Recently an Igalia engineer posted a NUMA Emulation patch for the Pi 5 to the Linux Kernel mailing list. He said it could improve performance of Geekbench 6 scores up to 6% for single-core, and 18% for multicore.
My testing didn't quite match those numbers, but I did see a significant and consistent performance increase across both Geekbench 6:
And High Performance Linpack:
Since the Pi 5's launch, a number of Pi case redesigns have launched, and there are a few new entrants with something to offer. Like Fractal's 'Baby North'... which, unfortunately, is only a prototype designed for their displays at Computex, and is not being planned for sale. At least not for now! I'll write more about this case later in this post.
The Pi 5's thermals are close enough to the Pi 4 that old cooling solutions work okay, but the port layout and inclusion of a power button means at least minimal redesigns are necessary.
Here are a few of the Pi 5 cases I've been testing (most for over a month, in various places), and my thoughts on each.
The official case for the Raspberry Pi 5 is like a saltine cracker.
For a few weeks I've been beta testing remote shell, the latest addition to Raspberry Pi Connect. Just a couple hours ago I was on a flight home from the new Micro Center in Charlotte.
One huge problem with VNC or remote desktop is how flaky it is if you have limited bandwidth or an unstable connection, like on an airplane.
It takes forever to start a screen sharing session, and the airplane's flaky WiFi usually causes the session to lock up, meaning you can't do much at all.
Remote terminal access, just relaying text commands, is the best solution for that problem. And sure, I have a VPN I could use with SSH to get to my Pi, but Raspberry Pi Connect just added support for remote shell access.
I gave away 480 Raspberry Pi Picos at Open Sauce last weekend, and ran into a number of challenges doing so. All of them self-inflicted, of course. I didn't want to just hand them out like candy—or, well... that's exactly what I did:
My initial plan was to build a backpack mount for a full 480-Pico reel (they sell them in bulk like that, for pick-n-place machines). However, there was a major flaw with that design.
I needed to get through TSA, so I could fly to San Francisco. Driving was out of the question, as my wife is almost full-term and I could not leave her for more than a week in the middle of summer, when the kids have about 15 events per week!
A full reel would require a knife or scissors, and I didn't want to check the bag whatever I built was in.
I'm in full-on procrastination mode with Open Sauce coming up in 10 days and a project I haven't started on for it, so I decided to try building the stable AI PC with all the AI accelerator chips I own:
After my first faltering attempt in my testing of Raspberry Pi's new AI Kit, I decided to try building it again, but with a more 'proper' PCIe setup, with external 12V power to the PCIe devices, courtesy of an uPCIty Lite PCIe HAT for the Pi 5.
I'm... not sure it's that much less janky, but at least I had one board with a bunch of M.2 cards instead of many precariously stacked on top of each other!
Raspberry Pi today launched the AI Kit, a $70 addon which straps a Hailo-8L on top of a Raspberry Pi 5, using the recently-launched M.2 HAT (the Hailo-8L is of the M.2 M-key variety, and comes preinstalled).
The Hailo-8L's claim to fame is 3-4 TOPS/W efficiency, which, along with the Pi's 3-4W idle power consumption, puts it alongside Nvidia's edge devices like the Jetson Orin in terms of TOPS/$ and TOPS/W for price and efficiency.
Google's Coral TPU has been a popular choice for a machine learning/AI accelerator for the Pi for years now, but Google seems to have left the project on life support, after the Coral hardware was scalped for a couple years about as badly as the Raspberry Pi itself!
In the past, I've booted LibreELEC on the Raspberry Pi Compute Module 4 in my "This is not a TV" Sharp NEC display.
According to LibreELEC's Pi 5 blog post, the new BCM2712 SoC decodes 4K and 1080p content just fine in H.264, and supports HEVC 4K60 hardware decoding.
And they've tested AV1, VC1, and VP9 at 1080p with no issue, though 4K in non-native formats does encounter frame dropping.
I wanted to put the Pi through some testing of my own, now that the Pi 5's been out for months, and LibreELEC version 12 is stable.
Besides the Pi 5 being approximately 2.5x faster for general compute, the addition of other blocks of the Arm architecture in the Pi 5's upgrade to A76 cores promises to speed up other tasks, too.
On the Pi 4, popular image processing models for object detection, pose detection, etc. would top out at 2-5 fps using the built-in CPU. Accessories like the Google Coral TPU speed things up considerably (and are eminently useful in builds like my Frigate NVR), but a Coral adds on $60 to the cost of your Pi project.
With the Pi 5, if I can double or triple inference speed—even at the expense of maxing out CPU usage—it could be worth it for some things.
LattePanda's been building Intel-based SBCs for almost a decade, but until now, they've never attempted to unite an Intel x86 chip with the popular SoM-style form factor Raspberry Pi's dominated with their Compute Module boards.
This year they've introduced the LattePanda Mu, a SoM that marries an Intel N100 SoC with a new edge connector standard they've designed, using a DDR4 SODIMM form factor.
Right now they offer two carrier boards: a lite version with basic interfaces and a couple 2230-size M.2 slots for SSDs and wireless, and a full evaluation carrier that breaks out every hardware interface in a Mini ITX-sized motherboard.
For years, SBCs that aren't Raspberry Pis experimented with eMMC and M.2 storage interfaces. While the Raspberry Pi went from full-size SD card in the first generation to microSD in every generation following (Compute Modules excluded), other vendors like Radxa, Orange Pi, Banana Pi, etc. have been all over the place.
Still, most of the time a fallback microSD card slot remains.
But microSD cards—even the fastest UHS-II/A2/V90/etc. ones that advertise hundreds of MB/sec—are laggards when it comes to any kind of SBC workflow.
The two main reasons they're used are cost and size. They're tiny, and they don't cost much, especially if you don't shell out for industrial-rated microSD cards.
