@MLE_online asked for more garden-related posts so here's my contribution. We found a fairly large sweet potato tuber lurking in a bed that had been planted with rooted slips from a batch of greens bought at a local Asian grocery and gave it a chance to sprout in a pot of pearlite under some bright LED lights. No idea if the potato is tasty but we know the greens are good so this will propagate some more.
Notes from the bench: got all the dspod interfaces working and the primary audio effects application is working. Porting to ESP32P4 *and* upgrading to ESP-IDF V6.x required some changes to the way I handle the audio and CV DMA, but that's done and things are working.
Latest project on the bench is a new dspod daughtercard. This one is based on the ESP32P4 in the form of the M5Stack Stamp P4 module (used here because the bare ESP32P4 isn't available through distribution yet). This MCU has no wireless capability, but does include dual-core 360MHz RISC V CPUs, 32MB of PSRAM in package and all the rest of the typical peripherals you'd expect on an ESP32. So far I've got the I2S port talking to a codec. More to come as I bring up ADC, LCD, buttons and DSP algos.
Stan Lee and cat
#catsandartists
#caturday
Must be July 4th in Phoenix
Tim Koh and cat
#catsandartists
#caturday
This board is intended to work with the NO2 USB bootloader from @tnt that I've used on a few previous boards. I'll be testing that out soon, but there's a lot of verification to go through on the other interfaces first. For now I've confirmed the FPGA is alive with a quick LED blinky
Latest in the dspod series of digital audio experiments - the dspod UP5k based on an iCE40 UP5K FPGA. It's also got flash, PSRAM, ADC, codec, digital I/O, USB fullspeed and a programmable clock generator. These FPGAs are neither fast nor particularly "roomy", but I'm curious to see how much I can cram into one in this context.
I'm particularly pleased with how the assembly went on this - reflowed all the SMT parts rather than doing some by hand and I think it turned out very nicely. Now to test.
For the last few weeks I've had a new dspod on the bench that's based on the STM32U3C5CIT6 MCU. This is a low-power device with some unique new features. I've posted the hardware and firmware sources on Github along with some design notes here:
https://github.com/emeb/dspod/tree/main/dspod_u3c5
This includes driver test code for all the on-board resources including the HSP hardware signal processor and the ADF PDM-PCM filter peripheral. There's also a skeleton of a simple GUI-based multi-effects application.
Jean-Edern Hallier and cat
#catsandartists
#caturday
Last week I spent some time adding yet another STM32 family to the @1bitsquared Blackmagic project. The STM32U3xx family is now supported for program flash and debugging (option bytes not yet). I used it to bring up a new member of my dspod boards today - all working just fine.
Scenario that happens more often lately -- out for my daily bike ride, bent down & pedalling hard into a headwind. Meanwhile some old fart passes me, sitting bolt upright, pedalling easily, not breaking a sweat. Kinda demoralizing until I notice that they're on an e-bike.
I've been using the Blackmagic Probe from @1bitsquared to do embedded development on Arm parts for years. It's a very slick bit of kit that supports so many devices and is tightly integrated with the Gnu Debugger (gdb). Recently I've had several contributions merged in that support some new parts - the STM32H7RS used in one of my dspod boards as well as the STM32C542 on the Cheep4 module I mentioned the other day. If you do embedded, definitely give it a look.
I've been working on the Cheep4 Eurorack oscillator module for the last week or so, making progress bringing up the various hardware features. The basic oscillator functions are working, including a handful of different waveform generation modes - saws, quadrature sinusoids, 2-operator FM and additive synthesis. The additive algorithm makes use of the hardware CORDIC coprocessor in the STM32C542 MCU I'm using, allowing up to 64 harmonics at a 96kHz sample rate without overloading the CPU.
Now reading "Night Watch" by Pratchett - a story about what happens when a large, complex economy is ruled by idiots and enforced by cruelty. Yes, it's a simplistic view, and yes it's written partly for laughs, but there's a definite kernel of correctness in there that applies oh so strongly to the things we're seeing every day lately.
How long will it be before a kid without a lot of cash can afford to spin up an agent to knock out the next big thing? Is it even reasonable to imagine training and using these tools independently without billion dollar financing?
With the explosion of hardware pricing caused by the datacenter buildout, even low-end hardware for machine learning may soon be out of reach for the average nerd. Is the future of tech locked up behind a paywall and available only to those who can pay the rent?
fin
Free / student tier pricing may provide some relief for learners and onboarding, but the restrictions applied will not enable productivity at the professional level that is required for meaningful progress & keeping up with the leading edge.
We saw some of these effects decades ago in the engineering & scientific disciplines when tools for computation (Matlab) or EDA (PCB design, HDL synthesis, chip layout, etc) were out of reach for hobbyists. Today there are free tools for these apps.
3/x
When I see LadyAda talking about spinning up an agent to do driver development and how it costs upwards of $100 to run a few hour session I'm forced to see this as a barrier to entry for people who can't / won't afford that kind of expense. So *if* this development flow becomes the minimum viable approach to software dev in the future, only professionals and well-funded enterprises can stay in the game. Free / open-source projects and individual hobbyists will be stuck in the backwater.
2/x
Like many folks I'm concerned about the intrusion of AI/LLM/agents into software (and especially free / open-source) development for all the known reasons: ethics, deskilling, resource use, societal impacts, etc. but one hazard that I'm not seeing people talk about is cost. Using these tools at the levels that are emerging as maximally effective is *expensive*, even factoring in that the providers are underpricing to capture the markets.
1/x
Hayao Miyazaki and cat
#catsandartists
#caturday
