Raspberry Pi Compute modules

[Dirk Broer]

Whereas the original single-core ARM11 Compute Module was, in my eyes, a complete dud and I never saw the use -BOINC-wise- of owning one, this changed
with the introduction of the Turing Pi at the end of the economical life of the original Compute Module's form factor, by then the Compute Module 3+.
You could cram up to seven compute modules on a board the size of an mini iTX board, giving you a 28-core ARM computer(!), but with all the limitations of the
original Compute Module, being not enough RAM per core, the needed six-rail PSU limiting the CPU speed, no PCIe, etc.

So then came the much more powerful Compute Module 4 (in short: CM4) and every Tom(my), Dick(y) and Harrie(t) went out of his/her head and designed their
own carrier boards or embedded designs. There were also some who thought "better a good copy than bad original design" and made modules with the
same form factor as the CM4, but with additional connectors for extra features, so that their Compute Modules fare better on their own boards than on those of others.
As each module needs a carrier board anyway it isn't too bad, you just have to make sure you are not limiting your module by your carrier board

So far we have, besides the 2x 100-pins connectors equipped Raspberry CM4, Pine64's SOQuartz and Banana Pi CM4,
the Orange Pi CM4 (with 2x 100- plus 1x 24-pins connectors), the Radxa Rock CM3 and CM5 (both with 3x 100-pins connectors)
and the Banana Pi CM5 (with 2x 100- plus 2x 70-pins connectors).

Feature	Pine64 SOQuartz	Orange Pi CM4	Radxa Rock CM3	Raspberry Pi CM4	Banana Pi CM4	Banana Pi CM5	Radxa Rock CM5
SOC	Rockchip RK3566	Rockchip RK3566	Rockchip RK3566	Broadcom BCM2711	Amlogic A311D	Amlogic A311D2	Rockchip RK3588S
CPU	Quad-Core Cortex-A55 @1800 MHz	Quad-Core Cortex-A55 @1800 MHz	Quad-Core Cortex-A55 @2000 MHz	Quad-Core Cortex-A72 @1500 MHz*	Quad core ARM Cortex-A73 @2200 MHz and dual core ARM Cortex-A53 @1800 MHz	Quad core ARM Cortex-A73 @2200 MHz and quad core ARM Cortex-A53 @1800 MHz	Quad core ARM Cortex-A76 @2400 MHz and quad core ARM Cortex-A55 @1800 MHz
GPU	Mali-G52 2EE Bifrost GPU @800-950 MHz	Mali-G52 2EE Bifrost GPU @800-950 MHz	Mali-G52 2EE Bifrost GPU @800-950 MHz	Broadcom VideoCore VI @500 MHz	Arm Mali-G52 MP4 6EE Bifrost GPU @650-850 MHz	Arm Mali-G52 MP8 8EE Bifrost GPU @	Arm Mali-G610 MP4 Odin Valhall GPU @600 MHz
NPU	RKNN NPU 0.8 TOPS INT8	RKNN NPU 0.8 TOPS INT8	RKNN NPU 0.8 TOPS INT8	-	VeriSilicon NPU 5 TOPS INT8	? NPU 3.2 TOPS INT8	RKxx NPU 6 TOPS INT8
RAM Memory	2, 4 or 8GB LPDDR4-3200	1, 2, 4 or 8GB LPDDR4-3200	1, 2, 4 or 8GB LPDDR4-3200	1, 2, 4 or 8GB LPDDR4-3200	4GB LPDDR4	4 or 8GB LPDDR4	4, 8 or 16GB LPDDR4-4224
eMMC Storage	16, 32, 64 or 128GB on carrier board	16, 32, 64 or 128GB on module	0, 8, 16, 32, 64 or 128GB on module	0, 8, 16 or 32GB on module	16, 32, 64 or 128GB on module	16, 32, 64 or 128GB on module	0, 8, 16, 32, 64 or 128GB on module

* Just as easily overclocked and using the same settings as a regular Pi 4: mine runs @2000 MHz under my old AMD K6-III 400 Cooler of some 5(l) x5(w) x4(h) centimeters..

to be continued.

[Dirk Broer]

A year has passed, are there more compute modules to be found?
Compute Modules by SOC.jpg (click on the picture to make it bigger)
Wow! That's a whole lotta Compute Module 4 clones...let's try to condens the information a bit.
Compute Modules by Connector.jpg (click on the picture to make it bigger)
But are they any good, Geekbench-wise?
Compute Modules by Geekbench Score.jpg (click on the picture to make it bigger)

[Dirk Broer]

The six-core Banana Pi BPI-CM4 and the eight-core Orange Pi CM5 and Radxa-CM5 modules take the top marks, but keep in mind that only the BPI-CM4 can truly be fitted on a Raspberry Pi CM IO board with full compatibility and that -while it can be run from the Raspberry Pi CM IO board- the Radxa-CM5 really prefers its own 3x 100 pins Radxa-CM5 IO board - even the older 3x 100 pins Radxa-CM3 IO board won't do, while the Orange Pi CM5 also does best on its own 3x 100 Orange Pi CM5 Base Board