https://turingpi.com logo
Powered by
Rpi cm5
# │forum
e
With cm5 now released, does anyone know how well this works with Turing pi2. Currently looking ti upgrade the two compute modules 4 I have in my Turing pi2.
r
The CM5 has the same 2 100-pins board-to-board connectors that the CM4 has, but I've heard the pinout is slightly different. It's said the CM5 should largely be a drop-in replacement for the CM4, but it might need a slightly different adapter board for the Turing Pi 2, especially if you want to access the new features such as the PCIe 3.0 lane, possibly hooking that up to the NVMe storage on the bottom of the Turing Pi 2.
d
You probably want to wait. We don't know about compatibility between the TPi CM4 adapter and the RPi CM5. Based on my experience with the Radxa CM5, a module that works okay on the official RPi CM4 IO Board may not work correctly on the TPi CM4 adapter. Might be resolvable by reworking u-boot and the device tree, but it's too early to know.
l
I have tested the cm5 in a cm4 compute adapter but was not able to boot it in a turing pi 2.4. ( no status en activity leds) So there are changes needed in the adapterboard to accomidate with the changed pin layout
Small update, was able to boot the cm5 in slot 1 of the tpi but not in slot 3 or 4, so further testing is required
x
There's a few pinout changes https://datasheets.raspberrypi.com/cm5/cm5-datasheet.pdf Appendix C: Differences between CM4 and CM5
d
Does it boot in node slot 2? If so, that might indicate the PCIe Gen 2 lead routing is wrong. You could potentially test this theory with a USB 3.0 or SATA mPCIe card on node 1 or node 2. A schematic or a document containing signal routing of the Turing Pi CM4 adapter would be very helpful. I can try to use a multimeter to "tone out" the signal leads and build a table.
l
Will check it later today when i am back home
@Dan, i've tested your hypothesis and looks that is indeed the case. I already had a 2 SATA mPCIe cards plugged in for slot 1 and 2 and that are indeed the slots where the cm5 is able to boot
Slot 3 without any mPCIe of NVME connectivity won't work and Slot 4 with an NVME drive installed (non boot) is also non functional with an CM5
When the CM 5 is deployed in slot 3 or 4 the activity led lights up for an bit. Go's off and on two times and then stays on for a bit. So it looks like something cicles at that moment
d
I wouldn't expect any support for the underside M.2 NVMe slots unless Turing Machines decides to redesign the adapter. My suggestion to them would be to open source that piece of hardware. "The Cathedral and the Bazaar" comes to mind.
Do you have a UART-to-USB cable? At least you might be able to see what the loader and u-boot are doing.
l
I know that NVME support is not possible ATM. Did mention it for completeness and also forgot to remove it when tested it the first time 😅 Open souring the adapter would be nice so the community could develop muliple versions for supporting mPCIe and NVME. But that is out of our hands at this moment
Yes I do, using the BMC to interface with the modules UART would not work but have not tested it yet using a seperate adapter
d
Kind of to be expected. No doubt some crafting of the device tree/u-boot/Linux configuration will be required to get the picocom connection set up.
l
Alright, did some extra testing using the UART but without any sucess. Looks like these headers are not exposed when using a cm 4 or 5 as descibed in the docs https://docs.turingpi.com/docs/turing-pi2-specs-and-io-ports#uart EDIT: Did enable uart btw. And tested the converter first on the UART of the BMC
Hmm interesting, judging by the picture of the TPI website there was in the past a adapter with a unpopulated header. Has anyone seen this in the wild or is it just a internal dev version before the 1.0 adapter? https://cdn.discordapp.com/attachments/1311437734308479016/1313233543513640961/Gray_CM4-Adapter.png?ex=674f638b&is=674e120b&hm=108554d9a4ffd99a6d572bfa99cb1f392af2ce5dfd1184e4fadb3432b69b448b&
d
My guess is J2 was a PWM fan header. The more interesting thing is that 4 position micro switch. The final production CM4 adapter design deleted J2 and left the micro switch and at least some of the supporting circuitry unpopulated. Those things could have been there to allow trace cross talk/interference testing. I read someplace here that the CM4 adapter was a complex beast to get right.
l
There are not many 3 pin use cases with a ground an 2 unknown pins. So it was either a PWM or JTAG header in the past :). I've ordered a new TP2 today with some extra adapter boards, so I can always try to resolder the micro switch on one of my modules
d
I would hold off on purchasing a naked RPi CM5 module for use in Turing Pi v2s. It's "early days". There appear to be several issues: - The CM5 does appear in 
lsusb
output when mass storage device mode is enabled for flashing (
tpi advanced msd - n <node number>
). However, the USB vendor+device identifier field is not recognized. - The module is not visible for flashing over the USB-C OTG port when the node slot is put into flash mode (
tpi usb flash -n <node number>
). - Raspberry Pi OS 64-bit desktop seems stable. Ubuntu 22.04.01 server is not. - Flashing the RPi CM5 devkit from Windows 10 doesn't work because the BCM2711 and BCM2712 boot drivers won't load, even after turning off driver signature enforcement. The RPi CM5 devkit was left active, but idle, overnight to burn in a bit I'll move the module back to a TPu v2.5 node slot 1 to see whether there are any signs of life.
l
Thanks for testing! Were you able to boot up the CM5 in slot 3 or 4 in the TP 2.5?
My new board is still underway so I am not able to test it myself atm
d
Not yet. I'll try to get to it today.
My plan is to test the pre-flashed CM5 module (desktop) in node slot 1 to see whether it boots, has HDMI output, the UART works and the USB-A port can be found. Then I'll flash the lite 64-but image and test other node slots. Update: The pre-flashed CM5 module (desktop) does boot and has HDMI output in node slot 1 on a TPi v2.5 board. No USB-A port, either with or without compatibility mode set. 
picocom /dev/ttyS1 -b 115200
from the BMC apparently connects to something, but there is no output. I'll check to see whether there is any UART I/O on the GPIO header. Update #2: Moved the module to a TPi v2.4 board. Got the same results. HDMI output works. No USB-A port, even when the port is placed into "Host" mode. The signals must not be routed properly. I'm uncertain whether this is something that could be fixed through device tree changes. I'll test a dual-port USB 3.0 mPCIe card to determine whether it could be a temporary/permanent workaround.
c
Hello! I have two RPi CM5 8GB modules and CM4 adapters, and I successfully flashed Ubuntu 24.04.1 LTS over USB on my Turing Pi board. However, both modules are running extremely hot. Without any cooling and under no load, CPU temperature rises above 90°C. I also get following message > This power supply is not capable of supplying 5A; power to peripherals will be restricted I'm using the power supply from the official store. Has anyone else experienced this issue? Could the overheating be caused by an insufficient power supply? is there a way to fix this? Turing Pi 2 (v2.5.1) Build version v2024.05.1
d
The RPi CM5 is not currently supported on any Turing Pi v2 cluster board. Please proceed with extreme caution with this configuration because you may void any warranty on the Turing Pi v2, the PicoPSU and associated transformer, the Turing Machines CM4 adapters and the Raspberry Pi CM5 modules. I'm not making this statement glibly: proceed at your own risk.
c
Thank you! I didn’t expect any issues with the original Raspberry Pi. I haven’t had any issues with the Radxa CM5, so this is quite unfortunate. What’s more confusing is that the Turing Pi Twitter account confirmed that “CM5 is a direct replacement for the CM4 on this board.” https://x.com/turingpi/status/1863898857291014638
d
The Radxa CM5 mostly works on the Turing Machines CM4 adapter in a Turing Pi v2.3/v2.4. The Turing Pi v2.5 USB-A port routing in Compatibility Mode uses the wrong USB port on the Radxa CM5. The RPi CM4 uses USB1 for this function, but the Radxa CM5 does not. I believe this issue can be resolved with a modified cm5-rpi-cm4-io device tree and/or changes to u-boot/Ubuntu. This is a project that I probably won't be able to tackle until I can get some form of Turing Pi v2.5 documentation done. Right now I'm acting as a carbon-based user guide to the community. Obviously, not a scalable solution.
l
Hi, @castorp the CM5 shares a lot of the internals with a Raspberry pi 5 in my experience and is quite new ATM. I see a mention of running Ubuntu 24.04.1 on the module, however this is not supported configuration based on the Ubuntu docs. So maybe this issue will be gone when testing it with the official Raspberry OS instead of Ubuntu. Raspberry OS is more of a first party OS then Ubuntu and should have the best compatibility in place at this point in time https://cdn.discordapp.com/attachments/1311437734308479016/1319393847813476402/image.png?ex=6765ccc6&is=67647b46&hm=6d055c62302ef1b99d9c4c92ca2aac3bdb4e971bbb0c4c50e482e4f7ea79ac01&
Small update, I did some digging off the missing UART output of the CM5 and it look like the primary debug UART has been re routed through a seperate microcontroller instead of the GPIO pins. The Raspberry pi 5 has had the same change in that regard. The CM5 has a unpopulated pad whitch could be used to solder on a 3 pin molex connector Have bought some connectors so wil try to add the connector to the board. I will have around the holiday's some free time so this will be a holiday project for me 🙂
d
I tried Ubuntu 24.04.1 via the RPi installer on my RPi CM5 devkit. It boot loops. I went to the latest Raspberry Pi 64-bit OS for the RPi 5, which does boot, but doesn't function anywhere near fully on a Turing Pi v2 and only in node slot 1.
l
I think that that the support of the CM5 will be provided in the next non LTS release of Ubuntu. The last version is a tad to old for official support. I had planned to test my modules in my new TPi2.5 but the Chinese warehouse made a mistake. So I have received something but not my new board. ( Has already been escalated to Turing machines for investigation)
d
Yeah, I saw your case. Probably not a warehouse mistake. It will get fixed. I figure Ubuntu support will come, but requires an updated kernel, so likely 24.10 (Oracular). The RPi CM5 was only released at the end of November.
I ordered and received an active cooler for the RPi CM5. The male end of the fan cable uses a smaller connector than the Turing Pi v2, so the extension cables are incompatible. Not a big deal yet. I'll either make or purchase a short adapter cable. Tom's Hardware evaluated this heat sink (https://www.tomshardware.com/raspberry-pi/raspberry-pi-compute-module-5-receives-the-active-cooler-that-it-really-needs). Net: provides equivalent results to CM5 passive heatsink unless the CM5 is overclocked. In this environment, the active heatsink prevents the BCM2712 from thermal throttling. Installing this heatsink eliminates the need for the case-mounted fan on the Raspberry Pi CM5 dev kit. https://cdn.discordapp.com/attachments/1311437734308479016/1321992279040331776/PXL_20241226_222502963.jpg?ex=676f40c0&is=676def40&hm=e73b0e585fc06507bbd05faf2cbe4c70a6049e756e2c1f06cc973cdd5ee38c7b&
p
out of curiosity, when using raspbian you mentioned it doesnt function fully on the turing pi 2, can you expand on that?
d
Sorry, I've been occupied trying to support customers with no time to work on research/science projects. I'll take a wack at it again tonight. The USB-A port on the Turing Pi v2.5.2, even in "Compatibility Mode" does not function with an RPi CM5 in node slot 1. Might be resolvable with device tree changes.
l
Update: I was able to mod my cm5 with the inclusion of a small terminal for UART en SWD and could capture the boot logs in the tpi 2.4 using the Raspberry pi debug probe. Did also receive my new 2.5 board, so I can start the tests on this board also
l
Hi, i was wondering if there is an update? Im thinking about getting the latest turing pi board but it is a bit unclear to me if you can run the cm5 on it? Im fine with flashing it first on a different system
r
Nearly 4 months now since Pi CM5 release... are we ever going to get an adapter board?
Well, a year since CM5 release, radio silence from Turing Machines. Hell, official development on anything supporting the Turing pi 2 seems to be vaporware now. Certainly has turned me off of supporting kickstarters in the future. “Here’s your beta product… f-you on support”
r
Hey, I agree, but at least they delivered.
r
Did they really, though? I mean, yea, they delivered a board, but it is missing the supporting software to do half of what they currently advertise it to do.
h
has anyone been able to get cm5 working stable with the cm4 adapter?
y
What do you mean, specifically? We have a set of four boards here, with all peripherals lying around, which wait to be assembled. What are we going to run into? I'm aware of https://github.com/turing-machines/BMC-Firmware/issues/250
j
It boots, runs on my TP2.5.2 Board since multiple days with Raspbian. * In slot1 and 2 the MiniPCIe devices are detected, too. * The serial works also. * There are issue with CM5 in Slot 3 & 4 * And you cannot flash it currently via the BMC * I have a CM5 without eMMC and it does not recognize the SD-card on the adapter
h
thanks! have you tried nvme?
j
yes via minipcie->m2 adapter, the kernel command line might need pcie_aspm=off
c
Rpi5 works fine on the Turning Pi 1 board with a few small modifications to the Waverunner board.
b
on the turing pi 1 you say?
c
Yep. With a Weverunner board slightly modified to supply +5 to the board instead of 3.3 it works great.
I love the Turning Pi 1. As a bonus, the CM5 also runs the switch at full 1gbs, without any other upgrade.
The CM3 and CM4 were 100BaseT. Blah.
Wish they would release a driver to enable control of the Gigabit switch to create VLANs. Only thing I miss.
s
3061 Views
PreviousNext
Powered by