With the release of CM5, it is now possible to double the speed of the PCIe bus on the Compute Blade when working with the Compute Module from Raspberry Pi.
It’s quite easy to do, and so far all the tests I’ve done show high stability of the solution.
Compute Blade is fully prepared to work at these speeds and has an m.2 m-key port with very short lines to the compute module to avoid any electromagnetic interference and loss of speed or stability as much as possible.
Many off-the-shelf solutions can be used with Compute Blade as it uses an extended 22110 size slot.
The simplest example is using an NVMe disk, but there is more to it than that.
There are m.2 network cards, FPGAs, Ai-accelerators, and even graphics cards on the market. Furthermore, we have three new PCIe 3.0-enabled modules planned (to be added as they are released).
Jeff Geerling has put together a nice selection of devices tested in the field by the link
So, all you need to do is to add to config.txt (/boot/firmware/cmdline.txt)
at the very end of the file:
dtparam=pciex1
dtparam=pciex1_gen=3CM5 as well as Raspberry Pi 5 are not certified for PCIe 3.0 support and that’s probably why PCIe 2.0 is the default option.
That is, you should be prepared that in some cases or with some devices it will not work at all or will work with problems.
But as I said I haven’t had any problems so far because of the PCIe 3.0 activation.
Below are a couple of disk speed comparisons, benchmarks, and results under overclocking:
Default settings
Compute Blade v1-mk4
Samsung 970 EVO Plus
The most telling line is HDParm Disk Read, 421.38 MB/sec
PCIe 3.0
Compute Blade v1-mk4
Samsung 970 EVO Plus
The result in the same test shows a 94% performance boost!
HDParm Disk Read, 816.69 MB/sec
PCIe 3.0, CPU OC 3GHz
Compute Blade v1-mk4
Samsung 970 EVO Plus
CPU overclocking does not give a noticeable speed increase
PCIe 3.0, CPU OC 3GHz, GPU 1050MHz
Compute Blade v1-mk4
Samsung 970 EVO Plus
However, overclocking the CPU paired with the GPU gives a good boost due to the peculiarities of the architecture. But remember that good cooling is required, as consumption and heat dissipation increase by 10-20% in peaks.
As I said at the beginning of the article, it’s not limited to disks and there are other m.2 devices such as FPGAs or external network cards. They too will get a nice speed boost
Compute Blade is on sale now and you can find a distributor at the link (https://computeblade.com/)
P.S. What a shame we only have 1x PCIe 3.0 line. In general, there are five of them in the processor. However, if they were brought out on CM5 connectors, all existing carrier boards would be changed. Maybe in CM6 and Compute Blade v.II …
I’m a systems engineer in JetBrains company. Uptime Lab founder. I’m glad to see you on my website! I hope you find my content useful. Please subscribe to my Instagram and Twitter. I post the newest updates there.