Just sharing my $0.02 here - I have ethical objections to using OpenAI or Anthropic products so I was a reluctant adopter of LLMs at all. Local models address most, though not all, my moral objections so I’ve been using them for work and personal projects for about a month.
The hardware I have (32gb Macs and a gaming PC with 10gb 3080) can only get me to Qwen3.6-35B-A3B at various quants but that’s enough (200-400 PP, 20-30 TG).
It’s taken some time to learn how to best utilize it - some things take a bit of babysitting or direction - but it’s quite useful. Not having ever used CC I can’t compare but it’s been a great assistant or pair programmer for everything from embedded C++ to Vue. I wish I could run 27B as there have been moments when this model feels like it just can’t quite figure something out but those moments are quite rare. For a lot of tasks it’s a huge time saver and has proved super capable at digging into and fixing bugs given pretty vague instructions.
I’m making a digital ESP32 powered synthesizer at https://subalpinecircuits.com/. It’s been a ton of fun, learning so much about every part of the process (and I don’t use LLMs). Currently I’m learning FreeCAD and figuring out what my case design will be. Woodworking, CAM and DFM is a whole other world for a software guy!
This nails my primary frustration with all gen AI - why are we all seemingly okay with a few massive companies and their billionaire CEOs training models on the output of all human civilization and then selling it back to us with the promise of putting all workers out of a job? How’s that not theft?
Well, you as a human train and live as a human on the output of all human civilization and if you are very efficient put people out of work. 99% of human jobs were physical labor and now machines do the work of thousands with one human superviser (oil tanker, machine loom, dump truck, train, etc). If humans are put completely out of the loop, that is a new problem for humans (super intelligent AI that takes over will likely be very bad for us) but avoiding that problem is another ball of wax.
The electricity aspect has always been one I wish AI skeptics/critics wouldn’t lean on so much, because it seems like the one most likely to be solveable. I think we should be focusing on the economic aspects instead - if all this pans out as boosters promise, how does our economy exist when all white collar jobs go away? Who buys all the stuff, with what money?
I discovered the same thing with u8g2, and digging through the abstraction layers it felt like improving it was going to be impossible. Sending a single transaction with a framebuffer is so much simpler and faster.
SSD1322 looks great and might be something I look at for the future..
Interesting! That could be good way to boost the speeds here for sure, as I'm still pushing out a full framebuffer out with every update and am not usually updating the whole screen.
(author here) I've been pondering this, yeah. I'm currently sharing the I2C bus with a DAC and that's working alright, but the refresh rate issue is enough to make me consider SPI. I know the SPI peripheral supports DMA as well, and the I2C one doesn't (sort of? I know there's "async" transmit now but can't tell if that's really doing DMA)
The I2C peripheral is DMA - a write just queues an operation descriptor in a hardware FIFO, and interrupts fire whenever something exciting happens or the FIFO is drained. The synchronous implementation is just blocking on a condition set by the interrupt handler.
But this is exactly the kind of thing the SPI peripheral is meant for. This cannot be said for the I2C peripheral, with its measly fast mode support.
Depending on the chip and your usecase, you might also want to use I2S for your DAC.
