3 ms·

The recent Raspberry Pi models (4/400) have the potential to really expand computing globally. They're the first ones that have enough CPU power and RAM to do v

The recent Raspberry Pi models (4/400) have the potential to really expand computing globally. They're the first ones that have enough CPU power and RAM to do virtually everything you'd ever want to do with a desktop computer, with the exception of playing the latest games (and perhaps ML? I don't know much about it), but they cost $45 and run from a USB power supply. It's amazing how much you can do with them.

MontyCarloHall 4y ago
>with the exception of … perhaps ML? I don't know much about it. It's amazing how much you can do with them. Indeed. Cloud VMs can fill in the needs unmet by the Pi—the amateur ML hacker generally only needs high performance computing resources infrequently enough that renting them on-demand from cloud providers is an extremely economical solution.
ffhhj 4y ago
It's out of stock everywhere.
- LeoPanthera 4y ago
  That would be the "potential" part, but I don't think the supply chain problems are limited to the Raspberry Pi. (Also, it's often in stock, check https://rpilocator.com https://rpilocator.com )
georgia_peach 4y ago
The only fly in the ointment is the lack of SATA or NVMe on the base model. I think there are riser boards out there for that, but it's all very boutique.
- jhgb 4y ago
  While it's true that SATA or NVMe would be ideal, I just recently started booting a Pi 4 straight from a Samsung T7 SSD and it's night and day compared to the SD card even over just USB. I'm sure we'll get to better interfaces one day, but so far I'm very happy regardless.
- sgtnoodle 4y ago
  The USB 2.0 controller seems to be relatively inefficient on the pi 4, or at least the driver is. An RTLSDR dongle uses almost 50% CPU just shovelling samples when it should only use less than 1%.
  - georgia_peach 4y ago
    I wonder how well PXE boot + NFS shares would fare in comparison? Hasn't a lot of the USB stack been pushed into user space in more recent operating systems?
    - sgtnoodle 4y ago
      Well, the Ethernet interface is PCI-E, so there's a good chance it would be efficient. One thing to keep in mind is that USB 3.0 is a completely different independent controller from 2.0. So, just because USB 2.0 sucks doesn't necessarily mean 3.0 does. It's just unfortunate because you would expect 2.0 to be well optimized by now. I assume the kernel driver must be using the CPU to shovel data from the USB 2.0 controller rather than DMA.

MontyCarloHall 4y ago

>with the exception of … perhaps ML? I don't know much about it. It's amazing how much you can do with them. Indeed. Cloud VMs can fill in the needs unmet by the Pi—the amateur ML hacker generally only needs high performance computing resources infrequently enough that renting them on-demand from cloud providers is an extremely economical solution.

ffhhj 4y ago

It's out of stock everywhere.

LeoPanthera 4y ago
That would be the "potential" part, but I don't think the supply chain problems are limited to the Raspberry Pi. (Also, it's often in stock, check https://rpilocator.com https://rpilocator.com )

georgia_peach 4y ago

The only fly in the ointment is the lack of SATA or NVMe on the base model. I think there are riser boards out there for that, but it's all very boutique.

jhgb 4y ago
While it's true that SATA or NVMe would be ideal, I just recently started booting a Pi 4 straight from a Samsung T7 SSD and it's night and day compared to the SD card even over just USB. I'm sure we'll get to better interfaces one day, but so far I'm very happy regardless.
sgtnoodle 4y ago
The USB 2.0 controller seems to be relatively inefficient on the pi 4, or at least the driver is. An RTLSDR dongle uses almost 50% CPU just shovelling samples when it should only use less than 1%.
- georgia_peach 4y ago
  I wonder how well PXE boot + NFS shares would fare in comparison? Hasn't a lot of the USB stack been pushed into user space in more recent operating systems?
  - sgtnoodle 4y ago
    Well, the Ethernet interface is PCI-E, so there's a good chance it would be efficient. One thing to keep in mind is that USB 3.0 is a completely different independent controller from 2.0. So, just because USB 2.0 sucks doesn't necessarily mean 3.0 does. It's just unfortunate because you would expect 2.0 to be well optimized by now. I assume the kernel driver must be using the CPU to shovel data from the USB 2.0 controller rather than DMA.