4 ms·

While Atmel MCUs are nice and robust (especially avr-gcc and avr-libc are great advantages) they seen like quite high cost. Cheaper ARMs probably are eating the

by jwatte 11y ago

While Atmel MCUs are nice and robust (especially avr-gcc and avr-libc are great advantages) they seen like quite high cost. Cheaper ARMs probably are eating their lunch. Meanwhile, I've never been friends with the microchip Dev tools. Clunky, expensive, proprietary. Here's hoping we'll get AVR ergonomics and PIC economics... But it might be too late to matter. Cortex-M0 eats the low end, M4 the mid end, and the phone SoCs the high end. (Even Intel may be threatened at some point... Different business though)

VLM 11y ago
The low end is something like the PIC 10F family, when compared to the Cortex-M0 its about a tenth the cost and a fiftieth the power consumed. Lighting an LED takes 100x the power of running a 10F series at full speed. I can buy some 10F series PICs cheaper than a 555 timer chip, which is pretty amazing.
- zokier 11y ago
  > The low end is something like the PIC 10F family, when compared to the Cortex-M0 its about a tenth the cost and a fiftieth the power consumed I don't really believe either of those claims. At 1ku quantity on Digikey (which admittedly is not the cheapest vendor) cheapest Cortex-M0 chips cost $0.378 and low-power chips (NXP KL03) start from $0.504. That aforementioned KL03 is advertised to run at 50 µA/MHz. Meanwhile cheapest PIC costs $0.35 and is rated 175µA at 4 MHz, which equals to 43.75 µA/MHz. Of course the power consumption figures are not directly comparable, and something like µA/MIPS might be better metric anyways. But based on this there is no way PIC is anywhere near fifty times more power efficient or ten times cheaper.
  - emeraldd 11y ago
    Several of the PIC models have a low power mode where they run off an internal RC circuit somewhere in the KHz range. Just enough to wake up on an interrupt or some external stimuli and really pushes their power usage way down. For a PIC10KF320/322 you're looking at: Sleep: 20nA, 1.8V Operating: 25uA @ 1Mhz, 1.8V Watchdog Timer: 500nA, 1.8V Of course that's dependent on what' you're doing with the io pins etc. See:http://ww1.microchip.com/downloads/en/DeviceDoc/40001585D.pdf http://ww1.microchip.com/downloads/en/DeviceDoc/40001585D.pd... edit: fix typo
    - makomk 11y ago
      For comparison, TI's MSP432 series is specified at 90 uA/MHz running, 850nA sleep with RTC enabled - and that's a Cortex M4 part with hardware FPU, not even an M0. I think Atmel's newest low-power M0 parts are even more impressive. Plus the Internet of Things means companies are wanting to run increasingly demanding code on low-power embedded devices. 8-bit chips are getting squeezed hard by ARM.
buserror 11y ago
Well there still is a case for the AVR -- you get zero pipelining effects, no flash latency etc; it's like a tiny, slow, scalar CPLD for all intent and purpose. Cortex can't beat that, you can tight loop an AVR to cycle precision and get pretty cool return for your running frequency; the Cortex are nowhere as linear. [0] Altho, I agree that the bigger parts are useless... xmegas and such have absolutely no reason to exist IMO. On the other hand the CM3 and especially M4's are awesome beast; and the parts like the STM32's are just peerless in terms of features -- they are just borderline scary. On the other hand, I was worried a little while back that there was nothing between these trusted 8 bits cores and the ARMs -- but nowadays there is; Espressif esp8266 is pretty cool, but the next one in the pipeline is awesome [1]; 2 (well 3 but) cores, and more peripherals than you can shake a stick at. Thats the first SoC that for me can give the STM32 (and NXP equivalent) a good run for their money (and win, in my opinion). Dislaimer: [0] I know the AVR very well, I'm the author of simavr: http://github.com/buserror/simavr http://github.com/buserror/simavr -- [1] I'm also one of the beta tester for the ESP31/32 new chipset.
- revelation 11y ago
  People care about time, not cycles, and chances are with whatever crystal is running your AVR, you will still not be comfortable despite the lack of post-1990 CPU features in ATmegas. And finally, the idea of tight loops engineered to cycle precision is dying a slow death in general. You just smother the hard timing requirements with excess frequency and performance. Curious thing about the ESP8266: it doesn't use ARM or any of the other popular kids on the block but a Tensilica Xtensa architecture.
ArkyBeagle 11y ago
Microchip dev tools are free. An ICD 3 isn't, but they're relatively inexpensive. Not sure, but I think the latest MPLAB-X is Eclipse. I have no complaints with them. Atmel AVR will redesign things out from under you. We could not get a new JTAG interface to work with an old chip.
IshKebab 11y ago
> But it might be too late to matter. Cortex-M0 eats the low end, M4 the mid end, and the phone SoCs the high end. You know Atmel makes Cortex M chips too right? They're very good in my experience - e.g. they have some of the very few Cortex chips that support USB High Speed (480 Mb/s).

by jwatte 11y ago

VLM 11y ago

The low end is something like the PIC 10F family, when compared to the Cortex-M0 its about a tenth the cost and a fiftieth the power consumed. Lighting an LED takes 100x the power of running a 10F series at full speed. I can buy some 10F series PICs cheaper than a 555 timer chip, which is pretty amazing.

zokier 11y ago
> The low end is something like the PIC 10F family, when compared to the Cortex-M0 its about a tenth the cost and a fiftieth the power consumed I don't really believe either of those claims. At 1ku quantity on Digikey (which admittedly is not the cheapest vendor) cheapest Cortex-M0 chips cost $0.378 and low-power chips (NXP KL03) start from $0.504. That aforementioned KL03 is advertised to run at 50 µA/MHz. Meanwhile cheapest PIC costs $0.35 and is rated 175µA at 4 MHz, which equals to 43.75 µA/MHz. Of course the power consumption figures are not directly comparable, and something like µA/MIPS might be better metric anyways. But based on this there is no way PIC is anywhere near fifty times more power efficient or ten times cheaper.
- emeraldd 11y ago
  Several of the PIC models have a low power mode where they run off an internal RC circuit somewhere in the KHz range. Just enough to wake up on an interrupt or some external stimuli and really pushes their power usage way down. For a PIC10KF320/322 you're looking at: Sleep: 20nA, 1.8V Operating: 25uA @ 1Mhz, 1.8V Watchdog Timer: 500nA, 1.8V Of course that's dependent on what' you're doing with the io pins etc. See:http://ww1.microchip.com/downloads/en/DeviceDoc/40001585D.pdf http://ww1.microchip.com/downloads/en/DeviceDoc/40001585D.pd... edit: fix typo
  - makomk 11y ago
    For comparison, TI's MSP432 series is specified at 90 uA/MHz running, 850nA sleep with RTC enabled - and that's a Cortex M4 part with hardware FPU, not even an M0. I think Atmel's newest low-power M0 parts are even more impressive. Plus the Internet of Things means companies are wanting to run increasingly demanding code on low-power embedded devices. 8-bit chips are getting squeezed hard by ARM.

buserror 11y ago

Well there still is a case for the AVR -- you get zero pipelining effects, no flash latency etc; it's like a tiny, slow, scalar CPLD for all intent and purpose. Cortex can't beat that, you can tight loop an AVR to cycle precision and get pretty cool return for your running frequency; the Cortex are nowhere as linear. [0] Altho, I agree that the bigger parts are useless... xmegas and such have absolutely no reason to exist IMO. On the other hand the CM3 and especially M4's are awesome beast; and the parts like the STM32's are just peerless in terms of features -- they are just borderline scary. On the other hand, I was worried a little while back that there was nothing between these trusted 8 bits cores and the ARMs -- but nowadays there is; Espressif esp8266 is pretty cool, but the next one in the pipeline is awesome [1]; 2 (well 3 but) cores, and more peripherals than you can shake a stick at. Thats the first SoC that for me can give the STM32 (and NXP equivalent) a good run for their money (and win, in my opinion). Dislaimer: [0] I know the AVR very well, I'm the author of simavr: http://github.com/buserror/simavr http://github.com/buserror/simavr -- [1] I'm also one of the beta tester for the ESP31/32 new chipset.

revelation 11y ago
People care about time, not cycles, and chances are with whatever crystal is running your AVR, you will still not be comfortable despite the lack of post-1990 CPU features in ATmegas. And finally, the idea of tight loops engineered to cycle precision is dying a slow death in general. You just smother the hard timing requirements with excess frequency and performance. Curious thing about the ESP8266: it doesn't use ARM or any of the other popular kids on the block but a Tensilica Xtensa architecture.

ArkyBeagle 11y ago

Microchip dev tools are free. An ICD 3 isn't, but they're relatively inexpensive. Not sure, but I think the latest MPLAB-X is Eclipse. I have no complaints with them. Atmel AVR will redesign things out from under you. We could not get a new JTAG interface to work with an old chip.

IshKebab 11y ago

> But it might be too late to matter. Cortex-M0 eats the low end, M4 the mid end, and the phone SoCs the high end. You know Atmel makes Cortex M chips too right? They're very good in my experience - e.g. they have some of the very few Cortex chips that support USB High Speed (480 Mb/s).