What would you like to see in The Next Chip?

[Deouss]

Are you guys planning a chip based on Risc-V (64bit) ?
If price of KPU K210 drops it may be very very attractive choice comparing to esp.
Its specs are just insane. If not Risc-V then maybe some Xtensa 64bit cpu with higher clocks and better memory
Cheers

[Baldhead]

I agree with Deouss.

"Are you guys planning a chip based on Risc-V (64bit) ?
If price of KPU K210 drops it may be very very attractive choice comparing to esp.
Its specs are just insane. If not Risc-V then maybe some Xtensa 64bit cpu with higher clocks and better memory
Cheers"

[ESP_Sprite]

On using RiscV: Let's say that we're not a member of the RiscV foundation just because we had the spare cash and picked them at random apart from that, I don't want to divulge too much on roadmap matters. Is there a specific reason that you want to have a 64-bit processor for embedded stuff; why would a 32-bit one not be feasible?

[jrakus]

Is there a specific reason that you want to have a 64-bit processor for embedded stuff; why would a 32-bit one not be feasible?

I think that some guys don't catch the difference between embedded and "embedded", and would like to see ESP transforming into another RPi running full operating system. Anyway - if Espressif make such one-board computer (or maybe some kind of SoM), based on Risc-V architecture it will definitely shake the industry. But ESP32 in any version should be dedicated to bare-bone embedded systems, with constrains like supply, board space, I/O speed and ease of usage.

I think that next ESP32 chip shouldn't be a revolution. A little more peripherals (one more UART, one more I2C/I2S), little more RAM/Flash memory (+20%) and some improvements of existing ones (accurate ADC/DAC) will be fine. Version without on-board antenna (only RF pin) also will be interesting for very space constrained projects.

[Baldhead]

Hi,

K210.

8MB of internal ram.
"Generic dma".

The external 8MB of esp32 are unusable in dma.

[PaulNi]

Is there a specific reason that you want to have a 64-bit processor for embedded stuff; why would a 32-bit one not be feasible?

I think that some guys don't catch the difference between embedded and "embedded", and would like to see ESP transforming into another RPi running full operating system. Anyway - if Espressif make such one-board computer (or maybe some kind of SoM), based on Risc-V architecture it will definitely shake the industry. But ESP32 in any version should be dedicated to bare-bone embedded systems, with constrains like supply, board space, I/O speed and ease of usage.

I think that next ESP32 chip shouldn't be a revolution. A little more peripherals (one more UART, one more I2C/I2S), little more RAM/Flash memory (+20%) and some improvements of existing ones (accurate ADC/DAC) will be fine. Version without on-board antenna (only RF pin) also will be interesting for very space constrained projects.

Can't state the same better than you. 64 bit in embedded, just a lot more gates for little benefit. Pretty much the only good reason is having "infinite" virtual memory — that will be good for extremely VM wasteful languages like Java, but that's "lazy programming."

Java is of course a trendy lang, but programming MCUs in anything, but C or C++ is really not taking off.

Having more FPUs would be good for languages that don't use integer math as such like Javascript, but again, this sound like throwing tenths of megagates for just like 1-2% more potential clientele for the company.

Vector extensions can also be though off for more DSP like functionality for voice synthesis/recognition, or doing something with video, but at 240mhz double cores it should be already be doable without increasing the gate count. Doing more "wider" computations with vector extensions, VLIW, or such will of course add performance for a small, linear increase in power consumption, but really for what?

Just seconding the thoughts that ESP32 in its current iteration with double 240mhz cores is a great overkill for the amount I/O it can drive. Either I/O options can be extended, and more thought put to them, or a more cost optimised version can be made with a single beefy core. The later, as I understood, was the S2 chip.

[Deouss]

On using RiscV: Let's say that we're not a member of the RiscV foundation just because we had the spare cash and picked them at random apart from that, I don't want to divulge too much on roadmap matters. Is there a specific reason that you want to have a 64-bit processor for embedded stuff; why would a 32-bit one not be feasible?

64bit cpu would of course have more ways of manipulating registers, memory and theoretically more registers than 32 plus it could perform more precise double-precision computations without headaches)
As far as I remember memcpy on 64bit can be 2x faster on today's cpus. Just advantages - not overkill. Risc-V has also compressed instructions despite 64bit architecture.
So far esp32 cannot perform very well with vga displays and larger rgb pannels because of limited peripheral and cpu speed
I do not understand any engineer who uses python or javascripts to program MCU if you aim at performance and precision systems.
64bit MCU with high clocked peripherals would be perfect for that

[jrakus]

So far esp32 cannot perform very well with vga displays and larger rgb pannels because of limited peripheral and cpu speed
I do not understand any engineer who uses python or javascripts to program MCU if you aim at performance and precision systems.
64bit MCU with high clocked peripherals would be perfect for that

We are almost in 2020's - I do not understand any engineer who still uses tiny MCU to build another smartphone, that's the waste of time. ESP32 was designed for doing simple network-based tasks without the user interaction and should stay at this niche. If You need a computer just buy Raspberry.

[Deouss]

We are almost in 2020's - I do not understand any engineer who still uses tiny MCU to build another smartphone, that's the waste of time. ESP32 was designed for doing simple network-based tasks without the user interaction and should stay at this niche. If You need a computer just buy Raspberry.

I am talking about the future step-up chip not the current one that you favor so much.
Why the new chip couldn't compete with f.ex. K210 ? Tensilica LX6 is very good cpu and I think it should be paired with faster peripherals and maybe get faster memory and clock. I have nothing against esp32 - just looking forward for some step-up module.

[jrakus]

I am talking about the future step-up chip not the current one that you favor so much.
Why the new chip couldn't compete with f.ex. K210 ? Tensilica LX6 is very good cpu and I think it should be paired with faster peripherals and maybe get faster memory and clock. I have nothing against esp32 - just looking forward for some step-up module.

I think you're taking about some another chip and IMHO this topic is not about, but about next version of ESP32.

K210 and ESP are designed to do different tasks. They couldn't compete. K210 SoC is rather a CPU than MCU, is very fast, have plenty of RAM, rich audio/video processing skills, but doesn't have Wi-Fi and Blue. Evaluation boards with K210 have... ESP32 for Wi-Fi - what's a shame. ESP32 doesn't have so much computing power, but provides connectivity with 3 lines of code. And have Flash memory in-package, where is the one in K210?

To summarize: it will be very nice and very competing to another industry players if Espressif make a SoC/SoM/SBC with 64-bit multi-core based high computing power and built-in wireless connectivity. But it shouldn't be a replacement for ESP32 - such thing will be bigger, more power hungry and most important: will be more expensive. ESP32 really need new version, but it should be version, not totally new chip. Doubling the bits also double the price. I don' think it also double users satisfaction and Espressif income.