while it was really just their interest in entering low margin industries to begin with that secured them the mobile market. Intel was never interested in low margins, so they never entered that market segment. If you want a good CPU, you have to spend money on it.

rmah 5y ago
Silicon chip production costs are based on three factors: area, layers and resolution. That's pretty much it. The complexity of the chip's design (on a single layer) is not really that major a factor. So.. Cost = Area * Num_Layers * 1/Process_Scale. As far as I'm aware Apple M1 chips are not an order of magnitude (much less two) larger in area, layer count or resolution vs Intel/AMD or similar ARM chips. Thus it's highly unlikely for them to cost 100x more to produce. 2x or 3x more expensive I could believe, 100x... nope.
- stefan_ 5y ago
  That ignores defects, where chip design and things like market segmentation come into play. Defect in a flash chip is easily worked around, but how do you sell a M1 with a broken CPU core? No M.5 out there.
  - cwizou 5y ago
    FYI, Apple does have two variants of the M1, with 7 or 8 GPU cores (those have a large area and are likely targets for defects) : https://en.wikipedia.org/wiki/Apple_M1 https://en.wikipedia.org/wiki/Apple_M1 Defects in CPUs (not GPUs) for segmentation hasn't really been a thing for a while, maybe the Phenom X3, and that was already a bit of a stretch from what I recall as most X3s had a perfectly functional 4th core. A cheap cellphone SoC is definitely not pennies, <10$ is a better estimate and it's usually without RAM included which is not always accounted for in comparisons. I vastly agree with rmah and most estimates do too.
sensorsbroken 5y ago
What's a smarket?
gwmnxnp_516a 5y ago
On mobile systems, the major concern is low power usage and battery duration, not performance, hence the dominance of ARM-based chips in this field. The great appeal of M1 chip is the performance and low battery power usage. ARM is the Microsft of the semiconductor world. The company design and licenses CPU cores for several market segments, including: ARM cortex-M series cores intended for MCU - microcontrollers SOC (System-On-Chip); CPU-cores intended for embedded systems processors (MPU) such as ARM-Cortex-A series intended for embedded systems and mobile devices. And there are also ARM cores designed for servers, for instance A64FX Fujitsu chip that powers the Japanese supercomputer Fugaku, until now the fastest in the world. ARM cores highly customized with microarchitecture, such as Apple M1 and some Samsung ARM chips, are designed under architecture license. A MCU is a self-contained chip containing up to 1MB of RAM, Flash memory up to 1 MB for the firmware; and several other peripherals such as digital IO, analog-to-digital converter, UART RS232 and so on. They are not comparable to Intel chips. MCUs are used in keyboards, mouses, car engine control unit, toys, home appliances and so on. The difference between the cores intended for MCUs and cores intended for MPUs is that cores for MPUs need external flash memory chips and external RAM memory. MPU chips and also they contains MMU - memory management unit for supporting virtual memory and operating systems that need this feature, such as iOS, Windows CE, Linux, BSD and etc.

rmah 5y ago

Silicon chip production costs are based on three factors: area, layers and resolution. That's pretty much it. The complexity of the chip's design (on a single layer) is not really that major a factor. So.. Cost = Area * Num_Layers * 1/Process_Scale. As far as I'm aware Apple M1 chips are not an order of magnitude (much less two) larger in area, layer count or resolution vs Intel/AMD or similar ARM chips. Thus it's highly unlikely for them to cost 100x more to produce. 2x or 3x more expensive I could believe, 100x... nope.

stefan_ 5y ago
That ignores defects, where chip design and things like market segmentation come into play. Defect in a flash chip is easily worked around, but how do you sell a M1 with a broken CPU core? No M.5 out there.
- cwizou 5y ago
  FYI, Apple does have two variants of the M1, with 7 or 8 GPU cores (those have a large area and are likely targets for defects) : https://en.wikipedia.org/wiki/Apple_M1 https://en.wikipedia.org/wiki/Apple_M1 Defects in CPUs (not GPUs) for segmentation hasn't really been a thing for a while, maybe the Phenom X3, and that was already a bit of a stretch from what I recall as most X3s had a perfectly functional 4th core. A cheap cellphone SoC is definitely not pennies, <10$ is a better estimate and it's usually without RAM included which is not always accounted for in comparisons. I vastly agree with rmah and most estimates do too.

sensorsbroken 5y ago

What's a smarket?

gwmnxnp_516a 5y ago

On mobile systems, the major concern is low power usage and battery duration, not performance, hence the dominance of ARM-based chips in this field. The great appeal of M1 chip is the performance and low battery power usage. ARM is the Microsft of the semiconductor world. The company design and licenses CPU cores for several market segments, including: ARM cortex-M series cores intended for MCU - microcontrollers SOC (System-On-Chip); CPU-cores intended for embedded systems processors (MPU) such as ARM-Cortex-A series intended for embedded systems and mobile devices. And there are also ARM cores designed for servers, for instance A64FX Fujitsu chip that powers the Japanese supercomputer Fugaku, until now the fastest in the world. ARM cores highly customized with microarchitecture, such as Apple M1 and some Samsung ARM chips, are designed under architecture license. A MCU is a self-contained chip containing up to 1MB of RAM, Flash memory up to 1 MB for the firmware; and several other peripherals such as digital IO, analog-to-digital converter, UART RS232 and so on. They are not comparable to Intel chips. MCUs are used in keyboards, mouses, car engine control unit, toys, home appliances and so on. The difference between the cores intended for MCUs and cores intended for MPUs is that cores for MPUs need external flash memory chips and external RAM memory. MPU chips and also they contains MMU - memory management unit for supporting virtual memory and operating systems that need this feature, such as iOS, Windows CE, Linux, BSD and etc.