Woah, 375Mhz overclock, from a stock 133Mhz:
https://github.com/tvlad1234/pico-rv32ima/blob/main/pico-rv3...
Almost 3x, without cooling...!
https://github.com/tvlad1234/pico-rv32ima/blob/main/pico-rv3...
Almost 3x, without cooling...!
Anything over 200mhz and the SPI flash doesn't work anymore either. It can be pretty hard to fit all working memory and code into the 256KB.
The chip was designed to replace Arduino boards, many of which run below 50mhz, and only process 8bits at a time. 133mhz, 32bit dual core is a sizable jump in performance already.
https://github.com/pimoroni/picosystem (see the note next to the no_overclock compile flag)
https://www.trustedparts.com/en/part/winbond/W25Q16JVSSIQ%20...
The flash chip they are using is only rated to 133mhz, so they are setting the SPI clock divider.
in hardware.cpp
#ifndef NO_OVERCLOCK
// dividing the clock by two ensures we keep the spi transfer to
// around 62.5mhz as per the st7789 datasheet when overclocking
sm_config_set_clkdiv_int_frac(&c, 2, 1);
#endif
It means they probably tried to do too little in each clock cycle. Some multi-cycle operations could have been made single cycle...
Or, they designed the part for one silicon process, but ended up changing path and manufacturing on a different one late in the design process.
Either way, the outcome is the users will be paying more per part than they needed to.
There are very few embedded applications where you need more computation power than a RP2040 can offer without being limited by RAM or flash. Sure, they could have tried pushing it even harder, but why bother when it is already enough?
To be fair to them, I think it was a fairly inexperienced design team making their first silicon, in relatively low volumes (10's of millions). They weren't an established player trying to make a 1 billion+ production run super competitive product where fractions of a cent matter.
Clock signals on big motherboards are great RF radiators and hard to shield. But on a SoC, they aren't much of an issue.