New Pico2

The polarized inducor is strange. I'd expect that a small shielded inductor would work fine. It is interesting to have a switching regulator on a CPU chip... near a 12-bit ADC!

The Pi designs struggle to save pennies and microwatts, which not all of us care about.

I've had buck converters sensitive to inductor orientation before. Last time was with MIC4576 It was sensitive to the orientation of an inductor with the winding horizontal, similar to that shown in the RP2350 data. But the inductor had no polarity markings. It was changed to a stand up toroid which had no orientation issues.

There are products other than the Raspberry Pi available for those with more money to thr^H^H^Hspend.

Also no floating point, if I’ve understood correctly?

It’s a slightly odd device, isn’t it?

If you wanted to explore RISC-V then there’s more flexible options. If you just wanted a microcontroller for something and didn’t care too much about CPU architecture then the dual-architecture thing is wasted. Two entire CPU cores that you don’t get to use.

I just had to re-check the datasheet and you're right, the Hazard3 cores do not handle floating point. That's deeply disappointing.

The M33 cores do support single and double precision floating point, single floating point within the M33 cores and double precision maths using one of its three coprocessors (gpio, maths and the other I forget). I guess we could just ring the doorbell to get the ARM core to do the maths for the Hazard3 core. :-D

Yes the datasheet says the unused cores are gated off.

..or not...

(I have no dog in this fight. I grew up on CISC. I use whatever works. All I care about is price power performance )

Microcontroller RISC-V cores are so small that I presume it's just thrown on there as a test to see how things pan out. You can do a basic RV32 in a few hundred lines of SystemVerilog and, if it doesn't have its own on-chip memory, it takes pretty minimal area. It's the area for on-chip RAM, on-chip flash, registers, caches and TLBs that costs.

It sounds like one of their devs had a hobby core on his github and they put it on the chip just because they could. What I'm more interested in is what QA they did on it and what tools they used. It's easy to write a core but embarassing if it doesn't do what you think it does[1]. Perhaps the 'enabling/disabling' mechanism is a chicken bit to wall it off in case something bad is discovered in it.

Theo

I wish I could take that statement back to say 1975.

In those days you'd 'write' the core... by drawing the transistors with pens (or sticky tapes, hence 'tape out').

But those cores gave us computing power which we could use to design the next generation of chips. Repeat for 50 years and now the computer does the drawing and we just have to tell it what to draw (and write the compiler, of course).

Theo

I like the Pi for more reasons than the cost. The chips have long-term availability guarentees, the documentation is excellent, and there are zillions of kids out there who are familiar with the Pi culture.

Even ignoring the 70 cent price, the RP2040 is an amazing part.

Our computer thinking evolved when CPUs filled rooms and cost megabucks, and RAM cost a dollar per byte. Things have changed.

Virtual memory worked around the cost of RAM and encouraged complexity and bloat. Similarly, c calling conventions gave us hazards.

It's time to rethink things. CPUs and RAM are cheap, bloated buggy code is expensive.

In real life, done is better than fun.

Long term, the ARM license may be a liability.

Arm are an investor in RPi, so not sure why they would give them a hard time over licensing.

Theo

I agree. Being able to build a wifi connected mcrocontroller project for under £10 including PCB and case is a win for me.

I grew up with microcomputers that had 32K of RAM and tape players. In some ways I agree, in other ways I disgree.

We now have glorious displays with a billion colours, mulitmedia capability and different ways to interact with these machines. 

I think it's worth the tradeoffs.

There's limited forms of memory protection that can mitigate not having a MMU.

Floating point only available with the M33 cores. Hazard3 cores don't.

Maybe the RISC-V cores are to "send a message" to ARM?

WTF?

They arrived today (quite good for ordering on Sunday)

The tiny one lives up to its name, about the size of a mini SIM card, even the PGA one is only an inch square, has nothing but 1/10" pin layout, should be good for embedding ... I'll get a picoplus2 when the second batch become available